Validating therapy decisions by reasons for therapy switch in relapsing-remitting multiple sclerosis

Validating therapy decisions by reasons for therapy switch in relapsing-remitting multiple sclerosis | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Article Validating therapy decisions by reasons for therapy switch in relapsing-remitting multiple sclerosis Anna Maria Sakr, Joachim Havla, Ulrich Mansmann This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-8295331/v1 This work is licensed under a CC BY 4.0 License Status: Under Review Version 1 posted 5 You are reading this latest preprint version Abstract Models for individualized treatment recommendations in relapsing–remitting multiple sclerosis often select therapies based on predicted risks of relapse and/or 3-month confirmed disability progression (3m-CDP). Recommendations should be interpretable, align with guidelines, and reflect patient perspectives. We validated recently developed prognostic algorithms in eight subgroups defined by reason for therapy switch (intolerance, lack of efficacy, pregnancy desire, serious adverse events, programmed stop, personal convenience, therapy initiation, drug-holiday end). From the OFSEP registry, we analyzed 3768 therapy cycles (2017–2021) with six commonly used therapies (interferon beta, glatiramer acetate, teriflunomide, dimethyl fumarate, fingolimod, natalizumab). Algorithms produced ranked therapy lists per outcome; certainty was quantified by entropy (0 = clear; 4.78 = maximal uncertainty). Recommendations were more consistent for relapse (median entropy 0.54, IQR 0.39–0.61) than for 3m-CDP (0.95, IQR 0.88–1.01). Across subgroups, natalizumab most often ranked highest for relapse, whereas teriflunomide or interferon beta frequently ranked highest for 3m-CDP. Calibration varied across switcher subgroups and outcomes, while discrimination was comparable to overall set: C-index 0.776 (95% CI 0.753–0.797) for 3m-CDP and 0.638 (95% CI 0.615–0.661) for relapse. Guideline and relapse-based recommendations did not always align, notably in the pregnancy-desire subgroup. For 3m-CDP, guideline-based recommendations are scarce, precluding systematic comparison with model outputs. Agreement between the best-ranked therapy for relapse and 3m-CDP was zero (κ = 0), underscoring outcome-dependent divergence. Algorithm-generated recommendations depend on the chosen outcome and may diverge from guideline-based practice or patient perspectives. Transparent communication of uncertainty and outcome trade-offs is essential for shared decision-making. Health sciences/Diseases Health sciences/Medical research Health sciences/Neurology Biological sciences/Neuroscience Clinical Decision Support Relapsing-Remitting Multiple Sclerosis Therapy Switches Clinical Practice Guidelines Evidence-Based Practice Figures Figure 1 Figure 2 Introduction Disease Context and Burden Multiple sclerosis is a common chronic autoimmune disease of the central nervous system, characterized by demyelinating inflammatory lesions and progressive neurodegeneration leading to disability. In 2020, the estimated global burden of MS was approximately 2.8 million people (35.9 per 100,000), showing a continued increase in both prevalence and incidence compared with 2013 [1]. The rise in MS incidence varies across disease phenotypes and between sex. A higher incidence is observed for relapsing-remitting multiple sclerosis (RRMS) than for primary progressive MS (PPMS)—the latter remaining stable or declining—and more in women than in men [2]. RRMS predominantly affects younger adults and women compared to men [1,3] reflecting both biological and possibly sociocultural determinants of disease risk and progression. Prognostic Prediction Models in MS: A Critical Gap Despite extensive research into disease mechanisms and therapy, prognostic prediction models for MS are not yet integrated into clinical practice. A recent Cochrane review concluded that no well-developed and independently externally validated prognostic models currently exist [4,5]. Yet such models are urgently needed, as they can support individualized clinical decision-making by predicting the heterogeneous outcomes among persons with MS (PwMS). We [6] recently developed and validated two prognostic models following the methodology of a promising commercial black-box tool, that used data from the NeuroTransData MS registry, designed to predict relapse and disease progression under specific treatments [7]. The original tool aimed to support physicians in optimal therapy selection in persons with RRMS (PwRRMS) [8]. Our initial goal was to externally validate those published models in new data; however, because model coefficients were proprietary and unavailable, we conducted an exact replication study to validate the methodological framework. Before such models can be implemented in clinical practice, multiple external validations across diverse settings, time periods, and patient populations are essential [9]. Changes in diagnostic criteria [10–14], epidemiological trends [2,15], disease course and classification [15], therapeutic management [16], and predictor or outcome definitions [17], can all lead to heterogeneous model performance [18]. Independent validations therefore increase confidence in a model’s generalizability and transportability to real-world clinical care. Study Population and Modelling Approach Our study included adult PwRRMS from the French OFSEP registry (Observatoire Français de la Sclérose en Plaques) [19] who initiated or switched, between January 2011 and December 2021, to six at this time commonly used disease-modifying therapies (DMTs): Interferon beta (IF), glatiramer acetate (GA), teriflunomide (TERI), dimethyl fumarate (DMF), fingolimod (FTY), or natalizumab (NA). Pre-switch clinical histories and demographic factors were used to predict key clinical outcomes: the number of relapses and the occurrence of disease progression confirmed at three months (3m-CDP). Therapy cycles starting before January 2017 served model development and internal validation, while those after were used for temporal external validation. By incorporating interactions between therapy types and patient characteristics, the models provide personalized predictions under each therapy, producing list of therapies ranked by predicted outcome. This can aid neurologists and future patients with similar histories to select the most promising treatment. We used a Bayesian machine learning algorithm that makes probabilistically informed predictions, rather than just a single educated guess [20]. The algorithm accounts for the variability in clinical practice and therefore may recommend multiple plausible treatment rankings. Presenting alternative rankings influences medical decision-making and counselling. Figure 1 illustrates how our models abstract clinical experience from data and generate treatment recommendations as ranked lists. Bayesian models also quantify individual treatment recommendation (ITR) certainty [21], distinguishing between patients with clear, unique recommendations (scenario 1, easier to treat) and those with broader treatment uncertainty (scenario 2, more complex cases). Guidelines Context Optimal therapy selection remains an unmet need [22] with major implications for multiple sclerosis morbidity and mortality [23]. The European (ECTRIMS/EAN) and the American Academy of Neurology (AAN) guidelines published in 2018— the first comprehensive reviews of DMT use [24–26]—were applicable to most of our validation period and likely informed the clinical practice recorded in the OFSEP registry. These guidelines emphasize individualized decision-making, primarily guided by patient characteristics, comorbidities, disease activity, drug safety, cost and accessibility, and patient preferences, rather than a single optimal strategy for all. In addition, they provide mainly consensus-based recommendations given the lack of evidence on most of the addressed clinical questions [22]. Switching DMTs is advised for inadequate response, intolerance, poor adherence, safety concerns, or pregnancy planning; escalation for lack of response or breakthrough disease activity; and de-escalation for adverse effects, advancing age, long disease duration, advanced disability or pregnancy [27]. Oral therapies may replace injectables in case of intolerance. (see Supplementary Table S1 ). Unlike ECTRIMS/EAN which specifies no strategy, the AAN recommends alemtuzumab, fingolimod, and natalizumab as possible initial therapies for highly active MS. For those switching due to breakthrough activity, it highlights evidence supporting alemtuzumab, fingolimod, natalizumab, and ocrelizumab, while advising consideration of individual factors such as adherence, safety, degree of disease activity, and the DMTs’ mechanisms of action. National guidelines, including the 2024 version of the German Neurological Society (DGN) [28] and an Italian Delphi [29] similarly endorse early high-efficacy therapy for high-risk or therapy escalation for suboptimally-controlled patients. Evolving Treatment Paradigms and Real-World Practice The treatment paradigm of RRMS is evolving. Two ongoing randomized controlled trials—DELIVER-MS [30] (NCT03535298) and TREAT-MS [31] (NCT03500328)—are comparing the effectiveness of early high-efficacy treatment (EHT) versus escalation in RRMS. Observational studies increasingly support the benefits of early use of high-efficacy DMTs for long-term disease control [32–34], yet real-world implementation remains heterogeneous [35]. A population based-analysis of PwMS in France from 2010 to 2015 [36], designed to describe real-world DMT use, found that 57% of patients initiated treatment with first-line DMTs (i.e., low efficacy), 13% received second-line agents (i.e., moderate or high efficacy; either following escalation or as EHT), 7% received off-label DMTs (likely reflecting use in progressive MS) and 23% discontinued therapy early (presumably due to benign or evolved disease). The most frequently prescribed DMTs during this period were IF, GA, NA, FTY, DMF, and TERI. Analyses of German claims data (2017–2022; 2020–2022) [37,38] found that two thirds of patients still initiated low- or moderate-efficacy therapy, with only gradual increases in high-efficacy DMT use, mainly in PPMS, while their use in RRMS remained relatively stable [38]. This pattern indicates that while the therapeutic shift toward high-efficacy agents is underway, it remains recent and uneven across disease phenotypes. Study Aim and Scope While current guidelines emphasize individualized therapy selection, clinical decision-making in practice remains complex. This study examines the potential of the replicated prognostic models to support treatment-switch decisions in PwRRMS, integrating both guidelines’ principles and patient perspectives. Although the models do not explicitly incorporate the reason for therapy switch, the OFSEP registry documents several reasons that enable model exploration across real-world decision contexts: lack of efficacy (NoEff), intolerance (Intolerance), programmed stop (ProgStop ), personal convenience (Convenience), pregnancy desire (PregDes) , and serious adverse events (SAE ). In addition, two derived categories capture transitions into therapy: ending therapy naivety (TxNaive) and ending a drug holiday (TxHol) . To illustrate how these models could inform real-world therapy decisions, Table 1 presents three representative therapy-switch scenarios that exemplify common clinical situations captured in the OFSEP registry and highlight the types of questions this study aims to address. This study aims to further validate the previously developed models across diverse patient profiles by investigating model behaviour across four key areas: (1) the robustness of individualized treatment recommendations (ITRs), (2) the alignment of ITRs with therapy switching reasons and guideline, (3) the comparability of prognostic performance across switcher subgroups, and (4) the agreement between relapse and progression-specific ITRs. Table 1 Representative therapy switches’ scenarios, captured in the OFSEP registry, illustrating the clinical relevance of individualized treatment recommendations Scenario Clinical context Patient characteristics Clinical question Lack of efficacy Escalation after recent relapse 30-year-old woman; 6-year RRMS; teriflunomide for 3 years; no prior therapy; relapse within past 6 months; EDSS 1.5 Should treatment be escalated to reduce risk of future activity? Pregnancy desire Switch for pregnancy planning 32-year-old woman; 6-year RRMS; fingolimod for 2 years; no prior therapy: relapse 2 years ago; EDSS 1.5 Which therapy is safest and effective when planning pregnancy? Personal convenience Switch to improve convenience/adherence 45-year-old woman; 11-year RRMS; interferon beta for 4.5 years; no prior therapy; relapse > 3 years ago; EDSS 1.5 Is an oral therapy appropriate given stable disease? Results Patient population Of the 3768 patients, 1177 (31%) were TxNaive, 1006 (27%) were on TxHol, 625 (17%) switched due to NoEff, 464 (12%) due to Intolerance, 116 (3%) due to ProgStop, 96 (3%) due to Convenience, 73 (2%) due to PregDes, and 25 (0.7%) due to SAE. Additionally, 102 (2.7%) had combined reasons, and 84 (2%) had a missing or unknown reason (Supplementary Fig. S2). Supplementary Fig. S3 shows that Intolerance and NoEff switches mainly occurred from platform DMTs, while ProgStop primarily from NA. Convenience and PregDes switches were mainly from IFN and FTY, respectively. Actual switches by subgroup can be found in Supplementary Table S3, Supplementary Table S4, and Supplementary Fig. S4. Table 2 summarizes predictors and clinical outcomes. Most patients were female and under 50, except for the SAE subgroup, where 32% were aged 50 or older. Most patients had not previously received second-line therapies, except for 74% in Progstop. TxNaive patients had a median time to therapy initiation of 2.1 years (range: 0.50 to 52), while median therapy discontinuation duration was 0.8 years (range: 0.25 to 22). The Convenience subgroup had the longest previous therapy duration (median: 4.4 years, range: 0.19 to 19), while the Intolerance subgroup had the shortest (median: 1.6 years, range: 0.04 to 22). In most subgroups, over a year had passed since the last relapse for the majority of patients, except for NoEff (48%) and TxNaive (56%). Table 2 Overview of predictors at baseline and clinical outcomes at baseline in the development set (Dev Set), the overall validation set (Val Set) and by subgroup of therapy switchers Predictor Dev Set (N = 5517) Val Set (N = 3768) TxHol (N = 1006) Intolerance (N = 538) NoEff (N = 674) Convenience (N = 126) PregDes (N = 85) ProgStop (N = 137) SAE (N = 31) TxNaive (N = 1177) Female 4,155 (75%) 2,864 (76%) 838 (83%) 412 (77%) 498 (74%) 84 (67%) 84 (99%) a 98 (72%) 21 (68%) 836 (71%) Age ≤ 30 916 (17%) 728 (19%) 168 (17%) 82 (15%) 152 (23%) 20 (16%) 23 (27%) 17 (12%) 5 (16%) 275 (23%) 31 to 40 1,843 (33%) 1,382 (37%) 424 (42%) 174 (32%) 230 (34%) 34 (27%) 57 (67%) 48 (35%) 7 (23%) 418 (36%) 41 to 50 1,724 (31%) 974 (26%) 234 (23%) 164 (30%) 191 (28%) 40 (32%) 5 (5.9%) 40 (29%) 9 (29%) 288 (24%) ≥ 51 1,034 (19%) 684 (18%) 180 (18%) 118 (22%) 101 (15%) 32 (25%) 0 (0%) 32 (23%) 10 (32%) 196 (17%) DMT count 0 1,435 (26%) 1,177 (31%) 0 (0%) 0 (0%) 0 (0%) 0 (0%) 0 (0%) 0 (0%) 0 (0%) 1,177(100%) 1 2,057 (37%) 1,325 (35%) 454 (45%) 307 (57%) 418 (62%) 72 (57%) 34 (40%) 34 (25%) 14 (45%) 0 (0%) 2 1,149 (21%) 729 (19%) 306 (30%) 143 (27%) 168 (25%) 25 (20%) 32 (38%) 48 (35%) 10 (32%) 0 (0%) ≥ 3 876 (16%) 537 (14%) 246 (24%) 88 (16%) 88 (13%) 29 (23%) 19 (22%) 55 (40%) 7 (23%) 0 (0%) Second-line 1,051 (19%) 518 (14%) 214 (21%) 61 (11%) 65 (9.6%) 10 (7.9%) 47 (55%) 101 (74%) 15 (48%) 0 (0%) Current therapy DMF 150 (2.7%) 333 (8.8%) 0 (0%) 159 (30%) 138 (20%) 11 (8.7%) 19 (22%) 4 (2.9%) 6 (19%) 0 (0%) IF 1,227 (22%) 443 (12%) 0 (0%) 177 (33%) 169 (25%) 76 (60%) 4 (4.7%) 19 (14%) 3 (9.7%) 0 (0%) GA 595 (11%) 244 (6.5%) 0 (0%) 82 (15%) 128 (19%) 24 (19%) 3 (3.5%) 12 (8.8%) 2 (6.5%) 0 (0%) TERI 67 (1.2%) 290 (7.7%) 0 (0%) 72 (13%) 184 (27%) 7 (5.6%) 12 (14%) 7 (5.1%) 5 (16%) 0 (0%) FTY 89 (1.6%) 136 (3.6%) 0 (0%) 29 (5.4%) 47 (7.0%) 3 (2.4%) 37 (44%) 5 (3.6%) 10 (32%) 0 (0%) NA 519 (9.4%) 139 (3.7%) 0 (0%) 19 (3.5%) 8 (1.2%) 5 (4.0%) 10 (12%) 90 (66%) 5 (16%) 0 (0%) NoDMT 2,870 (52%) 2,183 (58%) 1,006 (100%) 0 (0%) 0 (0%) 0 (0%) 0 (0%) 0 (0%) 0 (0%) 1,177 (100%) Current duration 2.03 (0.01, 0.86, 5.12, 46.96) 1.63 (0.04, 0.76, 4.10, 51.82) 0.80 (0.25, 0.46, 1.44, 22.13) 1.56 (0.04, 0.60, 3.65, 22.05) 2.44 (0.12, 1.38, 4.48, 18.82) 4.41 (0.19, 1.34, 7.82, 19.27) 2.23 (0.12, 1.25, 3.72, 10.42) 2.68 (0.16, 1.52, 7.28, 19.99) 1.86 (0.06, 0.50, 4.10, 17.47) 2.14 (0.50, 0.96, 5.91, 51.82) b Index therapy DMF 1,086 (20%) 738 (20%) 195 (19%) 123 (23%) 68 (10%) 35 (28%) 5 (5.9%) 14 (10%) 9 (29%) 285 (24%) IF 901 (16%) 284 (7.5%) 116 (12%) 40 (7.4%) 6 (0.9%) 6 (4.8%) 14 (16%) 3 (2.2%) 4 (13%) 99 (8.4%) GA 663 (12%) 538 (14%) 141 (14%) 69 (13%) 8 (1.2%) 6 (4.8%) 50 (59%) 6 (4.4%) 4 (13%) 255 (22%) TERI 930 (17%) 953 (25%) 261 (26%) 214 (40%) 27 (4.0%) 64 (51%) 0 (0%) 12 (8.8%) 5 (16%) 373 (32%) FTY 1,381 (25%) 767 (20%) 170 (17%) 67 (12%) 341 (51%) 13 (10%) 0 (0%) 90 (66%) 3 (9.7%) 84 (7.1%) NA 556 (10%) 488 (13%) 123 (12%) 25 (4.6%) 224 (33%) 2 (1.6%) 16 (19%) 12 (8.8%) 6 (19%) 81 (6.9%) Index duration 1.37 (0.00, 0.61, 2.20, 5.91) 1.29 (0.00, 0.58, 2.41, 4.87) 1.20 (0.00, 0.52, 2.23, 4.87) 1.19 (0.01, 0.52, 2.26, 4.85) 1.63 (0.01, 0.85, 2.70, 4.80) 1.66 (0.03, 0.82, 2.52, 4.55) 0.70 (0.02, 0.34, 1.01, 4.55) 1.41 (0.03, 0.60, 2.96, 4.79) 1.05 (0.03, 0.60, 2.56, 4.73) 1.29 (0.01, 0.56, 2.38, 4.70) Onset distance c 7.50 (0.50, 3.22, 13.48, 54.96) 5.79 (0.50, 2.31, 11.91, 51.83) 8.09 (0.52, 4.74, 13.39, 43.42) 6.56 (0.53, 2.94, 12.94, 41.52) 6.17 (0.62, 3.05, 11.92, 41.52) 10.97 (0.65, 5.68, 16.20, 35.92) 5.58 (1.27, 3.42, 8.47, 17.31) 10.82 (0.68, 5.24, 16.96, 39.61) 8.27 (0.63, 5.21, 13.12, 41.79) 2.14 (0.50, 0.97, 5.92, 51.83) c Relapse distance < 3 months 407 (7.4%) 234 (6.2%) 71 (7.1%) 15 (2.8%) 66 (9.8%) 3 (2.4%) 2 (2.4%) 5 (3.6%) 0 (0%) 73 (6.2%) ≥ 3 months and < 1year 1,961 (36%) 1175 (31%) 177 (18%) 114 (21%) 254 (38%) 15 (12%) 11 (13%) 12 (8.8%) 8 (26%) 590 (50%) 1–3 years 1,561 (28%) 1173 (31%) 322 (32%) 190 (35%) 177 (26%) 28 (22%) 40 (47%) 45 (33%) 9 (29%) 362 (31%) ≥ 3 years 1,588 (29%) 1186 (31%) 436 (43%) 219 (41%) 177 (26%) 80 (63%) 32 (38%) 75 (55%) 14 (45%) 152 (13%) Baseline EDSS ≤ 1.5 2,411 (44%) 2040 (54%) 490 (49%) 302 (56%) 366 (54%) 70 (56%) 65 (76%) 55 (40%) 18 (58%) 690 (59%) 2 to 2.5 1,336 (24%) 898 (24%) 247 (25%) 127 (24%) 157 (23%) 26 (21%) 13 (15%) 35 (26%) 6 (19%) 290 (25%) 3 to 3.5 759 (14%) 434 (12%) 127 (13%) 62 (12%) 78 (12%) 12 (9.5%) 3 (3.5%) 22 (16%) 1 (3.2%) 121 (10%) 4 to 6 1,011 (18%) 396 (11%) 142 (14%) 47 (8.7%) 73 (11%) 18 (14%) 4 (4.7%) 25 (18%) 6 (19%) 76 (6.5%) Relapse count d 0.55 (0.72) 0.45 (0.63) 0.29 (0.54) 0.28 (0.54) 0.58 (0.68) 0.17 (0.43) 0.16 (0.40) 0.15 (0.45) 0.32 (0.65) 0.67 (0.67) Relapse-free e 3,129 (57%) 2349 (62%) 756 (75%) 408 (76%) 353 (52%) 108 (86%) 72 (85%) 120 (88%) 23 (74%) 509 (43%) Clinical outcomes 3m-CDP free 5,005 (91%) 3449 (92%) 929 (92%) 497 (92%) 610 (91%) 117 (93%) 80 (94%) 126 (92%) 28 (90%) 1,070 (91%) Relapse-free f 4,225 (77%) 3177 (84%) 849 (84%) 457 (85%) 584 (87%) 109 (87%) 63 (74%) 116 (85%) 25 (81%) 982 (83%) Table 3 compares observed relapse activity pre- and post-switch. The highest mean difference (MD) in annualized relapse rate (ARR) was in TxNaive (MD = 0.47, 95% CI: 0.42–0.51) and NoEff (MD = 0.4, 95% CI: 0.34–0.46). Relapse-free proportions improved in these groups (NoEff: 52% to 87%, odds ratio (OR) = 8.45, 95% CI: 5.81–12.69; TxNaive: 43% to 83%, OR = 7.76, 95% CI: 6.04–10.10). TxHol and Intolerance had similar pre-switch disease activity and comparable post-switch relapse reduction. Convenience had lower relapse activity pre-switch and remained stable post-switch. SAE showed a similar ARR pre- and post-switch but an increased relapse-free proportion. PregDes and ProgStop had pre-switch the lowest ARR (0.16 and 0.15) and highest relapse-free proportions (85% and 88%), but post-switch ARR increased (PregDes: MD = -0.21, 95% CI: -0.39 to -0.04; ProgStop: MD = -0.04, 95% CI: -0.15 to 0.07). Relapse-free proportions decreased in both groups (PregDes: OR = 0.50, 95% CI: 0.20–1.17; ProgStop: OR = 0.75, 95% CI: 0.32–1.69). Table 3 Relapse activity pre- and post-therapy switch. Reason (N) Mean ARR pre-switch (SD) Mean ARR post-switch (SD) Mean difference of ARR (95% CI) a Relapse-free pre-switch, N (%) Relapse-free post-switch, N (%) Odds ratio of relapse-free proportion (95% CI) b Most frequent switch, N (%) Therapy naive (N = 1177) 0.67 (0.67) 0.21 (0.52) 0.47 (0.42 to 0.51) 509 (43.25%) 982 (83.43%) 7.76 (6.04 to 10.10) NoDMT to TERI 373 (31.69%) Lack of efficacy (N = 674) 0.58 (0.68) 0.17 (0.48) 0.40 (0.34 to 0.46) 353 (52.37%) 584 (86.65%) 8.45 (5.81 to 12.69) TERI to FTY 113 (16.77%) Overall validation set (N = 3768) 0.45 (0.63) 0.20 (0.51) 0.25 (0.22 to 0.27) 2349 (62.34%) 3177 (84.32%) 3.52 (3.11 to 3.99) NoDMT to TERI 634 (16.83%) Development set (N = 5517) 0.55 (0.72) 0.33 (0.72) 0.21 (0.19 to 0.24) 3129 (56.72%) 4225 (76.58%) 2.87 (2.61 to 3.16) NoDMT to IF 779 (14.12%) Intolerance (N = 538) 0.28 (0.54) 0.18 (0.48) 0.10 (0.04 to 0.16) 408 (75.84%) 457 (84.94%) 1.82 (1.31 to 2.53) IF to TERI 90 (16.73%) Drug holiday (N = 1006) 0.29 (0.54) 0.20 (0.50) 0.09 (0.05 to 0.14) 756 (75.15%) 849 (84.39%) 1.81 (1.43 to 2.29) NoDMT to TERI 261 (25.94%) Personal convenience (N = 126) 0.17 (0.43) 0.16 (0.45) 0.01 (-0.11 to 0.12) 108 (85.71%) 109 (86.51%) 1.06 (0.50 to 2.25) IF to TERI 48 (38.1%) Serious adverse events (N = 31) 0.32 (0.65) 0.32 (0.75) 0.00 (-0.35 to 0.35) 23 (74.19%) 25 (80.65%) 1.50 (0.36 to 7.23) FTY to NA 4 (12.90%) Programmed stop (N = 137) 0.15 (0.45) 0.19 (0.49) -0.04 (-0.15 to 0.07) 120 (87.59%) 116 (84.67%) 0.75 (0.32 to 1.69) NA to FTY 80 (58.39%) Pregnancy desire (N = 85) 0.16 (0.40) 0.38 (0.76) -0.21 (-0.39 to -0.04) 72 (84.71%) 63 (74.12%) 0.5 (0.20 to 1.17) FTY to GA 18 (21.18%) Sample size For both outcomes, the available sample size ( N = 3768) exceeded the minimum required for reliable intercept and slope estimation. For the 3m-CDP outcome, assuming an 8% prevalence and a C-index of 0.777, the required sample size was 222 corresponding to 18 events and 8.88 events per predictor parameter. For the relapse outcome, assuming a 16% prevalence and a C-index of 0.639, the required sample size was 544, with 88 events and 43.52 events per predictor. Robustness of model ITRs Across all subgroups, the median of individual patient entropies for the recommended treatment triplet was 0.54 for the relapse model (interquartile range: 0.39–0.61; Supplementary Fig. S5). Boxplots of individual entropy values reveal substantial variability: some patients had high entropy, indicating very heterogeneous results, and some patients had entropy of 0, indicating a highly consistent recommendation of the treatment triplet (NA, DMF, FTY). For the 3m-CDP model, entropy was generally higher, with a median of 0.95 across subgroups (interquartile range: 0.88–1.01; Supplementary Fig. S6). Recommended therapy switches and outcome agreement Subgroup-specific best-ranked therapies are presented in Fig. 2 , and Supplementary Fig. S7-S9. Across subgroups, NA consistently ranked first for the relapse outcome. Recommending NA may suit patients with reason for therapy switching being TxNaive, TxHol, NoEff and Intolerance. However, it can be less suitable for all NoEff patients (for example patients with neutralizing antibodies against NA), ProgStop (especially patients needing a DMT with lower PML risk), Convenience (patients preferring oral therapies), PregDes (patients with stable disease requiring therapy de-escalation) and SAE (patients with risk of malignancies). For 3m-CDP, the first choices were TERI and IF. The agreement between relapse and 3m-CDP outcomes regarding the first-choice therapies was very low ( kappa = 0, Supplementary Table S4). For the relapse outcome, the most frequently recommended treatment triplets were (NA, FTY, DMF) and (NA, DMF, FTY) (Supplementary Fig. S10). Conversely, 3m-CDP outcome recommendations were more variable, with common triplets including (IF, TERI, DMF), (IF, DMF, TERI), (TERI, DMF, IF), (TERI, DMF, NA), (TERI, GA, DMF) and (TERI, DMF, GA) (Supplementary Fig. S11). Compliance of algorithmic output with guidelines Guidelines provide general and less specific recommendations (Supplementary Table S1 ) making direct comparisons with our Bayesian models recommendations difficult for most switching subgroups. The relapse model recommendations comply with guidelines for NoEff and SAE, but not for PregDes. Specifically, while guidelines recommend stopping DMT or de-escalating to IF or GA for women planning pregnancy, or using NA only if disease activity persists, our model primarily recommended the triplet (NA, FTY, DMF) for PregDes patients. The 3m-CDP model’s recommendations partially aligned with guidelines, which focus on reducing relapse or magnetic resonance imaging (MRI) disease activity. Model performance The recalibrated relapse model showed improved calibration, a C-index of 0.638 (95% CI: 0.615 to 0.661), and an MSE reduction from 0.326 to 0.248 showing improved global fit compared to the original model. The recalibrated 3m-CDP model exhibited good calibration, a C-index of 0.776 (95% CI: 0.753 to 0.797), and an MSE of 0.072 (Table 4 , Supplementary Table S6, Supplementary Fig. S12-S15). Table 4 highlights that the relapse model exhibited consistent and good calibration across most subgroups, except for the PregDes, SAE, and ProgStop. It slightly overpredicted the average number of relapses in NoEff, ProgStop, and Convenience, and slightly underpredicted it in PregDes, SAE and TxNaive. The 3m-CDP model performed consistently across subgroups but overestimated the average 3m-CDP occurrence risk in ProgStop by 5%, and by 1% in NoEff, PregDes and SAE compared to the mean observed proportions. Additional technical details are provided in Supplementary Fig. S12-S16. Table 4 Performance of the recalibrated models in the overall validation set and in the reason of switch subgroups. Outcome Performance measure (95% CI) Overall validation set (N = 3768) Drug holiday (N = 1006) Therapy naive (N = 1177) Intolerance (N = 538) Lack of efficacy (N = 674) Pregnancy desire (N = 85) Serious adverse events (N = 31) Programmed stop (N = 137) Personal convenience (N = 126) 3m-CDP Calibration intercept 0 (-0.120 to 0.120) 0.017 (-0.224 to 0.259) 0.089 (-0.119 to 0.297) 0.018 (-0.315 to 0.351) -0.082 (-0.350 to 0.185) -0.220 (-1.157 to 0.716) -0.176 (-1.432 to 1.081) -0.547 (-1.190 to 0.095) -0.053 (-0.757 to 0.650) Calibration slope 1.001 (0.865 to 1.137) 0.991 (0.713 to 1.268) 1.040 (0.801 to 1.278) 0.989 (0.626 to 1.352) 1.002 (0.669 to 1.335) 0.969 (-0.009 to 1.948) 1.047 (-0.352 to 2.446) 0.888 (0.196 to 1.581) 1.005 (0.209 to 1.802) C-index 0.776 (0.753 to 0.797) 0.773 (0.723 to 0.816) 0.783 (0.745 to 0.816) 0.785 (0.714 to 0.843) 0.746 (0.687 to 0.798) 0.787 (0.588 to 0.906) 0.786 (0.454 to 0.942) 0.768 (0.656 to 0.852) 0.779 (0.523 to 0.919) MSE 0.072 0.066 0.077 0.065 0.080 0.053 0.079 0.076 0.059 RRMSE (%) 97 97 96 96 97 97 94 101 94 Predicted proportion (range) 0.08 (0 to 0.60) 0.08 (0 to 0.41) 0.08 (0 to 0.52) 0.08 (0 to 0.52) 0.10 (0 to 0.60) 0.07 (0 to 0.32) 0.11 (0 to 0.39) 0.13 (0 to 0.50) 0.07 (0 to 0.28) Observed proportion (range) 0.08 (0 to 1) 0.08 (0 to 1) 0.09 (0 to 1) 0.08 (0 to 1) 0.09 (0 to 1) 0.06 (0 to 1) 0.10 (0 to 1) 0.08 (0 to 1) 0.07 (0 to 1) Number of Relapses Calibration intercept -0.004 (-0.093 to 0.086) -0.007 (-0.180 to 0.167) 0.119 (-0.038 to 0.275) -0.020 (-0.261 to 0.220) -0.269 (-0.494 to -0.044) 0.540 (0.043 to 1.036) 0.152 (-0.773 to 1.077) -0.237 (-0.71 to 0.237) -0.202 (-0.721 to 0.317) Calibration slope 0.973 (0.795 to 1.151) 0.897 (0.564 to 1.229) 0.871 (0.564 to 1.178) 1.361 (0.833 to 1.889) 1.082 (0.592 to 1.572) 2.186 (0.818 to 3.553) 0.621 (-0.668 to 1.909) 0.274 (-0.500 to 1.049) 1.052 (0.001 to 2.103) C-index 0.638 (0.615 to 0.661) 0.632 (0.587 to 0.676) 0.622 (0.580 to 0.661) 0.688 (0.634 to 0.738) 0.639 (0.580 to 0.693) 0.755 (0.623 to 0.852) 0.617 (0.338 to 0.835) 0.533 (0.401 to 0.659) 0.637 (0.493 to 0.759) MSE 0.248 0.243 0.258 0.223 0.226 0.543 0.538 0.255 0.192 RRMSE (%) 98 98 98 98 99 98 98 102 98 Mean predicted (range) 0.20 (0.01 to 0.95) 0.20 (0.01 to 0.66) 0.19 (0.01 to 0.56) 0.19 (0.01 to 0.67) 0.21 (0.01 to 0.88) 0.22 (0.03 to 0.53) 0.23 (0.02 to 0.77) 0.25 (0.02 to 0.63) 0.20 (0.02 to 0.67) Mean observed (range) 0.20 (0 to 4) 0.20 (0 to 4) 0.21 (0 to 4) 0.18 (0 to 3) 0.17 (0 to 3) 0.38 (0 to 4) 0.32 (0 to 3) 0.19 (0 to 3) 0.16 (0 to 3) Discussion This study validated previously developed Bayesian prognostic models [6] across eight patient subgroups defined by their reason for therapy switch. The models predicted relapse and 3-month confirmed disability progression (3m-CDP) under six treatments over a specific time horizon. Although switching reasons were not explicitly modelled, we examined whether such factors were indirectly captured through patient characteristics during model development. Because switching is often driven by lack of efficacy, intolerance, safety concerns, or personal preferences, exploring how model behaviour reflects real-world decision-making is clinically relevant. While neurologists likely followed guidelines and general principles (Supplementary Table S1 ), it remains uncertain whether one general model is sufficient for clinical use or if subgroup-specific models are required. We examined four key aspects related to shared decision-making (SDM) in PwRRMS following a therapy switch: (1) the robustness of individualized treatment recommendations (ITRs), (2) their alignment with switching reasons and guidelines, (3) the comparability of prognostic performance across subgroups, and (4) the agreement between relapse and progression-specific ITRs. Our analysis included 3768 adult PwRRMS treated between January 2017 and December 2021 in the French OFSEP registry. Eligible patients had an EDSS below or equal to 6 and had received therapy at least 6 months after disease onset. Patients were treated with IF, GA, TERI, DMF, FTY, or NA with roughly one-third being treatment-naive, one-third resuming therapy after its discontinuation, and the remainder switching therapies from platform DMTs— primarily due to lack of efficacy or intolerance (Supplementary Fig. S3). Tallantyre et al. [39] found similar switching patterns in their study on real-world DMT persistence and discontinuation reasons, supporting the representativeness of our cohort for current PwRRMS management. In our study, 93% of therapy-naïve patients initiated a low- or moderate-efficacy DMT. This aligns with nationwide analyses from Germany (2017–2022 and 2020–2022) showing that 85–86% of adults with MS started therapy with low- or moderate-efficacy agents at this time [37,38]. Both studies noted a gradual shift from the traditional escalation approach—initiating lower-efficacy treatments and escalating upon breakthrough disease activity—toward the early high-efficacy treatment (EHT) strategy. However, this transition remains incomplete and is not reflected in the analyzed data set. In particular, increased use of high-efficacy therapy was mainly observed in PPMS, which could be driven by the 2018 approval of ocrelizumab [40], while use in RRMS remained stable [38]. In addition, because MS predominantly affects women of childbearing age and at this time only a few DMTs could be safely used during pregnancy or lactation, lower-efficacy yet safer agents remained an important therapeutic option for managing this patient subgroup [27,41,42]. These observations indicate that, despite emerging evidence supporting EHT, the escalation approach still dominated clinical practice during our study period (2017–2021). The limited representation of recently approved high-efficacy DMTs—such as anti-CD20 antibodies (ocrelizumab, ofatumumab, ublituximab or rituximab (off-label)), immune-reconstitution therapies (cladribine, alemtuzumab), and newer S1P modulators (ozanimod, ponesimod, siponimod)—reflects their gradual introduction into the market and adoption. Similar low uptake was reported by Stratil et al. [38]. Likewise, most of these agents were not yet included among the therapies listed in the French Multiple Sclerosis Society (SFSEP) consensus guidelines on DMT switching [43], further confirming that our data are representative of real-world MS management at that time and the data is still informative for this particular group of drugs. To illustrate clinical applicability, Table 5 revisits the three patient scenarios introduced in Table 1 , presenting model-based treatment rankings and associated certainty (entropy) for each outcome, alongside brief clinical interpretation. Decision-making is straightforward in some patients but complex in others [22], and this heterogeneity is reflected in the OFSEP data and, consequently, in the algorithm built from it. Traditional models often overlook such uncertainty by providing a single “best” answer per patient, whereas our Bayesian framework quantifies probabilistic uncertainty that more closely resembles clinical reasoning. Combined with decision curve analysis based on the Chalkou et al. [44] method, this framework demonstrates how interpretable, uncertainty-aware models can inform treatment recommendations. Our work complements the Big Multiple Sclerosis Data Network initiative, which aims to improve methodological standards and real-world data analytics in MS research [45]. Table 5 Model-based individualized treatment recommendations in the three therapy-switch cases. Scenario Clinical Outcome Top model-recommended therapies (triplet) Entropy (certainty) Clinical interpretation Lack of efficacy At least one relapse (NA, FTY, DMF): 83% (NA, FTY, TERI): 17% 0.456 Patient had a breakthrough relapse while on TERI. Escalation to higher-efficacy therapy in this case is guideline-consistent. Switch to NA or FTY is clinically plausible; Consistent outputs (2 triplets) → higher certainty 3m-CDP (DMF, IFN, TERI):27% (GA, IFN, TERI): 21% (GA, TERI, DMF): 10% + 7 other triplets: 42% 2.169 Model recommends a horizontal switch to mild- moderate efficacy therapies. It is difficult to evaluate as no guideline-defined strategy for progression-based switching and limited evidence on DMTs efficacy for preventing disability progression; Inconsistent outputs (10 triplets) → lower certainty Pregnancy desire At least one relapse (NA, DMF, FTY): 98% (NA, FTY, DMF): 2% 0.098 NA can be given in early pregnancy if high disease activity but with risks. The other model recommendations are not advised in pregnancy → requires clinical judgment given safety considerations; Consistent outputs → higher certainty 3m-CDP (TERI, DMF, IFN): 94% (TERI, IFN, DMF): 6% 0.227 IFN recommended by model is a safe option in case of disease activity and aligns with guidelines, but TERI and DMF are contraindicated/not recommended in pregnancy → requires clinical judgment given safety considerations; Consistent outputs → higher certainty Personal convenience At least one relapse (DMF, NA, FTY): 66% (FTY, NA, DMF): 31% (NA, DMF, FTY): 2% (NA, FTY, DMF): 1% 0.762 The recommended options DMF and FTY are oral therapies and align with patient preference of switching from injectable IFN, which seems clinically reasonable. No official guideline recommendation so comparison not possible; Less consistent outputs → moderate certainty 3m-CDP (IFN, DMF, TERI): 65% (IFN, TERI, DMF): 24% (IFN, TERI, GA): 5% (TERI, IFN, DMF): 5% (TERI, IFN, GA): 1% 0.968 Patient can stay on IFN or switch to DMF or TERI. Options are clinically acceptable given patient older age and stable disease, but partially divergent from relapse model; Less consistent outputs → moderate certainty Current guidelines likewise address uncertainty only indirectly by offering general rather than specific recommendations [22,25,26], and an update might be necessary in light of emerging evidence. For example, they advise monitoring and reevaluating patients after therapy discontinuation, but specify no strategies for re-initiation or switching. While alemtuzumab was listed as potential initial therapy by the 2018 AAN guidelines for high risk patients, the European Medicines Agency has restricted its use in 2019 as first-line therapy due to safety concerns [46], leading to its limited use in practice, as also shown by Papukchieva et al. [37]. We show, whenever comparison with guidelines was possible, that model recommendations did not always align with guidelines. While natalizumab consistently ranked highest for relapse prevention—supporting both model validity and aligning with clinical expectations—recommendations for the pregnancy desire subgroup included contraindicated options (FTY, TERI), highlighting the importance of contextual interpretation and clinician oversight. These discrepancies raise broader questions about how divergent model outputs and guideline recommendations should be handled and integrated in the SDM process between patient and neurologist. Uncertainty is an intrinsic part of MS management, and probabilistic predictions can make it more transparent to both physicians and patients. However, clinicians must still interpret model outputs within the clinical, safety, and personal context of each patient. SDM can be strengthened through patient-preparation tools—such as MS-SUPPORT [47]—that help patients clarify their treatment goals, become more informed about their disease and therapy options, and improve their readiness for collaborative decision-making. This in turn can help improve adherence to therapy and disease outcomes. Some miscalibration of the model occurred in smaller subgroups, which could lead to over- or under-treatment. Because we did not oversample low-prevalence subgroups, further research should determine whether one general model—which does not explicitly account for switching reasons—is clinically satisfactory, or whether subgroup-specific models are necessary to improve accuracy. We also observed inconsistent treatment recommendations between both outcome models. While the 3m-CDP model recommended diverse treatment triplets for individual patients, the relapse model consistently suggested only two quite similar triplets. This divergence complicates prioritization of treatment goals and the practical implementation of model recommendations for therapy switches. As mentioned earlier, the models mirror the uncertainties encountered in clinical practice, where optimal strategies for individual patients remain unclear due to limited comparative evidence [48,49]. This challenge is also reflected in guideline recommendations [26], which are largely consensus-based, and in González-Lorenzo et al. [50] findings, whose network meta-analysis revealed uncertainties in DMT efficacy for 3m-CDP but produced rankings similar to our relapse model. Because a network meta-analysis is population-based and does not provide individualized effect rankings, we cautiously compared our results to theirs. Several limitations must be acknowledged. Key variables influencing therapy choice—comorbidities, concomitant medications, John Cunningham virus status, MRI findings, and adherence—were unavailable and may have affected model outputs. Furthermore, AAN guidelines recommend high efficacy therapies for highly active disease or switching within the same efficacy category for patients already on such therapies. The limited availability of therapeutic options in 2011–2021 is no longer comparable with the current therapeutic landscape. This explains the absence of newer agents from the analyses. However, this is consistent with findings from the previously cited French studies [36,43]. Updating the model with contemporary OFSEP data incorporating these therapies will be important for future validation. Strengths include the use of representative, prospectively collected, and fit-for-purpose registry data with low missingness for switching reasons, a large sample size across treatment groups, and the inclusion of diverse predefined subgroups with varying risk profiles and treatment responses (Tables 2 and 3 ), better reflecting the real-world PwRRMS target population. These features enhance generalizability compared with randomized trials, which—owing to strict eligibility criteria—often lack external validity. This underscores the value of registry-based prognostic modelling and the need for better research, particularly in heterogeneous patient subgroups, as emphasized by the European Medicines Agency’s guidelines [51]. In addition to supporting individualized decisions, prognostic models should be fair to avoid discrimination against individuals or groups [52]. Because MS disproportionately affects younger adults and women, including sex and age as predictors in our models helps account for key demographic differences and supports balanced recommendations. While the models demonstrated robust ITRs, some misalignment with guidelines persisted, highlighting the inherent complexity of treatment decisions even when supported by guidelines and algorithms. Physician expertise in balancing model insights with clinical guidelines and patient preferences remains key [53]. In addition, emerging evidence [15] increasingly supports viewing multiple sclerosis as a biological continuum rather than a collection of discrete subtypes. This perspective emphasizes that inflammatory and neurodegenerative mechanisms coexist throughout the disease course and may explain discrepancies between relapse- and progression-based predictions. In particular, progression independent of relapse activity (PIRA) occurs across all MS phenotypes and is associated with poor prognosis yet remains difficult to detect when defined solely by conventional endpoints such as EDSS. Incorporating patient data with different MS subtypes, patient-reported outcomes (e.g., fatigue, cognition, mental health) and functional measures into prognostic modelling may improve early identification of subtle progression and align model predictions with a more continuous and patient-centred understanding of MS. Finally, implementing model-based treatment decision support in clinical practice requires further validation and consideration, as clear evidence is currently lacking. To assess the utility of such models, a cluster-randomized trial would be ideal. Ultimately, such tools must operate in—and aim to improve—imperfect clinical care settings [54]. Materials and methods Data source and Participants This project was approved by the Ethics Committee of Ludwig-Maximilians-University Munich (No. 21-1174) and the OFSEP steering board. We used longitudinal data from 3768 patients in the French multiple sclerosis registry OFSEP [55], treated between January 1, 2017 and December 15, 2021, which marked the end of follow-up. OFSEP provides high-quality, standardized data collected prospectively during regular visits across 38 French centres [56]. OFSEP physicians obtained written consent from all participants. The study was announced on the OFSEP website ( https://www.ofsep.org/fr/etudes/extval-phrend ), allowing patients to provide dynamic consent. Eligible PwRRMS had a minimum disease duration of 6 months, an EDSS score below or equal to 6, and were treated with IF, GA, TERI, DMF, FTY, or NA, and either initiated or switched therapy. We considered patients as therapy-naive if they had not received any therapy for at least 6 months following disease onset and on a drug holiday if they had discontinued one of the studied DMTs for at least 3 months prior to re-initiating therapy. This threshold allows distinguishing clinically meaningful treatment gaps from standard washout periods required for therapy switches (see Supplementary Table S2). Variables Predictors were assessed before or at index therapy initiation (baseline) and included: previous therapy (current therapy) , and its duration ( current duration ), whether this therapy or the one prior to it was second-line ( second-line ), number of previous DMTs (DMT count) , number of relapses in the previous year (relapse count) , time since last relapse (relapse distance) , duration since disease onset ( onset distance ), age at therapy start, baseline Expanded Disability Status Scale (EDSS) , index therapy , and its duration. We used index duration as an offset term to account for therapy duration variability. Outcomes included the number of relapses and 3m-CDP occurrence during index therapy. 3m-CDP was defined as a sustained EDSS increase from baseline (≥ 1 point, or ≥ 0.5 if baseline EDSS > 5.5) with no improvement over 3 months, confirmed by a follow-up EDSS measured ≥ 3 months after the increase and within 12 months after therapy end, and occurring at least 3 months after any relapse. Data processing steps, predictor selection, and variable definitions are detailed in Sakr et al. [6]. Eight patient subgroups were defined based on the reason for therapy switching: lack of efficacy, intolerance, programmed stop, personal convenience, pregnancy desire, serious adverse events, ending therapy naivety or ending drug holiday. We performed our analyses in R, version 4.2.2. We also reported this work in accordance with the TRIPOD + AI statement [52] in the Supplementary Fig. S17 to ensure transparency and reproducibility. Sample size Although our previously developed models generally performed well, their predictions did not consistently align with observed outcome risks in some patient groups. Therefore, we updated the models’ intercept and slope using the validation set. To estimate a sufficient sample size for this update, we applied the method proposed by Riley et al. [57], which determines the minimum sample size needed for binary outcome prediction models to reduce overfitting and precisely estimate the average outcome risk. This method requires specifying the expected outcome proportion in the new dataset, the anticipated model performance (e.g., Cox-Snell R² or C-index), the number of model parameters, and a targeted shrinkage factor—here set to 10%. For our calculations, we used the original models’ C-index and the outcome prevalence in the validation set, defined as the proportion of patients experiencing at least one relapse or disease progression. Statistical analysis methods Robustness of ITRs The Bayesian machine learning algorithm generates a vector of six relapse or 3m-CDP probabilities under the six DMTs, ranked by the value of a favourable prognosis. To assess recommendation certainty, we drew 100 samples per patient from the risk posterior distribution, counting how often each DMT ranked, first, second, third, etc. To simplify, we focused on the top three DMTs in each sample—the ones with best prognoses—and refer to these as treatment triplet . We then assessed consistency of the recommended treatment triplet using entropy. With six DMTs, there are 120 (6 x 5 x 4) possible triplet rankings. Maximum entropy (-ln(1/120) = ln(120) = 4.78) indicates equal probability of the 120 possible results and thus complete uncertainty, while entropy of 0 reflects identical algorithmic recommendations across samples. Values near 0 indicate robustness, while we considered values above 2 as heterogeneous or uncertain. Compliance of recommendations with switching reasons and guidelines. We summarized relevant guidelines’ recommendations (Supplementary Table S1 ) and descriptively assessed whether subgroup-specific top-ranked ITRs align with both clinical guidelines and reason for therapy switching. Performance assessment We evaluated model performance in the overall validation set and predefined subgroups following established methods for clinical prediction models [58–60]. Performance assessment included calibration, discrimination and global fit. Calibration, assessed via calibration curves, intercept and slope, reflects the agreement between observed outcome proportion and predicted outcome probability [61,62]. Discrimination was evaluated by the C-index (similar to the area under the curve AUC). The mean squared error (MSE) and relative root MSE (RRMSE) quantified the model’s fit absolutely or relative to the outcome variance [63]. The RRMSE enables cross-setting comparisons. Presentation of model predictions We developed a web-based Shiny application (available at https://shiny.ibe.med.uni-muenchen.de/switch-ms ; permalink: https://switchms.annamariasakr.de accessible with the following credentials: user: msswitch, password: Roox3iey) (Supplementary Fig. S1 ). To use the calculator please push on the tab “Outcome predictions and Treatment recommendations”, than users can input the routinely collected patient data and generate predicted probabilities for relapse and 3m-CDP under the six DMTs. The app provides two therapy rankings based on: (1) best prognosis and (2) benefit outweighing harms (Decision Curves Analysis, Chalkou et al . [44]). Agreement between 3m-CDP and Relapse models’ recommendations We used Cohen’s kappa to measure the agreement between ITRs for relapse and 3m-CDP. First-ranked therapies under the best prognosis approach were categorized based on the German Neurological Society (DGN) classification [28]: (1) platform DMTs: IFN, GA, TERI, DMF, (2) moderate efficacy FTY and (3) high efficacy NA. Abbreviations AAN American Academy of Neurology AUC Area under the curve 3m-CDP Confirmed disease progression at 3 months CI Confidence interval Convenience Personal convenience DGN Deutsche Gesellschaft für Neurologie, German neurological society DMF Dimethyl fumarate DMT Disease-modifying therapy ECTRIMS/EAN European Committee of Treatment of Research in Multiple Sclerosis/ European Academy of Neurology EDSS Expanded disability status scale EHT Early high-efficacy treatment FTY Fingolimod GA Glatiramer acetate IF Interferon beta 1 ITRs Individual treatment recommendations MRI Magnetic resonance imaging MS Multiple sclerosis MSE Mean squared error NA Natalizumab NoEff Lack of efficacy OFSEP Observatoire Français de la Sclérose en Plaques OR Odds ratio PIRA Progression independent of relapse activity PML Progressive multifocal leukoencephalopathy PwMS Persons with multiple sclerosis PwRRMS Persons with relapsing-remitting multiple sclerosis PPMS Primary progressive multiple sclerosis PregDes Pregnancy desire ProgStop Programmed stop RRMSE Relative root mean squared error RRMS Relapsing-remitting multiple sclerosis SAE Serious adverse events SDM Shared Decision-Making SFSEP French Multiple Sclerosis Society TERI Teriflunomide TRIPOD + AI Transparent Reporting of a multivariable prediction model for individual Prognosis or Diagnosis + Artificial Intelligence TxHol Drug holiday TxNaive Therapy naive Declarations Acknowledgements OFSEP Data collection has been supported by a grant provided by the French State and handled by the “Agence Nationale de la Recherche," within the framework of the "France 2030" programme, under the reference ANR-10-COHO-002, Observatoire Français de la Sclérose en Plaques (OFSEP)/Eugène Devic EDMUS Foundation against multiple sclerosis. We also thank Nikolaus von Bomhard and Marcel Müller for their technical support in providing a secure server for publishing the Shiny app. ChatGPT was used to polish the text and R code. Authors’ contributions Anna-Maria Sakr : Data curation (lead), formal analysis (lead), methodology (equal), software (lead), visualization (lead), Conceptualization (equal), writing - original draft (equal), writing - review and editing (equal), Ulrich Mansmann : Conceptualization (equal), funding acquisition (lead), methodology (equal), formal analysis (supporting), supervision (lead), visualization (supporting), writing- original draft (equal), writing - review and editing (equal), Joachim Havla : Conceptualization (supporting), writing - review and editing (supporting). Additional Information Data availability The data supporting the results of this study cannot be made publicly available due to legal constraints. The data were obtained from the Observatoire Français de la Sclérose en Plaques (OFSEP), which served as our primary data source. Our specific data request is documented at https://www.ofsep.org/fr/etudes/extval-phrend and accessible at the OSF directory https://osf.io/yf3ks/files/osfstorage. Interested researchers may obtain the corresponding OFSEP dataset directly from the data-provider by using our data request. The analyses can be replicated using the analysis scripts (freely accessible in the public repository Gitlab, at https://gitlab.lrz.de/asakr/switch-ms; permanent link in case of relocation of the repository at switchms-code.annamariasakr.de) as well as the methodological details provided in the manuscript and supplementary materials. The authors did not have any special access privileges and the dataset can be requested by any qualified researcher through the standard application process. Access to the OFSEP data requires completion of a data use agreement with OFSEP. The corresponding application is accessible at: https://shiny.ibe.med.uni-muenchen.de/switch-ms (username: msswitch, password: Roox3iey). To use the calculator please push on the tab “Outcome predictions and Treatment recommendations “. Funding Anna Maria Sakr was supported by the Data integration for Future Medicine (DIFUTURE) grant (Bundesministerium für Bildung und Forschung (BMBF) 01ZZ1804C). The funding bodies played no role in the design of the study and collection, analysis, and interpretation of data and in writing the manuscript. Competing interests JH reports a grant for OCT research from the Friedrich‐Baur‐Stiftung, Horizon, Sanofi and Merck, personal fees and nonfinancial support from Alexion, Amgen, Bayer, Biogen, BMS, Merck, Novartis and Roche, and nonfinancial support of the Sumaira‐Foundation and Guthy‐Jackson Charitable Foundation, all outside the submitted work. AMS und UM declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article. Ethics approval The study was approved by the OFSEP Steering Committee (approval no. 0266) and the Ethics Committee of Ludwig-Maximilians-University Munich (approval no. 21-1174). All procedures were conducted in accordance with applicable ethical standards, guidelines, and regulations. The study is reported in accordance with the TRIPOD-AI statement. Consent to participate All participants in the OFSEP registry provided written informed consent through their treating physicians. The study was announced on the OFSEP website (https://www.ofsep.org/fr/etudes/extval-phrend) prior to initiation, allowing patients to provide project-specific dynamic consent. Consent for publication Not applicable References Walton, C. et al. Rising prevalence of multiple sclerosis worldwide: Insights from the Atlas of MS, third edition. Mult Scler 26 , 1816–1821 (2020). Koch-Henriksen, N. & Magyari, M. Apparent changes in the epidemiology and severity of multiple sclerosis. Nat Rev Neurol 17 , 676–688 (2021). Dunn, S. E., Gunde, E. & Lee, H. Sex-Based Differences in Multiple Sclerosis (MS): Part II: Rising Incidence of Multiple Sclerosis in Women and the Vulnerability of Men to Progression of this Disease. in Emerging and Evolving Topics in Multiple Sclerosis Pathogenesis and Treatments (eds AC, F. & JM, O.) vol. 2015 57–86 (Springer International Publishing). Reeve, K. et al. Prognostic models for predicting clinical disease progression, worsening and activity in people with multiple sclerosis. Cochrane Database of Systematic Reviews 2023 , (2023). Havla, J., Reeve, K., On, B. I., Mansmann, U. & Held, U. Prognostic models in multiple sclerosis: progress and challenges in clinical integration. Neurol. Res. Pract. 6 , 44 (2024). Sakr, A. M., Mansmann, U., Havla, J., Ön, B. I. & Ön, B. I. Framework for personalized prediction of treatment response in relapsing-remitting multiple sclerosis: a replication study in independent data. BMC Med Res Methodol 24 , 138 (2024). Stühler, E. et al. Framework for personalized prediction of treatment response in relapsing remitting multiple sclerosis. BMC medical research methodology 20 , 24 (2020). Braune, S., Stuehler, E., Heer, Y., van Hoevell, P. & Bergmann, A. PHREND(®)-A Real-World Data-Driven Tool Supporting Clinical Decisions to Optimize Treatment in Relapsing-Remitting Multiple Sclerosis. Frontiers in digital health 4 , 856829 (2022). Van Calster, B., Steyerberg, E. W., Wynants, L. & Van Smeden, M. There is no such thing as a validated prediction model. BMC Med 21 , 70 (2023). McDonald, W. I. et al. Recommended diagnostic criteria for multiple sclerosis: Guidelines from the International Panel on the Diagnosis of Multiple Sclerosis. Annals of Neurology 50 , 121–127 (2001). Polman, C. H. et al. Diagnostic Criteria for Multiple Sclerosis: 2005 Revisions to the ‘McDonald Criteria’. https://doi.org/10.1002/ana.206703 (2005) doi:10.1002/ana.206703. Polman, C. H. et al. Diagnostic criteria for multiple sclerosis: 2010 Revisions to the McDonald criteria. Annals of Neurology 69 , 292–302 (2011). Thompson, A. J. et al. Diagnosis of multiple sclerosis: 2017 revisions of the McDonald criteria. The Lancet Neurology 17 , 162–173 (2018). Montalban, X. et al. Diagnosis of multiple sclerosis: 2024 revisions of the McDonald criteria. The Lancet Neurology 24 , 850–865 (2025). Portaccio, E. et al. Multiple sclerosis: emerging epidemiological trends and redefining the clinical course. The Lancet Regional Health - Europe 44 , 100977 (2024). Arredondo-Robles, A. V., Rodríguez-López, K. P. & Ávila-Avilés, R. D. Clinical Management in Multiple Sclerosis. Neuroglia 6 , 6 (2025). Müller, J. et al. Harmonizing Definitions for Progression Independent of Relapse Activity in Multiple Sclerosis: A Systematic Review. JAMA Neurol 80 , 1232 (2023). Debray, T. P. A. et al. A guide to systematic review and meta-analysis of prediction model performance. BMJ i6460 (2017) doi:10.1136/bmj.i6460. Vukusic, S. et al. Observatoire Français de la Sclérose en Plaques (OFSEP): A unique multimodal nationwide MS registry in France. Multiple Sclerosis Journal 26 , 118–122 (2020). McElreath, R. Statistical Rethinking: A Bayesian Course with Examples in R and Stan . (CRC Press, Boca Raton London New York, 2020). Gelman, A., Carlin, J. B. & Stern, H. S. Bayesian Data Analysis Third edition ( with errors fixed as of 15 February 2021 ). (2021). Corboy, J. R., Weinshenker, B. G. & Wingerchuk, D. M. Comment on 2018 American Academy of Neurology guidelines on disease-modifying therapies in MS. Neurology 90 , 1106–1112 (2018). Koch-Henriksen, N. & Magyari, M. Apparent changes in the epidemiology and severity of multiple sclerosis. Nat Rev Neurol 17 , 676–688 (2021). Rae-Grant, A. et al. Comprehensive systematic review summary: Disease-modifying therapies for adults with multiple sclerosis: Report of the Guideline Development, Dissemination, and Implementation Subcommittee of the American Academy of Neurology. Neurology 90 , 789–800 (2018). Rae-Grant, A. et al. Practice guideline recommendations summary: Disease-modifying therapies for adults with multiple sclerosis: Report of the Guideline Development, Dissemination, and Implementation Subcommittee of the American Academy of Neurology. Neurology 90 , 777–788 (2018). Montalban, X. et al. ECTRIMS/EAN Guideline on the pharmacological treatment of people with multiple sclerosis. Multiple Sclerosis 24 , 96–120 (2018). Androdias, G. et al. De-escalating and discontinuing disease-modifying therapies in multiple sclerosis. Brain 148 , 1459–1478 (2025). Hemmer, B., Gehring, K. & et al. Diagnose und Therapie der Multiplen Sklerose, Neuromyelitis-optica-Spektrum-Erkrankungen und MOG-IgG-assoziierten Erkrankungen, S2k-Leitlinie, 2024, in: Deutsche Gesellschaft für Neurologie (Hrsg.), Leitlinien für Diagnostik und Therapie in der Neurologie. Online: www.dgn.org/leitlinien. Accessed on 19 November 2025. https://www.dgn.org/leitlinie/diagnose-und-therapie-der-multiplen-sklerose-neuromyelitis-optica-spektrum-erkrankungen-und-mog-igg-assoziierten-erkrankungen (2024). Filippi, M. et al. The use of high-efficacy disease-modifying therapies in multiple sclerosis: recommendations from an expert Delphi consensus. J Neurol 272 , 565 (2025). Ontaneda, D. et al. Determining the effectiveness of early intensive versus escalation approaches for the treatment of relapsing-remitting multiple sclerosis: The DELIVER-MS study protocol. Contemporary Clinical Trials 95 , 106009 (2020). Mowry, E. M. et al. The TRaditional versus Early Aggressive Therapy for MS (TREAT-MS) trial: Design and baseline characteristics of participants. Contemporary Clinical Trials 108117 (2025) doi:10.1016/j.cct.2025.108117. Selmaj, K., Cree, B. A. C., Barnett, M., Thompson, A. & Hartung, H.-P. Multiple sclerosis: time for early treatment with high-efficacy drugs. J Neurol 271 , 105–115 (2024). Harding, K. et al. Clinical Outcomes of Escalation vs Early Intensive Disease-Modifying Therapy in Patients With Multiple Sclerosis. JAMA Neurol 76 , 536 (2019). Merkel, B., Butzkueven, H., Traboulsee, A. L., Havrdova, E. & Kalincik, T. Timing of high-efficacy therapy in relapsing-remitting multiple sclerosis: A systematic review. Autoimmunity Reviews 16 , 658–665 (2017). Singer, B. A., Feng, J. & Chiong-Rivero, H. Early use of high-efficacy therapies in multiple sclerosis in the United States: benefits, barriers, and strategies for encouraging adoption. J Neurol 271 , 3116–3130 (2024). Leblanc, S., Roux, J., Tillaut, H., Le Page, E. & Leray, E. Disease-modifying therapy usage in patients with multiple sclerosis in France: A 6-year population-based study. Revue Neurologique 177 , 1250–1261 (2021). Papukchieva, S. et al. Shifting from the treat-to-target to the early highly effective treatment approach in patients with multiple sclerosis – real-world evidence from Germany. Ther Adv Neurol Disord 17 , 17562864241237857 (2024). Stratil, A.-S., Papukchieva, S., Joschko, N., Meuth, S. G. & Friedrich, B. Shifts in treatment initiation patterns among newly diagnosed multiple sclerosis patients in Germany: a claims data analysis from 2017 to 2022. Therapeutic Advances in Neurological Disorders 18 , 17562864251328576 (2025). Tallantyre, E. C. et al. Real‐world persistence of multiple sclerosis disease‐modifying therapies. Euro J of Neurology 31 , e16289 (2024). European Medicines Agency. Ocrevus : EPAR - Product Information. https://www.ema.europa.eu/en/medicines/human/EPAR/ocrevus. Accessed on 13 November 2025. Wang, Y., Wang, J. & Feng, J. Multiple sclerosis and pregnancy: Pathogenesis, influencing factors, and treatment options. Autoimmunity Reviews 22 , 103449 (2023). Vukusic, S. et al. Pregnancy and multiple sclerosis: 2022 recommendations from the French multiple sclerosis society. Mult Scler 29 , 11–36 (2023). Bigaut, K. et al. How to switch disease-modifying treatments in multiple sclerosis: Guidelines from the French Multiple Sclerosis Society (SFSEP). Multiple Sclerosis and Related Disorders 53 , 103076 (2021). Chalkou, K., Vickers, A. J., Pellegrini, F., Manca, A. & Salanti, G. Decision Curve Analysis for Personalized Treatment Choice between Multiple Options. Medical Decision Making 43 , 337–349 (2023). Trojano, M. et al. Big multiple sclerosis data network: novel modelling approaches for real-world data analysis. J Neurol 272 , 754 (2025). European Medicines Agency. Lemtrada: EPAR - Product Information. https://www.ema.europa.eu/en/medicines/human/EPAR/lemtrada. Accessed on 17 November 2025. Col, N. F. et al. Implementing Shared Decision-Making for Multiple Sclerosis: The MS-SUPPORT Tool. Multiple Sclerosis and Related Disorders 80 , 105092 (2023). He, D. et al. Teriflunomide for multiple sclerosis. Cochrane Database of Systematic Reviews 2016 , (2016). Xu, Z. et al. Dimethyl fumarate for multiple sclerosis. Cochrane Database of Systematic Reviews 2015 , (2015). Gonzalez-Lorenzo, M. et al. Immunomodulators and immunosuppressants for relapsing-remitting multiple sclerosis: a network meta-analysis. Cochrane Database of Systematic Reviews 2024 , (2024). Committee for Medicinal Products for Human Use (CHMP) (2019) Guideline on the investigation of subgroups in confirmatory clinical trials. European Medicines Agency. https://www.ema.europa.eu/en/investigation-subgroups-confirmatory-clinical-trials-scientific-guideline. Accessed on 19 November 2025. Collins, G. S. et al. TRIPOD+AI statement: updated guidance for reporting clinical prediction models that use regression or machine learning methods. BMJ e078378 (2024) doi:10.1136/bmj-2023-078378. Chitnis, T., Giovannoni, G. & Trojano, M. Complexity of MS management in the current treatment era. Neurology 90 , 761–762 (2018). Ghezzi, A. European and American Guidelines for Multiple Sclerosis Treatment. Neurol Ther 7 , 189–194 (2018). OFSEP. https://www.ofsep.org/en/. Confavreux, C., Compston, D. A., Hommes, O. R., McDonald, W. I. & Thompson, A. J. EDMUS, a European database for multiple sclerosis. J Neurol Neurosurg Psychiatry 55 , 671–676 (1992). Riley, R. D. et al. Minimum sample size for developing a multivariable prediction model: PART II - binary and time-to-event outcomes. Statistics in Medicine 38 , 1276–1296 (2019). Binuya, M. A. E., Engelhardt, E. G., Schats, W., Schmidt, M. K. & Steyerberg, E. W. Methodological guidance for the evaluation and updating of clinical prediction models: a systematic review. BMC Medical Research Methodology 22 , 1–14 (2022). Janssen, K. J. M., Moons, K. G. M., Kalkman, C. J., Grobbee, D. E. & Vergouwe, Y. Updating methods improved the performance of a clinical prediction model in new patients. Journal of Clinical Epidemiology 61 , 76–86 (2008). Steyerberg, E. W., Borsboom, G. J. J. M., van Houwelingen, H. C., Eijkemans, M. J. C. & Habbema, J. D. F. Validation and updating of predictive logistic regression models: A study on sample size and shrinkage. Statistics in Medicine 23 , 2567–2586 (2004). On behalf of Topic Group ‘Evaluating diagnostic tests and prediction models’ of the STRATOS initiative et al. Calibration: the Achilles heel of predictive analytics. BMC Med 17 , 230 (2019). Van Calster, B. et al. A calibration hierarchy for risk models was defined: from utopia to empirical data. J Clin Epidemiol 74 , 167–176 (2016). Steyerberg, E. W. Clinical Prediction Models. A Practical Approach to Development, Validation, and Updating. Statistics for Biology and Health. (2019). Additional Declarations Competing interest reported. JH reports a grant for OCT research from the Friedrich‐Baur‐Stiftung, Horizon, Sanofi and Merck, personal fees and nonfinancial support from Alexion, Amgen, Bayer, Biogen, BMS, Merck, Novartis and Roche, and nonfinancial support of the Sumaira‐Foundation and Guthy‐Jackson Charitable Foundation, all outside the submitted work. AMS und UM declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article. Supplementary Files Supplementary.docx Supplementary material is available online. Cite Share Download PDF Status: Under Review Version 1 posted Reviewers invited by journal 01 Apr, 2026 Editor assigned by journal 31 Mar, 2026 Editor invited by journal 18 Dec, 2025 Submission checks completed at journal 16 Dec, 2025 First submitted to journal 16 Dec, 2025 You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-8295331","acceptedTermsAndConditions":true,"allowDirectSubmit":false,"archivedVersions":[],"articleType":"Article","associatedPublications":[],"authors":[{"id":617274629,"identity":"8bec9bba-3e8b-4ec0-9d73-52bc1fe96f3e","order_by":0,"name":"Anna Maria Sakr","email":"","orcid":"","institution":"Ludwig-Maximilians-Universität München","correspondingAuthor":false,"prefix":"","firstName":"Anna","middleName":"Maria","lastName":"Sakr","suffix":""},{"id":617274633,"identity":"41ee8072-0ddc-41f0-9b06-b65cbd7afc3d","order_by":1,"name":"Joachim Havla","email":"","orcid":"","institution":"LMU Klinikum","correspondingAuthor":false,"prefix":"","firstName":"Joachim","middleName":"","lastName":"Havla","suffix":""},{"id":617274634,"identity":"a42a39a5-99e8-4477-938e-6af5ffef1f2d","order_by":2,"name":"Ulrich Mansmann","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAABNklEQVRIie2RP0vEMBiHEwp3S9S1Uu7uEwgpgRSp+FkaCtclLYKLg3CFg3MRu97HOBAytwS8pcW1g4MgOHU4PRBB/JNWEa3ngZtDniEJP/LkfZMAoNH8Z7YM1MwwVkP6njWJ8auyPf6zgtMvClin7JwU8r48vorIfCO7hnyvF5thlkIu+063yG4OgNtvKTSPhha/uD2kctPHUAxJbEZeCoUku6eRT6YgIG0l5dTgHcmERNRUO1lsciyf1WJWImohoJKWclmRJX+R7HyMnEcoXhulrjJSivOklFFbKTm2wok600AUQJF+Kh5WVdTLS6/dWFlRKzyTbKruYjLhkwmqaiWwZzknFsKB/aMxrhp7kCxJimxxJ/Z7SZeTBRTuAM9ze4mO3MHqX/mg7qHzPcJrBY1Go9Gs5g3T0HQmnwpiKAAAAABJRU5ErkJggg==","orcid":"","institution":"Ludwig-Maximilians-Universität München","correspondingAuthor":true,"prefix":"","firstName":"Ulrich","middleName":"","lastName":"Mansmann","suffix":""}],"badges":[],"createdAt":"2025-12-06 14:38:09","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-8295331/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-8295331/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":106403873,"identity":"d46f29d9-68ca-4e4d-8412-8213ac667712","added_by":"auto","created_at":"2026-04-08 09:15:07","extension":"jpg","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":702070,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eBayesian models as clinical decision support tools\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eClinicians typically lack direct access to large patient registries for case-based learning. Instead, models are used to abstract these complex data structures and present them to the clinician in an accessible, clinically meaningful format. Here, we conceptually illustrate how our models reflect the clinical experience embedded in the data and generate treatment recommendations as a ranked list of medications. This figure is not focused on a specific clinical scenario but rather sets the foundation for understanding how the models operate.\u003c/p\u003e","description":"","filename":"Fig1.jpg","url":"https://assets-eu.researchsquare.com/files/rs-8295331/v1/b467e641309854d182c07b28.jpg"},{"id":106309872,"identity":"d4896b13-20fe-4d1b-a938-72817fd6f96e","added_by":"auto","created_at":"2026-04-07 10:20:29","extension":"png","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":766381,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eActual switches and individual treatment recommendations for the 3m-CDP and relapse outcomes in the intolerance and lack of efficacy subgroups\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eFor each outcome and subgroup, the recommended therapy switch is stratified by the previously taken therapy. For example, patients treated with TERI who needed a switch due to intolerance mainly received DMF (51%). As the top-ranked therapy, the 3m-CDP model recommended staying on TERI (57%) or switching to IF (31%). The relapse model recommended a switch to NA (92%).\u003c/p\u003e","description":"","filename":"Fig2.png","url":"https://assets-eu.researchsquare.com/files/rs-8295331/v1/134821749e94ccd635e890e4.png"},{"id":106405587,"identity":"981fdb87-d24b-48f3-9e4c-f45eff303114","added_by":"auto","created_at":"2026-04-08 09:27:39","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":3987741,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-8295331/v1/5d769583-09fa-4dae-91a8-532dad89ff2c.pdf"},{"id":106404376,"identity":"921d4eb3-868a-4f00-b7b0-0b206578e987","added_by":"auto","created_at":"2026-04-08 09:15:54","extension":"docx","order_by":1,"title":"","display":"","copyAsset":false,"role":"supplement","size":5104755,"visible":true,"origin":"","legend":"\u003cp\u003eSupplementary material is available online.\u003c/p\u003e","description":"","filename":"Supplementary.docx","url":"https://assets-eu.researchsquare.com/files/rs-8295331/v1/25814674a3eac30dd374a758.docx"}],"financialInterests":"Competing interest reported. JH reports a grant for OCT research from the Friedrich‐Baur‐Stiftung, Horizon, Sanofi and Merck, personal fees and nonfinancial support from Alexion, Amgen, Bayer, Biogen, BMS, Merck, Novartis and Roche, and nonfinancial support of the Sumaira‐Foundation and Guthy‐Jackson Charitable Foundation, all outside the submitted work. AMS und UM declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.","formattedTitle":"Validating therapy decisions by reasons for therapy switch in relapsing-remitting multiple sclerosis","fulltext":[{"header":"Introduction","content":"\u003cdiv id=\"Sec2\" class=\"Section2\"\u003e \u003ch2\u003eDisease Context and Burden\u003c/h2\u003e \u003cp\u003eMultiple sclerosis is a common chronic autoimmune disease of the central nervous system, characterized by demyelinating inflammatory lesions and progressive neurodegeneration leading to disability. In 2020, the estimated global burden of MS was approximately 2.8\u0026nbsp;million people (35.9 per 100,000), showing a continued increase in both prevalence and incidence compared with 2013 [1]. The rise in MS incidence varies across disease phenotypes and between sex. A higher incidence is observed for relapsing-remitting multiple sclerosis (RRMS) than for primary progressive MS (PPMS)\u0026mdash;the latter remaining stable or declining\u0026mdash;and more in women than in men [2]. RRMS predominantly affects younger adults and women compared to men [1,3] reflecting both biological and possibly sociocultural determinants of disease risk and progression.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec3\" class=\"Section2\"\u003e \u003ch2\u003ePrognostic Prediction Models in MS: A Critical Gap\u003c/h2\u003e \u003cp\u003eDespite extensive research into disease mechanisms and therapy, prognostic prediction models for MS are not yet integrated into clinical practice. A recent Cochrane review concluded that no well-developed and independently externally validated prognostic models currently exist [4,5]. Yet such models are urgently needed, as they can support individualized clinical decision-making by predicting the heterogeneous outcomes among persons with MS (PwMS).\u003c/p\u003e \u003cp\u003eWe [6] recently developed and validated two prognostic models following the methodology of a promising commercial black-box tool, that used data from the NeuroTransData MS registry, designed to predict relapse and disease progression under specific treatments [7]. The original tool aimed to support physicians in optimal therapy selection in persons with RRMS (PwRRMS) [8]. Our initial goal was to externally validate those published models in new data; however, because model coefficients were proprietary and unavailable, we conducted an exact replication study to validate the methodological framework.\u003c/p\u003e \u003cp\u003eBefore such models can be implemented in clinical practice, multiple external validations across diverse settings, time periods, and patient populations are essential [9]. Changes in diagnostic criteria [10\u0026ndash;14], epidemiological trends [2,15], disease course and classification [15], therapeutic management [16], and predictor or outcome definitions [17], can all lead to heterogeneous model performance [18]. Independent validations therefore increase confidence in a model\u0026rsquo;s generalizability and transportability to real-world clinical care.\u003c/p\u003e \u003c/div\u003e\n\u003ch3\u003eStudy Population and Modelling Approach\u003c/h3\u003e\n\u003cp\u003eOur study included adult PwRRMS from the French OFSEP registry (Observatoire Fran\u0026ccedil;ais de la Scl\u0026eacute;rose en Plaques) [19] who initiated or switched, between January 2011 and December 2021, to six at this time commonly used disease-modifying therapies (DMTs): Interferon beta (IF), glatiramer acetate (GA), teriflunomide (TERI), dimethyl fumarate (DMF), fingolimod (FTY), or natalizumab (NA). Pre-switch clinical histories and demographic factors were used to predict key clinical outcomes: the number of relapses and the occurrence of disease progression confirmed at three months (3m-CDP). Therapy cycles starting before January 2017 served model development and internal validation, while those after were used for temporal external validation. By incorporating interactions between therapy types and patient characteristics, the models provide personalized predictions under each therapy, producing list of therapies ranked by predicted outcome. This can aid neurologists and future patients with similar histories to select the most promising treatment.\u003c/p\u003e \u003cp\u003eWe used a Bayesian machine learning algorithm that makes probabilistically informed predictions, rather than just a single educated guess [20]. The algorithm accounts for the variability in clinical practice and therefore may recommend multiple plausible treatment rankings. Presenting alternative rankings influences medical decision-making and counselling. Figure\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003e illustrates how our models abstract clinical experience from data and generate treatment recommendations as ranked lists. Bayesian models also quantify individual treatment recommendation (ITR) certainty [21], distinguishing between patients with clear, unique recommendations (scenario 1, easier to treat) and those with broader treatment uncertainty (scenario 2, more complex cases).\u003c/p\u003e \u003cp\u003e \u003c/p\u003e\n\u003ch3\u003eGuidelines Context\u003c/h3\u003e\n\u003cp\u003eOptimal therapy selection remains an unmet need [22] with major implications for multiple sclerosis morbidity and mortality [23]. The European (ECTRIMS/EAN) and the American Academy of Neurology (AAN) guidelines published in 2018\u0026mdash; the first comprehensive reviews of DMT use [24\u0026ndash;26]\u0026mdash;were applicable to most of our validation period and likely informed the clinical practice recorded in the OFSEP registry. These guidelines emphasize individualized decision-making, primarily guided by patient characteristics, comorbidities, disease activity, drug safety, cost and accessibility, and patient preferences, rather than a single optimal strategy for all. In addition, they provide mainly consensus-based recommendations given the lack of evidence on most of the addressed clinical questions [22]. Switching DMTs is advised for inadequate response, intolerance, poor adherence, safety concerns, or pregnancy planning; escalation for lack of response or breakthrough disease activity; and de-escalation for adverse effects, advancing age, long disease duration, advanced disability or pregnancy [27]. Oral therapies may replace injectables in case of intolerance. (see Supplementary Table \u003cspan refid=\"MOESM1\" class=\"InternalRef\"\u003eS1\u003c/span\u003e).\u003c/p\u003e \u003cp\u003eUnlike ECTRIMS/EAN which specifies no strategy, the AAN recommends alemtuzumab, fingolimod, and natalizumab as possible initial therapies for highly active MS. For those switching due to breakthrough activity, it highlights evidence supporting alemtuzumab, fingolimod, natalizumab, and ocrelizumab, while advising consideration of individual factors such as adherence, safety, degree of disease activity, and the DMTs\u0026rsquo; mechanisms of action. National guidelines, including the 2024 version of the German Neurological Society (DGN) [28] and an Italian Delphi [29] similarly endorse early high-efficacy therapy for high-risk or therapy escalation for suboptimally-controlled patients.\u003c/p\u003e\n\u003ch3\u003eEvolving Treatment Paradigms and Real-World Practice\u003c/h3\u003e\n\u003cp\u003eThe treatment paradigm of RRMS is evolving. Two ongoing randomized controlled trials\u0026mdash;DELIVER-MS [30] (NCT03535298) and TREAT-MS [31] (NCT03500328)\u0026mdash;are comparing the effectiveness of early high-efficacy treatment (EHT) versus escalation in RRMS. Observational studies increasingly support the benefits of early use of high-efficacy DMTs for long-term disease control [32\u0026ndash;34], yet real-world implementation remains heterogeneous [35].\u003c/p\u003e \u003cp\u003eA population based-analysis of PwMS in France from 2010 to 2015 [36], designed to describe real-world DMT use, found that 57% of patients initiated treatment with first-line DMTs (i.e., low efficacy), 13% received second-line agents (i.e., moderate or high efficacy; either following escalation or as EHT), 7% received off-label DMTs (likely reflecting use in progressive MS) and 23% discontinued therapy early (presumably due to benign or evolved disease). The most frequently prescribed DMTs during this period were IF, GA, NA, FTY, DMF, and TERI.\u003c/p\u003e \u003cp\u003eAnalyses of German claims data (2017\u0026ndash;2022; 2020\u0026ndash;2022) [37,38] found that two thirds of patients still initiated low- or moderate-efficacy therapy, with only gradual increases in high-efficacy DMT use, mainly in PPMS, while their use in RRMS remained relatively stable [38]. This pattern indicates that while the therapeutic shift toward high-efficacy agents is underway, it remains recent and uneven across disease phenotypes.\u003c/p\u003e\n\u003ch3\u003eStudy Aim and Scope\u003c/h3\u003e\n\u003cp\u003eWhile current guidelines emphasize individualized therapy selection, clinical decision-making in practice remains complex. This study examines the potential of the replicated prognostic models to support treatment-switch decisions in PwRRMS, integrating both guidelines\u0026rsquo; principles and patient perspectives. Although the models do not explicitly incorporate the reason for therapy switch, the OFSEP registry documents several reasons that enable model exploration across real-world decision contexts: \u003cem\u003elack of efficacy (NoEff), intolerance (Intolerance), programmed stop (ProgStop\u003c/em\u003e), \u003cem\u003epersonal convenience (Convenience), pregnancy desire (PregDes)\u003c/em\u003e, and \u003cem\u003eserious adverse events (SAE\u003c/em\u003e). In addition, two derived categories capture transitions into therapy: ending \u003cem\u003etherapy naivety (TxNaive)\u003c/em\u003e and ending a \u003cem\u003edrug holiday (TxHol)\u003c/em\u003e.\u003c/p\u003e \u003cp\u003eTo illustrate how these models could inform real-world therapy decisions, Table\u0026nbsp;\u003cspan refid=\"Tab1\" class=\"InternalRef\"\u003e1\u003c/span\u003e presents three representative therapy-switch scenarios that exemplify common clinical situations captured in the OFSEP registry and highlight the types of questions this study aims to address.\u003c/p\u003e \u003cp\u003eThis study aims to further validate the previously developed models across diverse patient profiles by investigating model behaviour across four key areas: (1) the robustness of individualized treatment recommendations (ITRs), (2) the alignment of ITRs with therapy switching reasons and guideline, (3) the comparability of prognostic performance across switcher subgroups, and (4) the agreement between relapse and progression-specific ITRs.\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab1\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 1\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eRepresentative therapy switches\u0026rsquo; scenarios, captured in the OFSEP registry, illustrating the clinical relevance of individualized treatment recommendations\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"4\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eScenario\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eClinical context\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003ePatient characteristics\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003eClinical question\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eLack of efficacy\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eEscalation after recent relapse\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e30-year-old woman; 6-year RRMS; teriflunomide for 3 years; no prior therapy; relapse within past 6 months; EDSS 1.5\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eShould treatment be escalated to reduce risk of future activity?\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003ePregnancy desire\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eSwitch for pregnancy planning\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e32-year-old woman; 6-year RRMS; fingolimod for 2 years; no prior therapy: relapse 2 years ago; EDSS 1.5\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eWhich therapy is safest and effective when planning pregnancy?\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003ePersonal convenience\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eSwitch to improve convenience/adherence\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e45-year-old woman; 11-year RRMS; interferon beta for 4.5 years; no prior therapy; relapse\u0026thinsp;\u0026gt;\u0026thinsp;3 years ago; EDSS 1.5\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eIs an oral therapy appropriate given stable disease?\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e"},{"header":"Results","content":"\u003cdiv id=\"Sec9\" class=\"Section2\"\u003e \u003ch2\u003ePatient population\u003c/h2\u003e \u003cp\u003eOf the 3768 patients, 1177 (31%) were TxNaive, 1006 (27%) were on TxHol, 625 (17%) switched due to NoEff, 464 (12%) due to Intolerance, 116 (3%) due to ProgStop, 96 (3%) due to Convenience, 73 (2%) due to PregDes, and 25 (0.7%) due to SAE. Additionally, 102 (2.7%) had combined reasons, and 84 (2%) had a missing or unknown reason (Supplementary Fig. S2). Supplementary Fig. S3 shows that Intolerance and NoEff switches mainly occurred from platform DMTs, while ProgStop primarily from NA. Convenience and PregDes switches were mainly from IFN and FTY, respectively. Actual switches by subgroup can be found in Supplementary Table S3, Supplementary Table S4, and Supplementary Fig. S4.\u003c/p\u003e \u003cp\u003eTable\u0026nbsp;\u003cspan refid=\"Tab2\" class=\"InternalRef\"\u003e2\u003c/span\u003e summarizes predictors and clinical outcomes. Most patients were female and under 50, except for the SAE subgroup, where 32% were aged 50 or older. Most patients had not previously received second-line therapies, except for 74% in Progstop. TxNaive patients had a median time to therapy initiation of 2.1 years (range: 0.50 to 52), while median therapy discontinuation duration was 0.8 years (range: 0.25 to 22). The Convenience subgroup had the longest previous therapy duration (median: 4.4 years, range: 0.19 to 19), while the Intolerance subgroup had the shortest (median: 1.6 years, range: 0.04 to 22). In most subgroups, over a year had passed since the last relapse for the majority of patients, except for NoEff (48%) and TxNaive (56%).\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab2\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 2\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eOverview of predictors at baseline and clinical outcomes at baseline in the development set (Dev Set), the overall validation set (Val Set) and by subgroup of therapy switchers\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"12\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c6\" colnum=\"6\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c7\" colnum=\"7\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c8\" colnum=\"8\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c9\" colnum=\"9\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c10\" colnum=\"10\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c11\" colnum=\"11\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c12\" colnum=\"12\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003ePredictor\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003eDev Set\u003c/p\u003e \u003cp\u003e(N\u0026thinsp;=\u0026thinsp;5517)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003eVal Set\u003c/p\u003e \u003cp\u003e(N\u0026thinsp;=\u0026thinsp;3768)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c5\"\u003e \u003cp\u003eTxHol\u003c/p\u003e \u003cp\u003e(N\u0026thinsp;=\u0026thinsp;1006)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c6\"\u003e \u003cp\u003eIntolerance (N\u0026thinsp;=\u0026thinsp;538)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c7\"\u003e \u003cp\u003eNoEff\u003c/p\u003e \u003cp\u003e(N\u0026thinsp;=\u0026thinsp;674)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c8\"\u003e \u003cp\u003eConvenience (N\u0026thinsp;=\u0026thinsp;126)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c9\"\u003e \u003cp\u003ePregDes\u003c/p\u003e \u003cp\u003e(N\u0026thinsp;=\u0026thinsp;85)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c10\"\u003e \u003cp\u003eProgStop\u003c/p\u003e \u003cp\u003e(N\u0026thinsp;=\u0026thinsp;137)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c11\"\u003e \u003cp\u003eSAE\u003c/p\u003e \u003cp\u003e(N\u0026thinsp;=\u0026thinsp;31)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c12\"\u003e \u003cp\u003eTxNaive\u003c/p\u003e \u003cp\u003e(N\u0026thinsp;=\u0026thinsp;1177)\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eFemale\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e4,155 (75%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e2,864 (76%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e838 (83%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e412 (77%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e498 (74%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003e84 (67%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c9\"\u003e \u003cp\u003e84 (99%)\u003csup\u003ea\u003c/sup\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c10\"\u003e \u003cp\u003e98 (72%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c11\"\u003e \u003cp\u003e21 (68%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c12\"\u003e \u003cp\u003e836 (71%)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eAge\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e\u0026le;\u0026thinsp;30\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e916 (17%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e728 (19%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e168 (17%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e82 (15%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e152 (23%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003e20 (16%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c9\"\u003e \u003cp\u003e23 (27%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c10\"\u003e \u003cp\u003e17 (12%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c11\"\u003e \u003cp\u003e5 (16%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c12\"\u003e \u003cp\u003e275 (23%)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e31 to 40\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e1,843 (33%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e1,382 (37%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e424 (42%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e174 (32%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e230 (34%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003e34 (27%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c9\"\u003e \u003cp\u003e57 (67%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c10\"\u003e \u003cp\u003e48 (35%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c11\"\u003e \u003cp\u003e7 (23%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c12\"\u003e \u003cp\u003e418 (36%)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e41 to 50\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e1,724 (31%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e974 (26%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e234 (23%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e164 (30%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e191 (28%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003e40 (32%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c9\"\u003e \u003cp\u003e5 (5.9%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c10\"\u003e \u003cp\u003e40 (29%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c11\"\u003e \u003cp\u003e9 (29%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c12\"\u003e \u003cp\u003e288 (24%)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e\u0026ge;\u0026thinsp;51\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e1,034 (19%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e684 (18%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e180 (18%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e118 (22%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e101 (15%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003e32 (25%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c9\"\u003e \u003cp\u003e0 (0%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c10\"\u003e \u003cp\u003e32 (23%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c11\"\u003e \u003cp\u003e10 (32%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c12\"\u003e \u003cp\u003e196 (17%)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eDMT count\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e0\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e1,435 (26%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e1,177 (31%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e0 (0%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e0 (0%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e0 (0%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003e0 (0%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c9\"\u003e \u003cp\u003e0 (0%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c10\"\u003e \u003cp\u003e0 (0%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c11\"\u003e \u003cp\u003e0 (0%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c12\"\u003e \u003cp\u003e1,177(100%)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e1\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e2,057 (37%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e1,325 (35%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e454 (45%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e307 (57%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e418 (62%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003e72 (57%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c9\"\u003e \u003cp\u003e34 (40%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c10\"\u003e \u003cp\u003e34 (25%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c11\"\u003e \u003cp\u003e14 (45%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c12\"\u003e \u003cp\u003e0 (0%)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e2\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e1,149 (21%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e729 (19%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e306 (30%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e143 (27%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e168 (25%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003e25 (20%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c9\"\u003e \u003cp\u003e32 (38%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c10\"\u003e \u003cp\u003e48 (35%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c11\"\u003e \u003cp\u003e10 (32%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c12\"\u003e \u003cp\u003e0 (0%)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e\u0026ge;\u0026thinsp;3\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e876 (16%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e537 (14%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e246 (24%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e88 (16%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e88 (13%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003e29 (23%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c9\"\u003e \u003cp\u003e19 (22%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c10\"\u003e \u003cp\u003e55 (40%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c11\"\u003e \u003cp\u003e7 (23%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c12\"\u003e \u003cp\u003e0 (0%)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eSecond-line\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e1,051 (19%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e518 (14%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e214 (21%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e61 (11%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e65 (9.6%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003e10 (7.9%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c9\"\u003e \u003cp\u003e47 (55%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c10\"\u003e \u003cp\u003e101 (74%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c11\"\u003e \u003cp\u003e15 (48%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c12\"\u003e \u003cp\u003e0 (0%)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eCurrent therapy\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eDMF\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e150 (2.7%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e333 (8.8%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e0 (0%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e159 (30%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e138 (20%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003e11 (8.7%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c9\"\u003e \u003cp\u003e19 (22%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c10\"\u003e \u003cp\u003e4 (2.9%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c11\"\u003e \u003cp\u003e6 (19%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c12\"\u003e \u003cp\u003e0 (0%)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eIF\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e1,227 (22%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e443 (12%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e0 (0%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e177 (33%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e169 (25%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003e76 (60%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c9\"\u003e \u003cp\u003e4 (4.7%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c10\"\u003e \u003cp\u003e19 (14%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c11\"\u003e \u003cp\u003e3 (9.7%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c12\"\u003e \u003cp\u003e0 (0%)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eGA\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e595 (11%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e244 (6.5%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e0 (0%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e82 (15%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e128 (19%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003e24 (19%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c9\"\u003e \u003cp\u003e3 (3.5%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c10\"\u003e \u003cp\u003e12 (8.8%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c11\"\u003e \u003cp\u003e2 (6.5%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c12\"\u003e \u003cp\u003e0 (0%)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eTERI\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e67 (1.2%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e290 (7.7%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e0 (0%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e72 (13%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e184 (27%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003e7 (5.6%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c9\"\u003e \u003cp\u003e12 (14%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c10\"\u003e \u003cp\u003e7 (5.1%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c11\"\u003e \u003cp\u003e5 (16%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c12\"\u003e \u003cp\u003e0 (0%)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eFTY\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e89 (1.6%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e136 (3.6%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e0 (0%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e29 (5.4%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e47 (7.0%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003e3 (2.4%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c9\"\u003e \u003cp\u003e37 (44%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c10\"\u003e \u003cp\u003e5 (3.6%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c11\"\u003e \u003cp\u003e10 (32%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c12\"\u003e \u003cp\u003e0 (0%)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eNA\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e519 (9.4%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e139 (3.7%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e0 (0%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e19 (3.5%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e8 (1.2%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003e5 (4.0%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c9\"\u003e \u003cp\u003e10 (12%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c10\"\u003e \u003cp\u003e90 (66%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c11\"\u003e \u003cp\u003e5 (16%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c12\"\u003e \u003cp\u003e0 (0%)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eNoDMT\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e2,870 (52%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e2,183 (58%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e1,006 (100%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e0 (0%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e0 (0%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003e0 (0%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c9\"\u003e \u003cp\u003e0 (0%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c10\"\u003e \u003cp\u003e0 (0%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c11\"\u003e \u003cp\u003e0 (0%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c12\"\u003e \u003cp\u003e1,177 (100%)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eCurrent duration\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e2.03 (0.01, 0.86, 5.12, 46.96)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e1.63 (0.04, 0.76, 4.10, 51.82)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e0.80 (0.25, 0.46, 1.44, 22.13)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e1.56 (0.04, 0.60, 3.65, 22.05)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e2.44 (0.12, 1.38, 4.48, 18.82)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003e4.41 (0.19, 1.34, 7.82, 19.27)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c9\"\u003e \u003cp\u003e2.23 (0.12, 1.25, 3.72, 10.42)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c10\"\u003e \u003cp\u003e2.68 (0.16, 1.52, 7.28, 19.99)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c11\"\u003e \u003cp\u003e1.86 (0.06, 0.50, 4.10, 17.47)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c12\"\u003e \u003cp\u003e2.14 (0.50, 0.96, 5.91, 51.82)\u003csup\u003eb\u003c/sup\u003e\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eIndex therapy\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eDMF\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e1,086 (20%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e738 (20%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e195 (19%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e123 (23%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e68 (10%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003e35 (28%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c9\"\u003e \u003cp\u003e5 (5.9%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c10\"\u003e \u003cp\u003e14 (10%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c11\"\u003e \u003cp\u003e9 (29%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c12\"\u003e \u003cp\u003e285 (24%)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eIF\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e901 (16%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e284 (7.5%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e116 (12%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e40 (7.4%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e6 (0.9%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003e6 (4.8%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c9\"\u003e \u003cp\u003e14 (16%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c10\"\u003e \u003cp\u003e3 (2.2%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c11\"\u003e \u003cp\u003e4 (13%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c12\"\u003e \u003cp\u003e99 (8.4%)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eGA\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e663 (12%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e538 (14%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e141 (14%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e69 (13%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e8 (1.2%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003e6 (4.8%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c9\"\u003e \u003cp\u003e50 (59%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c10\"\u003e \u003cp\u003e6 (4.4%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c11\"\u003e \u003cp\u003e4 (13%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c12\"\u003e \u003cp\u003e255 (22%)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eTERI\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e930 (17%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e953 (25%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e261 (26%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e214 (40%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e27 (4.0%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003e64 (51%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c9\"\u003e \u003cp\u003e0 (0%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c10\"\u003e \u003cp\u003e12 (8.8%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c11\"\u003e \u003cp\u003e5 (16%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c12\"\u003e \u003cp\u003e373 (32%)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eFTY\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e1,381 (25%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e767 (20%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e170 (17%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e67 (12%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e341 (51%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003e13 (10%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c9\"\u003e \u003cp\u003e0 (0%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c10\"\u003e \u003cp\u003e90 (66%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c11\"\u003e \u003cp\u003e3 (9.7%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c12\"\u003e \u003cp\u003e84 (7.1%)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eNA\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e556 (10%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e488 (13%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e123 (12%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e25 (4.6%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e224 (33%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003e2 (1.6%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c9\"\u003e \u003cp\u003e16 (19%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c10\"\u003e \u003cp\u003e12 (8.8%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c11\"\u003e \u003cp\u003e6 (19%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c12\"\u003e \u003cp\u003e81 (6.9%)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eIndex duration\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e1.37 (0.00, 0.61, 2.20, 5.91)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e1.29 (0.00, 0.58, 2.41, 4.87)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e1.20 (0.00, 0.52, 2.23, 4.87)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e1.19 (0.01, 0.52, 2.26, 4.85)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e1.63 (0.01, 0.85, 2.70, 4.80)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003e1.66 (0.03, 0.82, 2.52, 4.55)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c9\"\u003e \u003cp\u003e0.70 (0.02, 0.34, 1.01, 4.55)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c10\"\u003e \u003cp\u003e1.41 (0.03, 0.60, 2.96, 4.79)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c11\"\u003e \u003cp\u003e1.05 (0.03, 0.60, 2.56, 4.73)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c12\"\u003e \u003cp\u003e1.29 (0.01, 0.56, 2.38, 4.70)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eOnset distance\u003csup\u003ec\u003c/sup\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e7.50 (0.50, 3.22, 13.48, 54.96)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e5.79 (0.50, 2.31, 11.91, 51.83)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e8.09 (0.52, 4.74, 13.39, 43.42)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e6.56 (0.53, 2.94, 12.94, 41.52)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e6.17 (0.62, 3.05, 11.92, 41.52)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003e10.97 (0.65, 5.68, 16.20, 35.92)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c9\"\u003e \u003cp\u003e5.58 (1.27, 3.42, 8.47, 17.31)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c10\"\u003e \u003cp\u003e10.82 (0.68, 5.24, 16.96, 39.61)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c11\"\u003e \u003cp\u003e8.27 (0.63, 5.21, 13.12, 41.79)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c12\"\u003e \u003cp\u003e2.14 (0.50, 0.97, 5.92, 51.83)\u003csup\u003ec\u003c/sup\u003e\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eRelapse distance\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e\u0026lt;\u0026thinsp;3 months\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e407 (7.4%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e234 (6.2%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e71 (7.1%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e15 (2.8%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e66 (9.8%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003e3 (2.4%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c9\"\u003e \u003cp\u003e2 (2.4%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c10\"\u003e \u003cp\u003e5 (3.6%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c11\"\u003e \u003cp\u003e0 (0%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c12\"\u003e \u003cp\u003e73 (6.2%)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c2\" namest=\"c1\"\u003e \u003cp\u003e\u0026ge;\u0026thinsp;3 months and \u0026lt;\u0026thinsp;1year\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e1,961 (36%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e1175 (31%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e177 (18%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e114 (21%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e254 (38%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003e15 (12%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c9\"\u003e \u003cp\u003e11 (13%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c10\"\u003e \u003cp\u003e12 (8.8%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c11\"\u003e \u003cp\u003e8 (26%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c12\"\u003e \u003cp\u003e590 (50%)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e1\u0026ndash;3 years\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e1,561 (28%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e1173 (31%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e322 (32%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e190 (35%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e177 (26%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003e28 (22%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c9\"\u003e \u003cp\u003e40 (47%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c10\"\u003e \u003cp\u003e45 (33%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c11\"\u003e \u003cp\u003e9 (29%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c12\"\u003e \u003cp\u003e362 (31%)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e\u0026ge;\u0026thinsp;3 years\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e1,588 (29%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e1186 (31%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e436 (43%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e219 (41%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e177 (26%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003e80 (63%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c9\"\u003e \u003cp\u003e32 (38%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c10\"\u003e \u003cp\u003e75 (55%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c11\"\u003e \u003cp\u003e14 (45%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c12\"\u003e \u003cp\u003e152 (13%)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eBaseline EDSS\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e\u0026le;\u0026thinsp;1.5\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e2,411 (44%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e2040 (54%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e490 (49%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e302 (56%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e366 (54%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003e70 (56%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c9\"\u003e \u003cp\u003e65 (76%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c10\"\u003e \u003cp\u003e55 (40%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c11\"\u003e \u003cp\u003e18 (58%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c12\"\u003e \u003cp\u003e690 (59%)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e2 to 2.5\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e1,336 (24%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e898 (24%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e247 (25%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e127 (24%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e157 (23%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003e26 (21%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c9\"\u003e \u003cp\u003e13 (15%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c10\"\u003e \u003cp\u003e35 (26%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c11\"\u003e \u003cp\u003e6 (19%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c12\"\u003e \u003cp\u003e290 (25%)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e3 to 3.5\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e759 (14%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e434 (12%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e127 (13%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e62 (12%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e78 (12%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003e12 (9.5%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c9\"\u003e \u003cp\u003e3 (3.5%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c10\"\u003e \u003cp\u003e22 (16%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c11\"\u003e \u003cp\u003e1 (3.2%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c12\"\u003e \u003cp\u003e121 (10%)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e4 to 6\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e1,011 (18%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e396 (11%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e142 (14%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e47 (8.7%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e73 (11%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003e18 (14%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c9\"\u003e \u003cp\u003e4 (4.7%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c10\"\u003e \u003cp\u003e25 (18%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c11\"\u003e \u003cp\u003e6 (19%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c12\"\u003e \u003cp\u003e76 (6.5%)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eRelapse count\u003csup\u003ed\u003c/sup\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e0.55 (0.72)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e0.45 (0.63)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e0.29 (0.54)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e0.28 (0.54)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e0.58 (0.68)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003e0.17 (0.43)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c9\"\u003e \u003cp\u003e0.16 (0.40)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c10\"\u003e \u003cp\u003e0.15 (0.45)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c11\"\u003e \u003cp\u003e0.32 (0.65)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c12\"\u003e \u003cp\u003e0.67 (0.67)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eRelapse-free\u003csup\u003ee\u003c/sup\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e3,129 (57%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e2349 (62%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e756 (75%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e408 (76%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e353 (52%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003e108 (86%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c9\"\u003e \u003cp\u003e72 (85%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c10\"\u003e \u003cp\u003e120 (88%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c11\"\u003e \u003cp\u003e23 (74%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c12\"\u003e \u003cp\u003e509 (43%)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eClinical outcomes\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e3m-CDP free\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e5,005 (91%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e3449 (92%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e929 (92%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e497 (92%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e610 (91%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003e117 (93%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c9\"\u003e \u003cp\u003e80 (94%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c10\"\u003e \u003cp\u003e126 (92%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c11\"\u003e \u003cp\u003e28 (90%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c12\"\u003e \u003cp\u003e1,070 (91%)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eRelapse-free\u003csup\u003ef\u003c/sup\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e4,225 (77%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e3177 (84%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e849 (84%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e457 (85%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e584 (87%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003e109 (87%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c9\"\u003e \u003cp\u003e63 (74%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c10\"\u003e \u003cp\u003e116 (85%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c11\"\u003e \u003cp\u003e25 (81%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c12\"\u003e \u003cp\u003e982 (83%)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003cp\u003eTable\u0026nbsp;\u003cspan refid=\"Tab3\" class=\"InternalRef\"\u003e3\u003c/span\u003e compares observed relapse activity pre- and post-switch. The highest mean difference (MD) in annualized relapse rate (ARR) was in TxNaive (MD\u0026thinsp;=\u0026thinsp;0.47, 95% CI: 0.42\u0026ndash;0.51) and NoEff (MD\u0026thinsp;=\u0026thinsp;0.4, 95% CI: 0.34\u0026ndash;0.46). Relapse-free proportions improved in these groups (NoEff: 52% to 87%, odds ratio (OR)\u0026thinsp;=\u0026thinsp;8.45, 95% CI: 5.81\u0026ndash;12.69; TxNaive: 43% to 83%, OR\u0026thinsp;=\u0026thinsp;7.76, 95% CI: 6.04\u0026ndash;10.10). TxHol and Intolerance had similar pre-switch disease activity and comparable post-switch relapse reduction. Convenience had lower relapse activity pre-switch and remained stable post-switch. SAE showed a similar ARR pre- and post-switch but an increased relapse-free proportion. PregDes and ProgStop had pre-switch the lowest ARR (0.16 and 0.15) and highest relapse-free proportions (85% and 88%), but post-switch ARR increased (PregDes: MD = -0.21, 95% CI: -0.39 to -0.04; ProgStop: MD = -0.04, 95% CI: -0.15 to 0.07). Relapse-free proportions decreased in both groups (PregDes: OR\u0026thinsp;=\u0026thinsp;0.50, 95% CI: 0.20\u0026ndash;1.17; ProgStop: OR\u0026thinsp;=\u0026thinsp;0.75, 95% CI: 0.32\u0026ndash;1.69).\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab3\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 3\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eRelapse activity pre- and post-therapy switch.\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"8\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c6\" colnum=\"6\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c7\" colnum=\"7\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c8\" colnum=\"8\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eReason (N)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eMean ARR\u003c/p\u003e \u003cp\u003epre-switch (SD)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003eMean ARR\u003c/p\u003e \u003cp\u003epost-switch (SD)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003eMean difference\u003c/p\u003e \u003cp\u003eof ARR (95% CI)\u003csup\u003ea\u003c/sup\u003e\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c5\"\u003e \u003cp\u003eRelapse-free\u003c/p\u003e \u003cp\u003epre-switch, N (%)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c6\"\u003e \u003cp\u003eRelapse-free\u003c/p\u003e \u003cp\u003epost-switch, N (%)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c7\"\u003e \u003cp\u003eOdds ratio of\u003c/p\u003e \u003cp\u003erelapse-free\u003c/p\u003e \u003cp\u003eproportion\u003c/p\u003e \u003cp\u003e(95% CI)\u003csup\u003eb\u003c/sup\u003e\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c8\"\u003e \u003cp\u003eMost frequent switch,\u003c/p\u003e \u003cp\u003eN (%)\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eTherapy naive\u003c/p\u003e \u003cp\u003e(N\u0026thinsp;=\u0026thinsp;1177)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.67 (0.67)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.21 (0.52)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e0.47 (0.42 to 0.51)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e509 (43.25%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e982 (83.43%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e7.76 (6.04 to 10.10)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003eNoDMT to TERI\u003c/p\u003e \u003cp\u003e373 (31.69%)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eLack of efficacy\u003c/p\u003e \u003cp\u003e(N\u0026thinsp;=\u0026thinsp;674)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.58 (0.68)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.17 (0.48)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e0.40 (0.34 to 0.46)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e353 (52.37%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e584 (86.65%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e8.45 (5.81 to 12.69)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003eTERI to FTY\u003c/p\u003e \u003cp\u003e113 (16.77%)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eOverall validation set\u003c/b\u003e\u003c/p\u003e \u003cp\u003e\u003cb\u003e(N\u0026thinsp;=\u0026thinsp;3768)\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e\u003cb\u003e0.45 (0.63)\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e\u003cb\u003e0.20 (0.51)\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e\u003cb\u003e0.25 (0.22 to 0.27)\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e\u003cb\u003e2349 (62.34%)\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e\u003cb\u003e3177 (84.32%)\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e\u003cb\u003e3.52 (3.11 to 3.99)\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003e\u003cb\u003eNoDMT to TERI\u003c/b\u003e\u003c/p\u003e \u003cp\u003e\u003cb\u003e634 (16.83%)\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eDevelopment set\u003c/p\u003e \u003cp\u003e(N\u0026thinsp;=\u0026thinsp;5517)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.55\u0026nbsp;(0.72)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.33\u0026nbsp;(0.72)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e0.21\u0026nbsp;(0.19\u0026nbsp;to\u0026nbsp;0.24)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e3129\u0026nbsp;(56.72%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e4225\u0026nbsp;(76.58%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e2.87\u0026nbsp;(2.61\u0026nbsp;to\u0026nbsp;3.16)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003eNoDMT\u0026nbsp;to\u0026nbsp;IF\u003c/p\u003e \u003cp\u003e779\u0026nbsp;(14.12%)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eIntolerance\u003c/p\u003e \u003cp\u003e(N\u0026thinsp;=\u0026thinsp;538)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.28 (0.54)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.18 (0.48)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e0.10 (0.04 to 0.16)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e408 (75.84%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e457 (84.94%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e1.82 (1.31 to 2.53)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003eIF to TERI\u003c/p\u003e \u003cp\u003e90 (16.73%)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eDrug holiday\u003c/p\u003e \u003cp\u003e(N\u0026thinsp;=\u0026thinsp;1006)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.29 (0.54)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.20 (0.50)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e0.09 (0.05 to 0.14)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e756 (75.15%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e849 (84.39%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e1.81 (1.43 to 2.29)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003eNoDMT to TERI\u003c/p\u003e \u003cp\u003e261 (25.94%)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003ePersonal convenience\u003c/p\u003e \u003cp\u003e(N\u0026thinsp;=\u0026thinsp;126)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.17 (0.43)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.16 (0.45)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e0.01 (-0.11 to 0.12)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e108 (85.71%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e109 (86.51%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e1.06 (0.50 to 2.25)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003eIF to TERI\u003c/p\u003e \u003cp\u003e48 (38.1%)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eSerious adverse events\u003c/p\u003e \u003cp\u003e(N\u0026thinsp;=\u0026thinsp;31)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.32 (0.65)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.32 (0.75)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e0.00 (-0.35 to 0.35)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e23 (74.19%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e25 (80.65%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e1.50 (0.36 to 7.23)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003eFTY to NA\u003c/p\u003e \u003cp\u003e4 (12.90%)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eProgrammed stop\u003c/p\u003e \u003cp\u003e(N\u0026thinsp;=\u0026thinsp;137)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.15 (0.45)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.19 (0.49)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e-0.04 (-0.15 to 0.07)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e120 (87.59%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e116 (84.67%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e0.75 (0.32 to 1.69)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003eNA to FTY\u003c/p\u003e \u003cp\u003e80 (58.39%)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003ePregnancy desire\u003c/p\u003e \u003cp\u003e(N\u0026thinsp;=\u0026thinsp;85)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.16 (0.40)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.38 (0.76)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e-0.21 (-0.39 to -0.04)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e72 (84.71%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e63 (74.12%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e0.5 (0.20 to 1.17)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003eFTY to GA\u003c/p\u003e \u003cp\u003e18 (21.18%)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003c/div\u003e\n\u003ch3\u003eSample size\u003c/h3\u003e\n\u003cp\u003eFor both outcomes, the available sample size (\u003cem\u003eN\u003c/em\u003e\u0026thinsp;=\u0026thinsp;3768) exceeded the minimum required for reliable intercept and slope estimation. For the 3m-CDP outcome, assuming an 8% prevalence and a C-index of 0.777, the required sample size was 222 corresponding to 18 events and 8.88 events per predictor parameter. For the relapse outcome, assuming a 16% prevalence and a C-index of 0.639, the required sample size was 544, with 88 events and 43.52 events per predictor.\u003c/p\u003e \u003cdiv id=\"Sec11\" class=\"Section2\"\u003e \u003ch2\u003eRobustness of model ITRs\u003c/h2\u003e \u003cp\u003eAcross all subgroups, the median of individual patient entropies for the recommended treatment triplet was 0.54 for the relapse model (interquartile range: 0.39\u0026ndash;0.61; Supplementary Fig. S5). Boxplots of individual entropy values reveal substantial variability: some patients had high entropy, indicating very heterogeneous results, and some patients had entropy of 0, indicating a highly consistent recommendation of the treatment triplet (NA, DMF, FTY). For the 3m-CDP model, entropy was generally higher, with a median of 0.95 across subgroups (interquartile range: 0.88\u0026ndash;1.01; Supplementary Fig. S6).\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec12\" class=\"Section2\"\u003e \u003ch2\u003eRecommended therapy switches and outcome agreement\u003c/h2\u003e \u003cp\u003eSubgroup-specific best-ranked therapies are presented in Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003e, and Supplementary Fig. S7-S9. Across subgroups, NA consistently ranked first for the relapse outcome. Recommending NA may suit patients with reason for therapy switching being TxNaive, TxHol, NoEff and Intolerance. However, it can be less suitable for all NoEff patients (for example patients with neutralizing antibodies against NA), ProgStop (especially patients needing a DMT with lower PML risk), Convenience (patients preferring oral therapies), PregDes (patients with stable disease requiring therapy de-escalation) and SAE (patients with risk of malignancies). For 3m-CDP, the first choices were TERI and IF. The agreement between relapse and 3m-CDP outcomes regarding the first-choice therapies was very low (\u003cem\u003ekappa\u003c/em\u003e\u0026thinsp;=\u0026thinsp;0, Supplementary Table S4).\u003c/p\u003e \u003cp\u003eFor the relapse outcome, the most frequently recommended treatment triplets were (NA, FTY, DMF) and (NA, DMF, FTY) (Supplementary Fig. S10). Conversely, 3m-CDP outcome recommendations were more variable, with common triplets including (IF, TERI, DMF), (IF, DMF, TERI), (TERI, DMF, IF), (TERI, DMF, NA), (TERI, GA, DMF) and (TERI, DMF, GA) (Supplementary Fig. S11).\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec13\" class=\"Section2\"\u003e \u003ch2\u003eCompliance of algorithmic output with guidelines\u003c/h2\u003e \u003cp\u003eGuidelines provide general and less specific recommendations (Supplementary Table \u003cspan refid=\"MOESM1\" class=\"InternalRef\"\u003eS1\u003c/span\u003e) making direct comparisons with our Bayesian models recommendations difficult for most switching subgroups. The relapse model recommendations comply with guidelines for NoEff and SAE, but not for PregDes. Specifically, while guidelines recommend stopping DMT or de-escalating to IF or GA for women planning pregnancy, or using NA only if disease activity persists, our model primarily recommended the triplet (NA, FTY, DMF) for PregDes patients. The 3m-CDP model\u0026rsquo;s recommendations partially aligned with guidelines, which focus on reducing relapse or magnetic resonance imaging (MRI) disease activity.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec14\" class=\"Section2\"\u003e \u003ch2\u003eModel performance\u003c/h2\u003e \u003cp\u003eThe recalibrated relapse model showed improved calibration, a C-index of 0.638 (95% CI: 0.615 to 0.661), and an MSE reduction from 0.326 to 0.248 showing improved global fit compared to the original model. The recalibrated 3m-CDP model exhibited good calibration, a C-index of 0.776 (95% CI: 0.753 to 0.797), and an MSE of 0.072 (Table\u0026nbsp;\u003cspan refid=\"Tab4\" class=\"InternalRef\"\u003e4\u003c/span\u003e, Supplementary Table S6, Supplementary Fig. S12-S15).\u003c/p\u003e \u003cp\u003eTable\u0026nbsp;\u003cspan refid=\"Tab4\" class=\"InternalRef\"\u003e4\u003c/span\u003e highlights that the relapse model exhibited consistent and good calibration across most subgroups, except for the PregDes, SAE, and ProgStop. It slightly overpredicted the average number of relapses in NoEff, ProgStop, and Convenience, and slightly underpredicted it in PregDes, SAE and TxNaive. The 3m-CDP model performed consistently across subgroups but overestimated the average 3m-CDP occurrence risk in ProgStop by 5%, and by 1% in NoEff, PregDes and SAE compared to the mean observed proportions. Additional technical details are provided in Supplementary Fig. S12-S16.\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab4\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 4\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003ePerformance of the recalibrated models in the overall validation set and in the reason of switch subgroups.\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"11\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c6\" colnum=\"6\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c7\" colnum=\"7\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c8\" colnum=\"8\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c9\" colnum=\"9\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c10\" colnum=\"10\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c11\" colnum=\"11\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eOutcome\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003ePerformance measure\u003c/p\u003e \u003cp\u003e(95% CI)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003eOverall\u003c/p\u003e \u003cp\u003evalidation set\u003c/p\u003e \u003cp\u003e(N\u0026thinsp;=\u0026thinsp;3768)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003eDrug\u003c/p\u003e \u003cp\u003eholiday\u003c/p\u003e \u003cp\u003e(N\u0026thinsp;=\u0026thinsp;1006)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c5\"\u003e \u003cp\u003eTherapy naive (N\u0026thinsp;=\u0026thinsp;1177)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c6\"\u003e \u003cp\u003eIntolerance (N\u0026thinsp;=\u0026thinsp;538)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c7\"\u003e \u003cp\u003eLack of efficacy (N\u0026thinsp;=\u0026thinsp;674)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c8\"\u003e \u003cp\u003ePregnancy desire\u003c/p\u003e \u003cp\u003e(N\u0026thinsp;=\u0026thinsp;85)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c9\"\u003e \u003cp\u003eSerious\u003c/p\u003e \u003cp\u003eadverse events\u003c/p\u003e \u003cp\u003e(N\u0026thinsp;=\u0026thinsp;31)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c10\"\u003e \u003cp\u003eProgrammed stop\u003c/p\u003e \u003cp\u003e(N\u0026thinsp;=\u0026thinsp;137)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c11\"\u003e \u003cp\u003ePersonal\u003c/p\u003e \u003cp\u003econvenience (N\u0026thinsp;=\u0026thinsp;126)\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\" morerows=\"6\" rowspan=\"7\"\u003e \u003cp\u003e\u003cb\u003e3m-CDP\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e\u003cb\u003eCalibration intercept\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e0\u003c/p\u003e \u003cp\u003e(-0.120 to 0.120)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e0.017\u003c/p\u003e \u003cp\u003e(-0.224 to 0.259)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e0.089\u003c/p\u003e \u003cp\u003e(-0.119 to 0.297)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e0.018\u003c/p\u003e \u003cp\u003e(-0.315 to 0.351)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e-0.082\u003c/p\u003e \u003cp\u003e(-0.350 to 0.185)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003e-0.220\u003c/p\u003e \u003cp\u003e(-1.157 to 0.716)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c9\"\u003e \u003cp\u003e-0.176\u003c/p\u003e \u003cp\u003e(-1.432 to 1.081)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c10\"\u003e \u003cp\u003e-0.547\u003c/p\u003e \u003cp\u003e(-1.190 to 0.095)\u003c/p\u003e\u003c/td\u003e \u003ctd align=\"left\" colname=\"c11\"\u003e \u003cp\u003e-0.053\u003c/p\u003e \u003cp\u003e(-0.757 to 0.650)\u003c/p\u003e\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e\u003cb\u003eCalibration slope\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e1.001\u003c/p\u003e \u003cp\u003e(0.865 to 1.137)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e0.991\u003c/p\u003e \u003cp\u003e(0.713 to 1.268)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e1.040\u003c/p\u003e \u003cp\u003e(0.801 to 1.278)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e0.989\u003c/p\u003e \u003cp\u003e(0.626 to 1.352)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e1.002\u003c/p\u003e \u003cp\u003e(0.669 to 1.335)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003e0.969\u003c/p\u003e \u003cp\u003e(-0.009 to 1.948)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c9\"\u003e \u003cp\u003e1.047\u003c/p\u003e \u003cp\u003e(-0.352 to 2.446)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c10\"\u003e \u003cp\u003e0.888\u003c/p\u003e \u003cp\u003e(0.196 to 1.581)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c11\"\u003e \u003cp\u003e1.005\u003c/p\u003e \u003cp\u003e(0.209 to 1.802)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e\u003cb\u003eC-index\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e0.776\u003c/p\u003e \u003cp\u003e(0.753 to 0.797)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e0.773\u003c/p\u003e \u003cp\u003e(0.723 to 0.816)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e0.783\u003c/p\u003e \u003cp\u003e(0.745 to 0.816)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e0.785\u003c/p\u003e \u003cp\u003e(0.714 to 0.843)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e0.746\u003c/p\u003e \u003cp\u003e(0.687 to 0.798)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003e0.787\u003c/p\u003e \u003cp\u003e(0.588 to 0.906)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c9\"\u003e \u003cp\u003e0.786\u003c/p\u003e \u003cp\u003e(0.454 to 0.942)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c10\"\u003e \u003cp\u003e0.768\u003c/p\u003e \u003cp\u003e(0.656 to 0.852)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c11\"\u003e \u003cp\u003e0.779\u003c/p\u003e \u003cp\u003e(0.523 to 0.919)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e\u003cb\u003eMSE\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e0.072\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e0.066\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e0.077\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e0.065\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e0.080\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003e0.053\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c9\"\u003e \u003cp\u003e0.079\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c10\"\u003e \u003cp\u003e0.076\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c11\"\u003e \u003cp\u003e0.059\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e\u003cb\u003eRRMSE (%)\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e97\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e97\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e96\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e96\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e97\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003e97\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c9\"\u003e \u003cp\u003e94\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c10\"\u003e \u003cp\u003e101\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c11\"\u003e \u003cp\u003e94\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e\u003cb\u003ePredicted proportion\u003c/b\u003e\u003c/p\u003e \u003cp\u003e\u003cb\u003e(range)\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e0.08\u003c/p\u003e \u003cp\u003e(0 to 0.60)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e0.08\u003c/p\u003e \u003cp\u003e(0 to 0.41)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e0.08\u003c/p\u003e \u003cp\u003e(0 to 0.52)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e0.08\u003c/p\u003e \u003cp\u003e(0 to 0.52)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e0.10\u003c/p\u003e \u003cp\u003e(0 to 0.60)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003e0.07\u003c/p\u003e \u003cp\u003e(0 to 0.32)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c9\"\u003e \u003cp\u003e0.11\u003c/p\u003e \u003cp\u003e(0 to 0.39)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c10\"\u003e \u003cp\u003e0.13\u003c/p\u003e \u003cp\u003e(0 to 0.50)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c11\"\u003e \u003cp\u003e0.07\u003c/p\u003e \u003cp\u003e(0 to 0.28)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e\u003cb\u003eObserved proportion\u003c/b\u003e\u003c/p\u003e \u003cp\u003e\u003cb\u003e(range)\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e0.08\u003c/p\u003e \u003cp\u003e(0 to 1)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e0.08\u003c/p\u003e \u003cp\u003e(0 to 1)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e0.09\u003c/p\u003e \u003cp\u003e(0 to 1)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e0.08\u003c/p\u003e \u003cp\u003e(0 to 1)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e0.09\u003c/p\u003e \u003cp\u003e(0 to 1)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003e0.06\u003c/p\u003e \u003cp\u003e(0 to 1)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c9\"\u003e \u003cp\u003e0.10\u003c/p\u003e \u003cp\u003e(0 to 1)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c10\"\u003e \u003cp\u003e0.08\u003c/p\u003e \u003cp\u003e(0 to 1)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c11\"\u003e \u003cp\u003e0.07\u003c/p\u003e \u003cp\u003e(0 to 1)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\" morerows=\"6\" rowspan=\"7\"\u003e \u003cp\u003e\u003cb\u003eNumber of\u003c/b\u003e\u003c/p\u003e \u003cp\u003e\u003cb\u003eRelapses\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e\u003cb\u003eCalibration intercept\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e-0.004\u003c/p\u003e \u003cp\u003e(-0.093 to 0.086)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e-0.007\u003c/p\u003e \u003cp\u003e(-0.180 to 0.167)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e0.119\u003c/p\u003e \u003cp\u003e(-0.038 to 0.275)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e-0.020\u003c/p\u003e \u003cp\u003e(-0.261 to 0.220)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e-0.269\u003c/p\u003e \u003cp\u003e(-0.494 to -0.044)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003e0.540\u003c/p\u003e \u003cp\u003e(0.043 to 1.036)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c9\"\u003e \u003cp\u003e0.152\u003c/p\u003e \u003cp\u003e(-0.773 to 1.077)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c10\"\u003e \u003cp\u003e-0.237\u003c/p\u003e \u003cp\u003e(-0.71 to 0.237)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c11\"\u003e \u003cp\u003e-0.202\u003c/p\u003e \u003cp\u003e(-0.721 to 0.317)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e\u003cb\u003eCalibration slope\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e0.973\u003c/p\u003e \u003cp\u003e(0.795 to 1.151)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e0.897\u003c/p\u003e \u003cp\u003e(0.564 to 1.229)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e0.871\u003c/p\u003e \u003cp\u003e(0.564 to 1.178)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e1.361\u003c/p\u003e \u003cp\u003e(0.833 to 1.889)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e1.082\u003c/p\u003e \u003cp\u003e(0.592 to 1.572)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003e2.186\u003c/p\u003e \u003cp\u003e(0.818 to 3.553)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c9\"\u003e \u003cp\u003e0.621\u003c/p\u003e \u003cp\u003e(-0.668 to 1.909)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c10\"\u003e \u003cp\u003e0.274\u003c/p\u003e \u003cp\u003e(-0.500 to 1.049)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c11\"\u003e \u003cp\u003e1.052\u003c/p\u003e \u003cp\u003e(0.001 to 2.103)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e\u003cb\u003eC-index\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e0.638\u003c/p\u003e \u003cp\u003e(0.615 to 0.661)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e0.632\u003c/p\u003e \u003cp\u003e(0.587 to 0.676)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e0.622\u003c/p\u003e \u003cp\u003e(0.580 to 0.661)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e0.688\u003c/p\u003e \u003cp\u003e(0.634 to 0.738)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e0.639\u003c/p\u003e \u003cp\u003e(0.580 to 0.693)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003e0.755\u003c/p\u003e \u003cp\u003e(0.623 to 0.852)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c9\"\u003e \u003cp\u003e0.617\u003c/p\u003e \u003cp\u003e(0.338 to 0.835)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c10\"\u003e \u003cp\u003e0.533\u003c/p\u003e \u003cp\u003e(0.401 to 0.659)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c11\"\u003e \u003cp\u003e0.637\u003c/p\u003e \u003cp\u003e(0.493 to 0.759)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e\u003cb\u003eMSE\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e0.248\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e0.243\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e0.258\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e0.223\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e0.226\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003e0.543\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c9\"\u003e \u003cp\u003e0.538\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c10\"\u003e \u003cp\u003e0.255\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c11\"\u003e \u003cp\u003e0.192\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e\u003cb\u003eRRMSE (%)\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e98\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e98\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e98\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e98\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e99\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003e98\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c9\"\u003e \u003cp\u003e98\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c10\"\u003e \u003cp\u003e102\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c11\"\u003e \u003cp\u003e98\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e\u003cb\u003eMean\u003c/b\u003e\u003c/p\u003e \u003cp\u003e\u003cb\u003epredicted\u003c/b\u003e\u003c/p\u003e \u003cp\u003e\u003cb\u003e(range)\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e0.20\u003c/p\u003e \u003cp\u003e(0.01 to 0.95)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e0.20\u003c/p\u003e \u003cp\u003e(0.01 to 0.66)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e0.19\u003c/p\u003e \u003cp\u003e(0.01 to 0.56)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e0.19\u003c/p\u003e \u003cp\u003e(0.01 to 0.67)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e0.21\u003c/p\u003e \u003cp\u003e(0.01 to 0.88)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003e0.22\u003c/p\u003e \u003cp\u003e(0.03 to 0.53)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c9\"\u003e \u003cp\u003e0.23\u003c/p\u003e \u003cp\u003e(0.02 to 0.77)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c10\"\u003e \u003cp\u003e0.25\u003c/p\u003e \u003cp\u003e(0.02 to 0.63)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c11\"\u003e \u003cp\u003e0.20\u003c/p\u003e \u003cp\u003e(0.02 to 0.67)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e\u003cb\u003eMean\u003c/b\u003e\u003c/p\u003e \u003cp\u003e\u003cb\u003eobserved\u003c/b\u003e\u003c/p\u003e \u003cp\u003e\u003cb\u003e(range)\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e0.20\u003c/p\u003e \u003cp\u003e(0 to 4)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e0.20\u003c/p\u003e \u003cp\u003e(0 to 4)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e0.21\u003c/p\u003e \u003cp\u003e(0 to 4)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e0.18\u003c/p\u003e \u003cp\u003e(0 to 3)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e0.17\u003c/p\u003e \u003cp\u003e(0 to 3)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003e0.38\u003c/p\u003e \u003cp\u003e(0 to 4)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c9\"\u003e \u003cp\u003e0.32\u003c/p\u003e \u003cp\u003e(0 to 3)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c10\"\u003e \u003cp\u003e0.19\u003c/p\u003e \u003cp\u003e(0 to 3)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c11\"\u003e \u003cp\u003e0.16\u003c/p\u003e \u003cp\u003e(0 to 3)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003c/div\u003e"},{"header":"Discussion","content":"\u003cp\u003eThis study validated previously developed Bayesian prognostic models [6] across eight patient subgroups defined by their reason for therapy switch. The models predicted relapse and 3-month confirmed disability progression (3m-CDP) under six treatments over a specific time horizon. Although switching reasons were not explicitly modelled, we examined whether such factors were indirectly captured through patient characteristics during model development. Because switching is often driven by lack of efficacy, intolerance, safety concerns, or personal preferences, exploring how model behaviour reflects real-world decision-making is clinically relevant. While neurologists likely followed guidelines and general principles (Supplementary Table \u003cspan refid=\"MOESM1\" class=\"InternalRef\"\u003eS1\u003c/span\u003e), it remains uncertain whether one general model is sufficient for clinical use or if subgroup-specific models are required.\u003c/p\u003e \u003cp\u003eWe examined four key aspects related to shared decision-making (SDM) in PwRRMS following a therapy switch: (1) the robustness of individualized treatment recommendations (ITRs), (2) their alignment with switching reasons and guidelines, (3) the comparability of prognostic performance across subgroups, and (4) the agreement between relapse and progression-specific ITRs.\u003c/p\u003e \u003cp\u003eOur analysis included 3768 adult PwRRMS treated between January 2017 and December 2021 in the French OFSEP registry. Eligible patients had an EDSS below or equal to 6 and had received therapy at least 6 months after disease onset. Patients were treated with IF, GA, TERI, DMF, FTY, or NA with roughly one-third being treatment-naive, one-third resuming therapy after its discontinuation, and the remainder switching therapies from platform DMTs\u0026mdash; primarily due to lack of efficacy or intolerance (Supplementary Fig. S3). Tallantyre \u003cem\u003eet al.\u003c/em\u003e [39] found similar switching patterns in their study on real-world DMT persistence and discontinuation reasons, supporting the representativeness of our cohort for current PwRRMS management.\u003c/p\u003e \u003cp\u003eIn our study, 93% of therapy-na\u0026iuml;ve patients initiated a low- or moderate-efficacy DMT. This aligns with nationwide analyses from Germany (2017\u0026ndash;2022 and 2020\u0026ndash;2022) showing that 85\u0026ndash;86% of adults with MS started therapy with low- or moderate-efficacy agents at this time [37,38]. Both studies noted a gradual shift from the traditional escalation approach\u0026mdash;initiating lower-efficacy treatments and escalating upon breakthrough disease activity\u0026mdash;toward the early high-efficacy treatment (EHT) strategy. However, this transition remains incomplete and is not reflected in the analyzed data set. In particular, increased use of high-efficacy therapy was mainly observed in PPMS, which could be driven by the 2018 approval of ocrelizumab [40], while use in RRMS remained stable [38]. In addition, because MS predominantly affects women of childbearing age and at this time only a few DMTs could be safely used during pregnancy or lactation, lower-efficacy yet safer agents remained an important therapeutic option for managing this patient subgroup [27,41,42].\u003c/p\u003e \u003cp\u003eThese observations indicate that, despite emerging evidence supporting EHT, the escalation approach still dominated clinical practice during our study period (2017\u0026ndash;2021). The limited representation of recently approved high-efficacy DMTs\u0026mdash;such as anti-CD20 antibodies (ocrelizumab, ofatumumab, ublituximab or rituximab (off-label)), immune-reconstitution therapies (cladribine, alemtuzumab), and newer S1P modulators (ozanimod, ponesimod, siponimod)\u0026mdash;reflects their gradual introduction into the market and adoption. Similar low uptake was reported by Stratil \u003cem\u003eet al.\u003c/em\u003e [38]. Likewise, most of these agents were not yet included among the therapies listed in the French Multiple Sclerosis Society (SFSEP) consensus guidelines on DMT switching [43], further confirming that our data are representative of real-world MS management at that time and the data is still informative for this particular group of drugs.\u003c/p\u003e \u003cp\u003eTo illustrate clinical applicability, Table\u0026nbsp;\u003cspan refid=\"Tab5\" class=\"InternalRef\"\u003e5\u003c/span\u003e revisits the three patient scenarios introduced in Table\u0026nbsp;\u003cspan refid=\"Tab1\" class=\"InternalRef\"\u003e1\u003c/span\u003e, presenting model-based treatment rankings and associated certainty (entropy) for each outcome, alongside brief clinical interpretation. Decision-making is straightforward in some patients but complex in others [22], and this heterogeneity is reflected in the OFSEP data and, consequently, in the algorithm built from it. Traditional models often overlook such uncertainty by providing a single \u0026ldquo;best\u0026rdquo; answer per patient, whereas our Bayesian framework quantifies probabilistic uncertainty that more closely resembles clinical reasoning. Combined with decision curve analysis based on the Chalkou et al. [44] method, this framework demonstrates how interpretable, uncertainty-aware models can inform treatment recommendations. Our work complements the Big Multiple Sclerosis Data Network initiative, which aims to improve methodological standards and real-world data analytics in MS research [45].\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab5\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 5\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eModel-based individualized treatment recommendations in the three therapy-switch cases.\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"5\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eScenario\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eClinical Outcome\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003eTop model-recommended therapies (triplet)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003eEntropy\u003c/p\u003e \u003cp\u003e(certainty)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c5\"\u003e \u003cp\u003eClinical interpretation\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eLack of efficacy\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eAt least one relapse\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e(NA, FTY, DMF): 83%\u003c/p\u003e \u003cp\u003e(NA, FTY, TERI): 17%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e0.456\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003ePatient had a breakthrough relapse while on TERI. Escalation to higher-efficacy therapy in this case is guideline-consistent. Switch to NA or FTY is clinically plausible; \u003c/p\u003e \u003cp\u003eConsistent outputs (2 triplets) \u0026rarr; higher certainty\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e3m-CDP\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e(DMF, IFN, TERI):27%\u003c/p\u003e \u003cp\u003e(GA, IFN, TERI): 21%\u003c/p\u003e \u003cp\u003e(GA, TERI, DMF): 10%\u003c/p\u003e \u003cp\u003e+\u0026thinsp;7 other triplets: 42%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e2.169\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003eModel recommends a horizontal switch to mild- moderate efficacy therapies. It is difficult to evaluate as no guideline-defined strategy for progression-based switching and limited evidence on DMTs efficacy for preventing disability progression; \u003c/p\u003e \u003cp\u003eInconsistent outputs (10 triplets) \u0026rarr; lower certainty\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003ePregnancy desire\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eAt least one relapse\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e(NA, DMF, FTY): 98%\u003c/p\u003e \u003cp\u003e(NA, FTY, DMF): 2%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e0.098\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003eNA can be given in early pregnancy if high disease activity but with risks. The other model recommendations are not advised in pregnancy \u0026rarr; requires clinical judgment given safety considerations; \u003c/p\u003e \u003cp\u003eConsistent outputs \u0026rarr; higher certainty\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e3m-CDP\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e(TERI, DMF, IFN): 94%\u003c/p\u003e \u003cp\u003e(TERI, IFN, DMF): 6%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e0.227\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003eIFN recommended by model is a safe option in case of disease activity and aligns with guidelines, but TERI and DMF are contraindicated/not recommended in pregnancy \u0026rarr; requires clinical judgment given safety considerations; \u003c/p\u003e \u003cp\u003eConsistent outputs \u0026rarr; higher certainty\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003ePersonal convenience\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eAt least one relapse\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e(DMF, NA, FTY): 66%\u003c/p\u003e \u003cp\u003e(FTY, NA, DMF): 31%\u003c/p\u003e \u003cp\u003e(NA, DMF, FTY): 2%\u003c/p\u003e \u003cp\u003e(NA, FTY, DMF): 1%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e0.762\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003eThe recommended options DMF and FTY are oral therapies and align with patient preference of switching from injectable IFN, which seems clinically reasonable. No official guideline recommendation so comparison not possible; \u003c/p\u003e \u003cp\u003eLess consistent outputs \u0026rarr; moderate certainty\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e3m-CDP\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e(IFN, DMF, TERI): 65%\u003c/p\u003e \u003cp\u003e(IFN, TERI, DMF): 24%\u003c/p\u003e \u003cp\u003e(IFN, TERI, GA): 5%\u003c/p\u003e \u003cp\u003e(TERI, IFN, DMF): 5%\u003c/p\u003e \u003cp\u003e(TERI, IFN, GA): 1%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e0.968\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003ePatient can stay on IFN or switch to DMF or TERI. Options are clinically acceptable given patient older age and stable disease, but partially divergent from relapse model; \u003c/p\u003e \u003cp\u003eLess consistent outputs \u0026rarr; moderate certainty\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003cp\u003eCurrent guidelines likewise address uncertainty only indirectly by offering general rather than specific recommendations [22,25,26], and an update might be necessary in light of emerging evidence. For example, they advise monitoring and reevaluating patients after therapy discontinuation, but specify no strategies for re-initiation or switching. While alemtuzumab was listed as potential initial therapy by the 2018 AAN guidelines for high risk patients, the European Medicines Agency has restricted its use in 2019 as first-line therapy due to safety concerns [46], leading to its limited use in practice, as also shown by Papukchieva \u003cem\u003eet al.\u003c/em\u003e [37].\u003c/p\u003e \u003cp\u003eWe show, whenever comparison with guidelines was possible, that model recommendations did not always align with guidelines. While natalizumab consistently ranked highest for relapse prevention\u0026mdash;supporting both model validity and aligning with clinical expectations\u0026mdash;recommendations for the pregnancy desire subgroup included contraindicated options (FTY, TERI), highlighting the importance of contextual interpretation and clinician oversight. These discrepancies raise broader questions about how divergent model outputs and guideline recommendations should be handled and integrated in the SDM process between patient and neurologist. Uncertainty is an intrinsic part of MS management, and probabilistic predictions can make it more transparent to both physicians and patients. However, clinicians must still interpret model outputs within the clinical, safety, and personal context of each patient. SDM can be strengthened through patient-preparation tools\u0026mdash;such as MS-SUPPORT [47]\u0026mdash;that help patients clarify their treatment goals, become more informed about their disease and therapy options, and improve their readiness for collaborative decision-making. This in turn can help improve adherence to therapy and disease outcomes.\u003c/p\u003e \u003cp\u003eSome miscalibration of the model occurred in smaller subgroups, which could lead to over- or under-treatment. Because we did not oversample low-prevalence subgroups, further research should determine whether one general model\u0026mdash;which does not explicitly account for switching reasons\u0026mdash;is clinically satisfactory, or whether subgroup-specific models are necessary to improve accuracy.\u003c/p\u003e \u003cp\u003eWe also observed inconsistent treatment recommendations between both outcome models. While the 3m-CDP model recommended diverse treatment triplets for individual patients, the relapse model consistently suggested only two quite similar triplets. This divergence complicates prioritization of treatment goals and the practical implementation of model recommendations for therapy switches. As mentioned earlier, the models mirror the uncertainties encountered in clinical practice, where optimal strategies for individual patients remain unclear due to limited comparative evidence [48,49]. This challenge is also reflected in guideline recommendations [26], which are largely consensus-based, and in Gonz\u0026aacute;lez-Lorenzo \u003cem\u003eet al.\u003c/em\u003e [50] findings, whose network meta-analysis revealed uncertainties in DMT efficacy for 3m-CDP but produced rankings similar to our relapse model. Because a network meta-analysis is population-based and does not provide individualized effect rankings, we cautiously compared our results to theirs.\u003c/p\u003e \u003cp\u003eSeveral limitations must be acknowledged. Key variables influencing therapy choice\u0026mdash;comorbidities, concomitant medications, John Cunningham virus status, MRI findings, and adherence\u0026mdash;were unavailable and may have affected model outputs. Furthermore, AAN guidelines recommend high efficacy therapies for highly active disease or switching within the same efficacy category for patients already on such therapies. The limited availability of therapeutic options in 2011\u0026ndash;2021 is no longer comparable with the current therapeutic landscape. This explains the absence of newer agents from the analyses. However, this is consistent with findings from the previously cited French studies [36,43]. Updating the model with contemporary OFSEP data incorporating these therapies will be important for future validation.\u003c/p\u003e \u003cp\u003eStrengths include the use of representative, prospectively collected, and fit-for-purpose registry data with low missingness for switching reasons, a large sample size across treatment groups, and the inclusion of diverse predefined subgroups with varying risk profiles and treatment responses (Tables\u0026nbsp;\u003cspan refid=\"Tab2\" class=\"InternalRef\"\u003e2\u003c/span\u003e and \u003cspan refid=\"Tab3\" class=\"InternalRef\"\u003e3\u003c/span\u003e), better reflecting the real-world PwRRMS target population. These features enhance generalizability compared with randomized trials, which\u0026mdash;owing to strict eligibility criteria\u0026mdash;often lack external validity. This underscores the value of registry-based prognostic modelling and the need for better research, particularly in heterogeneous patient subgroups, as emphasized by the European Medicines Agency\u0026rsquo;s guidelines [51]. In addition to supporting individualized decisions, prognostic models should be fair to avoid discrimination against individuals or groups [52]. Because MS disproportionately affects younger adults and women, including sex and age as predictors in our models helps account for key demographic differences and supports balanced recommendations.\u003c/p\u003e \u003cp\u003eWhile the models demonstrated robust ITRs, some misalignment with guidelines persisted, highlighting the inherent complexity of treatment decisions even when supported by guidelines and algorithms. Physician expertise in balancing model insights with clinical guidelines and patient preferences remains key [53]. In addition, emerging evidence [15] increasingly supports viewing multiple sclerosis as a biological continuum rather than a collection of discrete subtypes. This perspective emphasizes that inflammatory and neurodegenerative mechanisms coexist throughout the disease course and may explain discrepancies between relapse- and progression-based predictions. In particular, progression independent of relapse activity (PIRA) occurs across all MS phenotypes and is associated with poor prognosis yet remains difficult to detect when defined solely by conventional endpoints such as EDSS. Incorporating patient data with different MS subtypes, patient-reported outcomes (e.g., fatigue, cognition, mental health) and functional measures into prognostic modelling may improve early identification of subtle progression and align model predictions with a more continuous and patient-centred understanding of MS.\u003c/p\u003e \u003cp\u003eFinally, implementing model-based treatment decision support in clinical practice requires further validation and consideration, as clear evidence is currently lacking. To assess the utility of such models, a cluster-randomized trial would be ideal. Ultimately, such tools must operate in\u0026mdash;and aim to improve\u0026mdash;imperfect clinical care settings [54].\u003c/p\u003e"},{"header":"Materials and methods","content":"\u003cdiv id=\"Sec17\" class=\"Section2\"\u003e \u003ch2\u003eData source and Participants\u003c/h2\u003e \u003cp\u003eThis project was approved by the Ethics Committee of Ludwig-Maximilians-University Munich (No. 21-1174) and the OFSEP steering board. We used longitudinal data from 3768 patients in the French multiple sclerosis registry OFSEP [55], treated between January 1, 2017 and December 15, 2021, which marked the end of follow-up. OFSEP provides high-quality, standardized data collected prospectively during regular visits across 38 French centres [56]. OFSEP physicians obtained written consent from all participants. The study was announced on the OFSEP website (\u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://www.ofsep.org/fr/etudes/extval-phrend\u003c/span\u003e\u003cspan address=\"https://www.ofsep.org/fr/etudes/extval-phrend\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e), allowing patients to provide dynamic consent.\u003c/p\u003e \u003cp\u003eEligible PwRRMS had a minimum disease duration of 6 months, an EDSS score below or equal to 6, and were treated with IF, GA, TERI, DMF, FTY, or NA, and either initiated or switched therapy. We considered patients as \u003cem\u003etherapy-naive\u003c/em\u003e if they had not received any therapy for at least 6 months following disease onset and on a \u003cem\u003edrug holiday\u003c/em\u003e if they had discontinued one of the studied DMTs for at least 3 months prior to re-initiating therapy. This threshold allows distinguishing clinically meaningful treatment gaps from standard washout periods required for therapy switches (see Supplementary Table S2).\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec18\" class=\"Section2\"\u003e \u003ch2\u003eVariables\u003c/h2\u003e \u003cp\u003ePredictors were assessed before or at index therapy initiation (baseline) and included: previous therapy \u003cem\u003e(current therapy)\u003c/em\u003e, and its duration (\u003cem\u003ecurrent duration\u003c/em\u003e), whether this therapy or the one prior to it was second-line (\u003cem\u003esecond-line\u003c/em\u003e), number of previous DMTs \u003cem\u003e(DMT count)\u003c/em\u003e, number of relapses in the previous year \u003cem\u003e(relapse count)\u003c/em\u003e, time since last relapse \u003cem\u003e(relapse distance)\u003c/em\u003e, duration since disease onset (\u003cem\u003eonset distance\u003c/em\u003e), \u003cem\u003eage\u003c/em\u003e at therapy start, baseline Expanded Disability Status Scale \u003cem\u003e(EDSS)\u003c/em\u003e, \u003cem\u003eindex therapy\u003c/em\u003e, and its \u003cem\u003eduration.\u003c/em\u003e We used \u003cem\u003eindex duration\u003c/em\u003e as an offset term to account for therapy duration variability.\u003c/p\u003e \u003cp\u003eOutcomes included the number of relapses and 3m-CDP occurrence during index therapy. 3m-CDP was defined as a sustained EDSS increase from baseline (\u0026ge;\u0026thinsp;1 point, or \u0026ge;\u0026thinsp;0.5 if baseline EDSS\u0026thinsp;\u0026gt;\u0026thinsp;5.5) with no improvement over 3 months, confirmed by a follow-up EDSS measured\u0026thinsp;\u0026ge;\u0026thinsp;3 months after the increase and within 12 months after therapy end, and occurring at least 3 months after any relapse. Data processing steps, predictor selection, and variable definitions are detailed in Sakr \u003cem\u003eet al.\u003c/em\u003e [6].\u003c/p\u003e \u003cp\u003eEight patient subgroups were defined based on the reason for therapy switching: lack of efficacy, intolerance, programmed stop, personal convenience, pregnancy desire, serious adverse events, ending therapy naivety or ending drug holiday. We performed our analyses in R, version 4.2.2. We also reported this work in accordance with the TRIPOD\u0026thinsp;+\u0026thinsp;AI statement [52] in the Supplementary Fig. S17 to ensure transparency and reproducibility.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec19\" class=\"Section2\"\u003e \u003ch2\u003eSample size\u003c/h2\u003e \u003cp\u003eAlthough our previously developed models generally performed well, their predictions did not consistently align with observed outcome risks in some patient groups. Therefore, we updated the models\u0026rsquo; intercept and slope using the validation set. To estimate a sufficient sample size for this update, we applied the method proposed by Riley \u003cem\u003eet al.\u003c/em\u003e [57], which determines the minimum sample size needed for binary outcome prediction models to reduce overfitting and precisely estimate the average outcome risk. This method requires specifying the expected outcome proportion in the new dataset, the anticipated model performance (e.g., Cox-Snell R\u0026sup2; or C-index), the number of model parameters, and a targeted shrinkage factor\u0026mdash;here set to 10%. For our calculations, we used the original models\u0026rsquo; C-index and the outcome prevalence in the validation set, defined as the proportion of patients experiencing at least one relapse or disease progression.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec20\" class=\"Section2\"\u003e \u003ch2\u003eStatistical analysis methods\u003c/h2\u003e \u003cdiv id=\"Sec21\" class=\"Section3\"\u003e \u003ch2\u003eRobustness of ITRs\u003c/h2\u003e \u003cp\u003eThe Bayesian machine learning algorithm generates a vector of six relapse or 3m-CDP probabilities under the six DMTs, ranked by the value of a favourable prognosis. To assess recommendation certainty, we drew 100 samples per patient from the risk posterior distribution, counting how often each DMT ranked, first, second, third, etc. To simplify, we focused on the top three DMTs in each sample\u0026mdash;the ones with best prognoses\u0026mdash;and refer to these as \u003cem\u003etreatment triplet\u003c/em\u003e. We then assessed consistency of the recommended treatment triplet using entropy. With six DMTs, there are 120 (6 x 5 x 4) possible triplet rankings. Maximum entropy (-ln(1/120)\u0026thinsp;=\u0026thinsp;ln(120)\u0026thinsp;=\u0026thinsp;4.78) indicates equal probability of the 120 possible results and thus complete uncertainty, while entropy of 0 reflects identical algorithmic recommendations across samples. Values near 0 indicate robustness, while we considered values above 2 as heterogeneous or uncertain.\u003c/p\u003e \u003cp\u003e \u003cb\u003eCompliance of recommendations with switching reasons and guidelines.\u003c/b\u003e \u003c/p\u003e \u003cp\u003eWe summarized relevant guidelines\u0026rsquo; recommendations (Supplementary Table \u003cspan refid=\"MOESM1\" class=\"InternalRef\"\u003eS1\u003c/span\u003e) and descriptively assessed whether subgroup-specific top-ranked ITRs align with both clinical guidelines and reason for therapy switching.\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv id=\"Sec22\" class=\"Section2\"\u003e \u003ch2\u003ePerformance assessment\u003c/h2\u003e \u003cp\u003eWe evaluated model performance in the overall validation set and predefined subgroups following established methods for clinical prediction models [58\u0026ndash;60]. Performance assessment included calibration, discrimination and global fit. Calibration, assessed via calibration curves, intercept and slope, reflects the agreement between observed outcome proportion and predicted outcome probability [61,62]. Discrimination was evaluated by the C-index (similar to the area under the curve AUC). The mean squared error (MSE) and relative root MSE (RRMSE) quantified the model\u0026rsquo;s fit absolutely or relative to the outcome variance [63]. The RRMSE enables cross-setting comparisons.\u003c/p\u003e \u003cdiv id=\"Sec23\" class=\"Section3\"\u003e \u003ch2\u003ePresentation of model predictions\u003c/h2\u003e \u003cp\u003eWe developed a web-based Shiny application (available at \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://shiny.ibe.med.uni-muenchen.de/switch-ms\u003c/span\u003e\u003cspan address=\"https://shiny.ibe.med.uni-muenchen.de/switch-ms\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e; permalink: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://switchms.annamariasakr.de\u003c/span\u003e\u003cspan address=\"https://switchms.annamariasakr.de\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e accessible with the following credentials: user: msswitch, password: Roox3iey) (Supplementary Fig. \u003cspan refid=\"MOESM1\" class=\"InternalRef\"\u003eS1\u003c/span\u003e). To use the calculator please push on the tab \u0026ldquo;Outcome predictions and Treatment recommendations\u0026rdquo;, than users can input the routinely collected patient data and generate predicted probabilities for relapse and 3m-CDP under the six DMTs. The app provides two therapy rankings based on: (1) best prognosis and (2) benefit outweighing harms (Decision Curves Analysis, Chalkou \u003cem\u003eet al\u003c/em\u003e. [44]).\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv id=\"Sec24\" class=\"Section2\"\u003e \u003ch2\u003eAgreement between 3m-CDP and Relapse models\u0026rsquo; recommendations\u003c/h2\u003e \u003cp\u003eWe used Cohen\u0026rsquo;s kappa to measure the agreement between ITRs for relapse and 3m-CDP. First-ranked therapies under the best prognosis approach were categorized based on the German Neurological Society (DGN) classification [28]: (1) platform DMTs: IFN, GA, TERI, DMF, (2) moderate efficacy FTY and (3) high efficacy NA.\u003c/p\u003e \u003c/div\u003e"},{"header":"Abbreviations","content":"\u003cdiv class=\"DefinitionList\"\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eAAN\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eAmerican Academy of Neurology\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eAUC\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eArea under the curve\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003e3m-CDP\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eConfirmed disease progression at 3 months\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eCI\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eConfidence interval\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eConvenience\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003ePersonal convenience\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eDGN\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eDeutsche Gesellschaft f\u0026uuml;r Neurologie, German neurological society\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eDMF\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eDimethyl fumarate\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eDMT\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eDisease-modifying therapy\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eECTRIMS/EAN\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eEuropean Committee of Treatment of Research in Multiple Sclerosis/ European Academy of Neurology\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eEDSS\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eExpanded disability status scale\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eEHT\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eEarly high-efficacy treatment\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eFTY\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eFingolimod\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eGA\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eGlatiramer acetate\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eIF\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eInterferon beta 1\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eITRs\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eIndividual treatment recommendations\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eMRI\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eMagnetic resonance imaging\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eMS\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eMultiple sclerosis\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eMSE\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eMean squared error\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eNA\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eNatalizumab\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eNoEff\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eLack of efficacy\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eOFSEP\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eObservatoire Fran\u0026ccedil;ais de la Scl\u0026eacute;rose en Plaques\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eOR\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eOdds ratio\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003ePIRA\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eProgression independent of relapse activity\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003ePML\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eProgressive multifocal leukoencephalopathy\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003ePwMS\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003ePersons with multiple sclerosis\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003ePwRRMS\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003ePersons with relapsing-remitting multiple sclerosis\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003ePPMS\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003ePrimary progressive multiple sclerosis\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003ePregDes\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003ePregnancy\u0026nbsp;desire\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eProgStop\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eProgrammed stop\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eRRMSE\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eRelative root mean squared error\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eRRMS\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eRelapsing-remitting multiple sclerosis\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eSAE\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eSerious adverse events\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eSDM\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eShared Decision-Making\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eSFSEP\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eFrench Multiple Sclerosis Society\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eTERI\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eTeriflunomide\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eTRIPOD\u0026thinsp;+\u0026thinsp;AI\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eTransparent Reporting of a multivariable prediction model for individual Prognosis or Diagnosis\u0026thinsp;+\u0026thinsp;Artificial Intelligence\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eTxHol\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eDrug holiday\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eTxNaive\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eTherapy naive\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003c/div\u003e"},{"header":"Declarations","content":"\u003cp\u003eAcknowledgements\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eOFSEP Data collection has been supported by a grant provided by the French State and handled by the “Agence Nationale de la Recherche,\" within the framework of the \"France 2030\" programme, under the reference ANR-10-COHO-002, Observatoire Français de la Sclérose en Plaques (OFSEP)/Eugène Devic EDMUS Foundation against multiple sclerosis. We also thank Nikolaus von Bomhard and Marcel Müller for their technical support in providing a secure server for publishing the Shiny app. ChatGPT was used to polish the text and R code.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eAuthors’ contributions\u0026nbsp;\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAnna-Maria Sakr\u003c/strong\u003e: Data curation (lead), formal analysis (lead), methodology (equal), software (lead), visualization (lead), Conceptualization (equal), writing - original draft (equal), writing - review and editing (equal), \u003cstrong\u003eUlrich Mansmann\u003c/strong\u003e: Conceptualization (equal), funding acquisition (lead), methodology (equal), formal analysis (supporting), supervision (lead), visualization (supporting), writing- original draft (equal), writing - review and editing (equal), \u003cstrong\u003eJoachim Havla\u003c/strong\u003e: Conceptualization (supporting), writing - review and editing (supporting). \u0026nbsp;\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eAdditional Information\u003c/p\u003e\n\u003cp\u003eData availability\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eThe data supporting the results of this study cannot be made publicly available due to legal constraints. The data were obtained from the Observatoire Français de la Sclérose en Plaques (OFSEP), which served as our primary data source. Our specific data request is documented at https://www.ofsep.org/fr/etudes/extval-phrend and accessible at the OSF directory https://osf.io/yf3ks/files/osfstorage. Interested researchers may obtain the corresponding OFSEP dataset directly from the data-provider by using our data request. The analyses can be replicated using the analysis scripts (freely accessible in the public repository Gitlab, at https://gitlab.lrz.de/asakr/switch-ms; permanent link in case of relocation of the repository at \u0026nbsp;switchms-code.annamariasakr.de) as well as the methodological details provided in the manuscript and supplementary materials. The authors did not have any special access privileges and the dataset can be requested by any qualified researcher through the standard application process. Access to the OFSEP data requires completion of a data use agreement with OFSEP.\u003c/p\u003e\n\u003cp\u003eThe corresponding application is accessible at: https://shiny.ibe.med.uni-muenchen.de/switch-ms (username: msswitch, password: Roox3iey). To use the calculator please push on the tab “Outcome predictions and Treatment recommendations “.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eFunding\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eAnna Maria Sakr was supported by the Data integration for Future Medicine (DIFUTURE) grant (Bundesministerium für Bildung und Forschung (BMBF) 01ZZ1804C). The funding bodies played no role in the design of the study and collection, analysis, and interpretation of data and in writing the manuscript.\u003c/p\u003e\n\u003cp\u003eCompeting interests\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eJH reports a grant for OCT research from the Friedrich‐Baur‐Stiftung, Horizon, Sanofi and Merck, personal fees and nonfinancial support from Alexion, Amgen, Bayer, Biogen, BMS, Merck, Novartis and Roche, and nonfinancial support of the Sumaira‐Foundation and Guthy‐Jackson Charitable Foundation, all outside the submitted work. AMS und UM declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eEthics approval\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eThe study was approved by the OFSEP Steering Committee (approval no. 0266) and the Ethics Committee of Ludwig-Maximilians-University Munich (approval no. 21-1174). \u0026nbsp;All procedures were conducted in accordance with applicable ethical standards, guidelines, and regulations.\u0026nbsp;The study is reported in accordance with the TRIPOD-AI statement.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eConsent to participate\u003c/p\u003e\n\u003cp\u003eAll participants in the OFSEP registry provided written informed consent through their treating physicians. The study was announced on the OFSEP website\u0026nbsp;(https://www.ofsep.org/fr/etudes/extval-phrend)\u0026nbsp;prior to initiation, allowing patients to provide project-specific dynamic consent.\u003c/p\u003e\n\u003cp\u003eConsent for publication\u003c/p\u003e\n\u003cp\u003eNot applicable\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\n\u003cli\u003eWalton, C. \u003cem\u003eet al.\u003c/em\u003e Rising prevalence of multiple sclerosis worldwide: Insights from the Atlas of MS, third edition. \u003cem\u003eMult Scler\u003c/em\u003e \u003cstrong\u003e26\u003c/strong\u003e, 1816\u0026ndash;1821 (2020).\u003c/li\u003e\n\u003cli\u003eKoch-Henriksen, N. \u0026amp; Magyari, M. Apparent changes in the epidemiology and severity of multiple sclerosis. \u003cem\u003eNat Rev Neurol\u003c/em\u003e \u003cstrong\u003e17\u003c/strong\u003e, 676\u0026ndash;688 (2021).\u003c/li\u003e\n\u003cli\u003eDunn, S. E., Gunde, E. \u0026amp; Lee, H. Sex-Based Differences in Multiple Sclerosis (MS): Part II: Rising Incidence of Multiple Sclerosis in Women and the Vulnerability of Men to Progression of this Disease. in \u003cem\u003eEmerging and Evolving Topics in Multiple Sclerosis Pathogenesis and Treatments\u003c/em\u003e (eds AC, F. \u0026amp; JM, O.) vol. 2015 57\u0026ndash;86 (Springer International Publishing).\u003c/li\u003e\n\u003cli\u003eReeve, K. \u003cem\u003eet al.\u003c/em\u003e Prognostic models for predicting clinical disease progression, worsening and activity in people with multiple sclerosis. \u003cem\u003eCochrane Database of Systematic Reviews\u003c/em\u003e \u003cstrong\u003e2023\u003c/strong\u003e, (2023).\u003c/li\u003e\n\u003cli\u003eHavla, J., Reeve, K., On, B. I., Mansmann, U. \u0026amp; Held, U. Prognostic models in multiple sclerosis: progress and challenges in clinical integration. \u003cem\u003eNeurol. Res. Pract.\u003c/em\u003e \u003cstrong\u003e6\u003c/strong\u003e, 44 (2024).\u003c/li\u003e\n\u003cli\u003eSakr, A. M., Mansmann, U., Havla, J., \u0026Ouml;n, B. I. \u0026amp; \u0026Ouml;n, B. I. Framework for personalized prediction of treatment response in relapsing-remitting multiple sclerosis: a replication study in independent data. \u003cem\u003eBMC Med Res Methodol\u003c/em\u003e \u003cstrong\u003e24\u003c/strong\u003e, 138 (2024).\u003c/li\u003e\n\u003cli\u003eSt\u0026uuml;hler, E. \u003cem\u003eet al.\u003c/em\u003e Framework for personalized prediction of treatment response in relapsing remitting multiple sclerosis. \u003cem\u003eBMC medical research methodology\u003c/em\u003e \u003cstrong\u003e20\u003c/strong\u003e, 24 (2020).\u003c/li\u003e\n\u003cli\u003eBraune, S., Stuehler, E., Heer, Y., van Hoevell, P. \u0026amp; Bergmann, A. PHREND(\u0026reg;)-A Real-World Data-Driven Tool Supporting Clinical Decisions to Optimize Treatment in Relapsing-Remitting Multiple Sclerosis. \u003cem\u003eFrontiers in digital health\u003c/em\u003e \u003cstrong\u003e4\u003c/strong\u003e, 856829 (2022).\u003c/li\u003e\n\u003cli\u003eVan Calster, B., Steyerberg, E. W., Wynants, L. \u0026amp; Van Smeden, M. There is no such thing as a validated prediction model. \u003cem\u003eBMC Med\u003c/em\u003e \u003cstrong\u003e21\u003c/strong\u003e, 70 (2023).\u003c/li\u003e\n\u003cli\u003eMcDonald, W. I. \u003cem\u003eet al.\u003c/em\u003e Recommended diagnostic criteria for multiple sclerosis: Guidelines from the International Panel on the Diagnosis of Multiple Sclerosis. \u003cem\u003eAnnals of Neurology\u003c/em\u003e \u003cstrong\u003e50\u003c/strong\u003e, 121\u0026ndash;127 (2001).\u003c/li\u003e\n\u003cli\u003ePolman, C. H. \u003cem\u003eet al.\u003c/em\u003e Diagnostic Criteria for Multiple Sclerosis: 2005 Revisions to the \u0026lsquo;McDonald Criteria\u0026rsquo;. https://doi.org/10.1002/ana.206703 (2005) doi:10.1002/ana.206703.\u003c/li\u003e\n\u003cli\u003ePolman, C. H. \u003cem\u003eet al.\u003c/em\u003e Diagnostic criteria for multiple sclerosis: 2010 Revisions to the McDonald criteria. \u003cem\u003eAnnals of Neurology\u003c/em\u003e \u003cstrong\u003e69\u003c/strong\u003e, 292\u0026ndash;302 (2011).\u003c/li\u003e\n\u003cli\u003eThompson, A. J. \u003cem\u003eet al.\u003c/em\u003e Diagnosis of multiple sclerosis: 2017 revisions of the McDonald criteria. \u003cem\u003eThe Lancet Neurology\u003c/em\u003e \u003cstrong\u003e17\u003c/strong\u003e, 162\u0026ndash;173 (2018).\u003c/li\u003e\n\u003cli\u003eMontalban, X. \u003cem\u003eet al.\u003c/em\u003e Diagnosis of multiple sclerosis: 2024 revisions of the McDonald criteria. \u003cem\u003eThe Lancet Neurology\u003c/em\u003e \u003cstrong\u003e24\u003c/strong\u003e, 850\u0026ndash;865 (2025).\u003c/li\u003e\n\u003cli\u003ePortaccio, E. \u003cem\u003eet al.\u003c/em\u003e Multiple sclerosis: emerging epidemiological trends and redefining the clinical course. \u003cem\u003eThe Lancet Regional Health - Europe\u003c/em\u003e \u003cstrong\u003e44\u003c/strong\u003e, 100977 (2024).\u003c/li\u003e\n\u003cli\u003eArredondo-Robles, A. V., Rodr\u0026iacute;guez-L\u0026oacute;pez, K. P. \u0026amp; \u0026Aacute;vila-Avil\u0026eacute;s, R. D. Clinical Management in Multiple Sclerosis. \u003cem\u003eNeuroglia\u003c/em\u003e \u003cstrong\u003e6\u003c/strong\u003e, 6 (2025).\u003c/li\u003e\n\u003cli\u003eM\u0026uuml;ller, J. \u003cem\u003eet al.\u003c/em\u003e Harmonizing Definitions for Progression Independent of Relapse Activity in Multiple Sclerosis: A Systematic Review. \u003cem\u003eJAMA Neurol\u003c/em\u003e \u003cstrong\u003e80\u003c/strong\u003e, 1232 (2023).\u003c/li\u003e\n\u003cli\u003eDebray, T. P. A. \u003cem\u003eet al.\u003c/em\u003e A guide to systematic review and meta-analysis of prediction model performance. \u003cem\u003eBMJ\u003c/em\u003e i6460 (2017) doi:10.1136/bmj.i6460.\u003c/li\u003e\n\u003cli\u003eVukusic, S. \u003cem\u003eet al.\u003c/em\u003e Observatoire Fran\u0026ccedil;ais de la Scl\u0026eacute;rose en Plaques (OFSEP): A unique multimodal nationwide MS registry in France. \u003cem\u003eMultiple Sclerosis Journal\u003c/em\u003e \u003cstrong\u003e26\u003c/strong\u003e, 118\u0026ndash;122 (2020).\u003c/li\u003e\n\u003cli\u003eMcElreath, R. \u003cem\u003eStatistical Rethinking: A Bayesian Course with Examples in R and Stan\u003c/em\u003e. (CRC Press, Boca Raton London New York, 2020).\u003c/li\u003e\n\u003cli\u003eGelman, A., Carlin, J. B. \u0026amp; Stern, H. S. Bayesian Data Analysis Third edition ( with errors fixed as of 15 February 2021 ). (2021).\u003c/li\u003e\n\u003cli\u003eCorboy, J. R., Weinshenker, B. G. \u0026amp; Wingerchuk, D. M. Comment on 2018 American Academy of Neurology guidelines on disease-modifying therapies in MS. \u003cem\u003eNeurology\u003c/em\u003e \u003cstrong\u003e90\u003c/strong\u003e, 1106\u0026ndash;1112 (2018).\u003c/li\u003e\n\u003cli\u003eKoch-Henriksen, N. \u0026amp; Magyari, M. Apparent changes in the epidemiology and severity of multiple sclerosis. \u003cem\u003eNat Rev Neurol\u003c/em\u003e \u003cstrong\u003e17\u003c/strong\u003e, 676\u0026ndash;688 (2021).\u003c/li\u003e\n\u003cli\u003eRae-Grant, A. \u003cem\u003eet al.\u003c/em\u003e Comprehensive systematic review summary: Disease-modifying therapies for adults with multiple sclerosis: Report of the Guideline Development, Dissemination, and Implementation Subcommittee of the American Academy of Neurology. \u003cem\u003eNeurology\u003c/em\u003e \u003cstrong\u003e90\u003c/strong\u003e, 789\u0026ndash;800 (2018).\u003c/li\u003e\n\u003cli\u003eRae-Grant, A. \u003cem\u003eet al.\u003c/em\u003e Practice guideline recommendations summary: Disease-modifying therapies for adults with multiple sclerosis: Report of the Guideline Development, Dissemination, and Implementation Subcommittee of the American Academy of Neurology. \u003cem\u003eNeurology\u003c/em\u003e \u003cstrong\u003e90\u003c/strong\u003e, 777\u0026ndash;788 (2018).\u003c/li\u003e\n\u003cli\u003eMontalban, X. \u003cem\u003eet al.\u003c/em\u003e ECTRIMS/EAN Guideline on the pharmacological treatment of people with multiple sclerosis. \u003cem\u003eMultiple Sclerosis\u003c/em\u003e \u003cstrong\u003e24\u003c/strong\u003e, 96\u0026ndash;120 (2018).\u003c/li\u003e\n\u003cli\u003eAndrodias, G. \u003cem\u003eet al.\u003c/em\u003e De-escalating and discontinuing disease-modifying therapies in multiple sclerosis. \u003cem\u003eBrain\u003c/em\u003e \u003cstrong\u003e148\u003c/strong\u003e, 1459\u0026ndash;1478 (2025).\u003c/li\u003e\n\u003cli\u003eHemmer, B., Gehring, K. \u0026amp; et al. Diagnose und Therapie der Multiplen Sklerose, Neuromyelitis-optica-Spektrum-Erkrankungen und MOG-IgG-assoziierten Erkrankungen, S2k-Leitlinie, 2024, in: Deutsche Gesellschaft f\u0026uuml;r Neurologie (Hrsg.), Leitlinien f\u0026uuml;r Diagnostik und Therapie in der Neurologie. Online: www.dgn.org/leitlinien. Accessed on 19 November 2025. https://www.dgn.org/leitlinie/diagnose-und-therapie-der-multiplen-sklerose-neuromyelitis-optica-spektrum-erkrankungen-und-mog-igg-assoziierten-erkrankungen (2024).\u003c/li\u003e\n\u003cli\u003eFilippi, M. \u003cem\u003eet al.\u003c/em\u003e The use of high-efficacy disease-modifying therapies in multiple sclerosis: recommendations from an expert Delphi consensus. \u003cem\u003eJ Neurol\u003c/em\u003e \u003cstrong\u003e272\u003c/strong\u003e, 565 (2025).\u003c/li\u003e\n\u003cli\u003eOntaneda, D. \u003cem\u003eet al.\u003c/em\u003e Determining the effectiveness of early intensive versus escalation approaches for the treatment of relapsing-remitting multiple sclerosis: The DELIVER-MS study protocol. \u003cem\u003eContemporary Clinical Trials\u003c/em\u003e \u003cstrong\u003e95\u003c/strong\u003e, 106009 (2020).\u003c/li\u003e\n\u003cli\u003eMowry, E. M. \u003cem\u003eet al.\u003c/em\u003e The TRaditional versus Early Aggressive Therapy for MS (TREAT-MS) trial: Design and baseline characteristics of participants. \u003cem\u003eContemporary Clinical Trials\u003c/em\u003e 108117 (2025) doi:10.1016/j.cct.2025.108117.\u003c/li\u003e\n\u003cli\u003eSelmaj, K., Cree, B. A. C., Barnett, M., Thompson, A. \u0026amp; Hartung, H.-P. Multiple sclerosis: time for early treatment with high-efficacy drugs. \u003cem\u003eJ Neurol\u003c/em\u003e \u003cstrong\u003e271\u003c/strong\u003e, 105\u0026ndash;115 (2024).\u003c/li\u003e\n\u003cli\u003eHarding, K. \u003cem\u003eet al.\u003c/em\u003e Clinical Outcomes of Escalation vs Early Intensive Disease-Modifying Therapy in Patients With Multiple Sclerosis. \u003cem\u003eJAMA Neurol\u003c/em\u003e \u003cstrong\u003e76\u003c/strong\u003e, 536 (2019).\u003c/li\u003e\n\u003cli\u003eMerkel, B., Butzkueven, H., Traboulsee, A. L., Havrdova, E. \u0026amp; Kalincik, T. Timing of high-efficacy therapy in relapsing-remitting multiple sclerosis: A systematic review. \u003cem\u003eAutoimmunity Reviews\u003c/em\u003e \u003cstrong\u003e16\u003c/strong\u003e, 658\u0026ndash;665 (2017).\u003c/li\u003e\n\u003cli\u003eSinger, B. A., Feng, J. \u0026amp; Chiong-Rivero, H. Early use of high-efficacy therapies in multiple sclerosis in the United States: benefits, barriers, and strategies for encouraging adoption. \u003cem\u003eJ Neurol\u003c/em\u003e \u003cstrong\u003e271\u003c/strong\u003e, 3116\u0026ndash;3130 (2024).\u003c/li\u003e\n\u003cli\u003eLeblanc, S., Roux, J., Tillaut, H., Le Page, E. \u0026amp; Leray, E. Disease-modifying therapy usage in patients with multiple sclerosis in France: A 6-year population-based study. \u003cem\u003eRevue Neurologique\u003c/em\u003e \u003cstrong\u003e177\u003c/strong\u003e, 1250\u0026ndash;1261 (2021).\u003c/li\u003e\n\u003cli\u003ePapukchieva, S. \u003cem\u003eet al.\u003c/em\u003e Shifting from the treat-to-target to the early highly effective treatment approach in patients with multiple sclerosis \u0026ndash; real-world evidence from Germany. \u003cem\u003eTher Adv Neurol Disord\u003c/em\u003e \u003cstrong\u003e17\u003c/strong\u003e, 17562864241237857 (2024).\u003c/li\u003e\n\u003cli\u003eStratil, A.-S., Papukchieva, S., Joschko, N., Meuth, S. G. \u0026amp; Friedrich, B. Shifts in treatment initiation patterns among newly diagnosed multiple sclerosis patients in Germany: a claims data analysis from 2017 to 2022. \u003cem\u003eTherapeutic Advances in Neurological Disorders\u003c/em\u003e \u003cstrong\u003e18\u003c/strong\u003e, 17562864251328576 (2025).\u003c/li\u003e\n\u003cli\u003eTallantyre, E. C. \u003cem\u003eet al.\u003c/em\u003e Real‐world persistence of multiple sclerosis disease‐modifying therapies. \u003cem\u003eEuro J of Neurology\u003c/em\u003e \u003cstrong\u003e31\u003c/strong\u003e, e16289 (2024).\u003c/li\u003e\n\u003cli\u003eEuropean Medicines Agency. Ocrevus : EPAR - Product Information. https://www.ema.europa.eu/en/medicines/human/EPAR/ocrevus. Accessed on 13 November 2025.\u003c/li\u003e\n\u003cli\u003eWang, Y., Wang, J. \u0026amp; Feng, J. Multiple sclerosis and pregnancy: Pathogenesis, influencing factors, and treatment options. \u003cem\u003eAutoimmunity Reviews\u003c/em\u003e \u003cstrong\u003e22\u003c/strong\u003e, 103449 (2023).\u003c/li\u003e\n\u003cli\u003eVukusic, S. \u003cem\u003eet al.\u003c/em\u003e Pregnancy and multiple sclerosis: 2022 recommendations from the French multiple sclerosis society. \u003cem\u003eMult Scler\u003c/em\u003e \u003cstrong\u003e29\u003c/strong\u003e, 11\u0026ndash;36 (2023).\u003c/li\u003e\n\u003cli\u003eBigaut, K. \u003cem\u003eet al.\u003c/em\u003e How to switch disease-modifying treatments in multiple sclerosis: Guidelines from the French Multiple Sclerosis Society (SFSEP). \u003cem\u003eMultiple Sclerosis and Related Disorders\u003c/em\u003e \u003cstrong\u003e53\u003c/strong\u003e, 103076 (2021).\u003c/li\u003e\n\u003cli\u003eChalkou, K., Vickers, A. J., Pellegrini, F., Manca, A. \u0026amp; Salanti, G. Decision Curve Analysis for Personalized Treatment Choice between Multiple Options. \u003cem\u003eMedical Decision Making\u003c/em\u003e \u003cstrong\u003e43\u003c/strong\u003e, 337\u0026ndash;349 (2023).\u003c/li\u003e\n\u003cli\u003eTrojano, M. \u003cem\u003eet al.\u003c/em\u003e Big multiple sclerosis data network: novel modelling approaches for real-world data analysis. \u003cem\u003eJ Neurol\u003c/em\u003e \u003cstrong\u003e272\u003c/strong\u003e, 754 (2025).\u003c/li\u003e\n\u003cli\u003eEuropean Medicines Agency. Lemtrada: EPAR - Product Information. https://www.ema.europa.eu/en/medicines/human/EPAR/lemtrada. Accessed on 17 November 2025.\u003c/li\u003e\n\u003cli\u003eCol, N. F. \u003cem\u003eet al.\u003c/em\u003e Implementing Shared Decision-Making for Multiple Sclerosis: The MS-SUPPORT Tool. \u003cem\u003eMultiple Sclerosis and Related Disorders\u003c/em\u003e \u003cstrong\u003e80\u003c/strong\u003e, 105092 (2023).\u003c/li\u003e\n\u003cli\u003eHe, D. \u003cem\u003eet al.\u003c/em\u003e Teriflunomide for multiple sclerosis. \u003cem\u003eCochrane Database of Systematic Reviews\u003c/em\u003e \u003cstrong\u003e2016\u003c/strong\u003e, (2016).\u003c/li\u003e\n\u003cli\u003eXu, Z. \u003cem\u003eet al.\u003c/em\u003e Dimethyl fumarate for multiple sclerosis. \u003cem\u003eCochrane Database of Systematic Reviews\u003c/em\u003e \u003cstrong\u003e2015\u003c/strong\u003e, (2015).\u003c/li\u003e\n\u003cli\u003eGonzalez-Lorenzo, M. \u003cem\u003eet al.\u003c/em\u003e Immunomodulators and immunosuppressants for relapsing-remitting multiple sclerosis: a network meta-analysis. \u003cem\u003eCochrane Database of Systematic Reviews\u003c/em\u003e \u003cstrong\u003e2024\u003c/strong\u003e, (2024).\u003c/li\u003e\n\u003cli\u003eCommittee for Medicinal Products for Human Use (CHMP) (2019) Guideline on the investigation of subgroups in confirmatory clinical trials. European Medicines Agency. https://www.ema.europa.eu/en/investigation-subgroups-confirmatory-clinical-trials-scientific-guideline. Accessed on 19 November 2025.\u003c/li\u003e\n\u003cli\u003eCollins, G. S. \u003cem\u003eet al.\u003c/em\u003e TRIPOD+AI statement: updated guidance for reporting clinical prediction models that use regression or machine learning methods. \u003cem\u003eBMJ\u003c/em\u003e e078378 (2024) doi:10.1136/bmj-2023-078378.\u003c/li\u003e\n\u003cli\u003eChitnis, T., Giovannoni, G. \u0026amp; Trojano, M. Complexity of MS management in the current treatment era. \u003cem\u003eNeurology\u003c/em\u003e \u003cstrong\u003e90\u003c/strong\u003e, 761\u0026ndash;762 (2018).\u003c/li\u003e\n\u003cli\u003eGhezzi, A. European and American Guidelines for Multiple Sclerosis Treatment. \u003cem\u003eNeurol Ther\u003c/em\u003e \u003cstrong\u003e7\u003c/strong\u003e, 189\u0026ndash;194 (2018).\u003c/li\u003e\n\u003cli\u003eOFSEP. https://www.ofsep.org/en/.\u003c/li\u003e\n\u003cli\u003eConfavreux, C., Compston, D. A., Hommes, O. R., McDonald, W. I. \u0026amp; Thompson, A. J. EDMUS, a European database for multiple sclerosis. \u003cem\u003eJ Neurol Neurosurg Psychiatry\u003c/em\u003e \u003cstrong\u003e55\u003c/strong\u003e, 671\u0026ndash;676 (1992).\u003c/li\u003e\n\u003cli\u003eRiley, R. D. \u003cem\u003eet al.\u003c/em\u003e Minimum sample size for developing a multivariable prediction model: PART II - binary and time-to-event outcomes. \u003cem\u003eStatistics in Medicine\u003c/em\u003e \u003cstrong\u003e38\u003c/strong\u003e, 1276\u0026ndash;1296 (2019).\u003c/li\u003e\n\u003cli\u003eBinuya, M. A. E., Engelhardt, E. G., Schats, W., Schmidt, M. K. \u0026amp; Steyerberg, E. W. Methodological guidance for the evaluation and updating of clinical prediction models: a systematic review. \u003cem\u003eBMC Medical Research Methodology\u003c/em\u003e \u003cstrong\u003e22\u003c/strong\u003e, 1\u0026ndash;14 (2022).\u003c/li\u003e\n\u003cli\u003eJanssen, K. J. M., Moons, K. G. M., Kalkman, C. J., Grobbee, D. E. \u0026amp; Vergouwe, Y. Updating methods improved the performance of a clinical prediction model in new patients. \u003cem\u003eJournal of Clinical Epidemiology\u003c/em\u003e \u003cstrong\u003e61\u003c/strong\u003e, 76\u0026ndash;86 (2008).\u003c/li\u003e\n\u003cli\u003eSteyerberg, E. W., Borsboom, G. J. J. M., van Houwelingen, H. C., Eijkemans, M. J. C. \u0026amp; Habbema, J. D. F. Validation and updating of predictive logistic regression models: A study on sample size and shrinkage. \u003cem\u003eStatistics in Medicine\u003c/em\u003e \u003cstrong\u003e23\u003c/strong\u003e, 2567\u0026ndash;2586 (2004).\u003c/li\u003e\n\u003cli\u003eOn behalf of Topic Group \u0026lsquo;Evaluating diagnostic tests and prediction models\u0026rsquo; of the STRATOS initiative \u003cem\u003eet al.\u003c/em\u003e Calibration: the Achilles heel of predictive analytics. \u003cem\u003eBMC Med\u003c/em\u003e \u003cstrong\u003e17\u003c/strong\u003e, 230 (2019).\u003c/li\u003e\n\u003cli\u003eVan Calster, B. \u003cem\u003eet al.\u003c/em\u003e A calibration hierarchy for risk models was defined: from utopia to empirical data. \u003cem\u003eJ Clin Epidemiol\u003c/em\u003e \u003cstrong\u003e74\u003c/strong\u003e, 167\u0026ndash;176 (2016).\u003c/li\u003e\n\u003cli\u003eSteyerberg, E. W. \u003cem\u003eClinical Prediction Models. A Practical Approach to Development, Validation, and Updating. Statistics for Biology and Health.\u003c/em\u003e (2019).\u003c/li\u003e\n\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":true,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":false,"hideJournal":false,"highlight":"","institution":"","isAcceptedByJournal":false,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"scientific-reports","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"scirep","sideBox":"Learn more about [Scientific Reports](http://www.nature.com/srep/)","snPcode":"","submissionUrl":"","title":"Scientific Reports","twitterHandle":"","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"stoa","reportingPortfolio":"Scientific Reports","inReviewEnabled":true,"inReviewRevisionsEnabled":true},"keywords":"Clinical Decision Support, Relapsing-Remitting Multiple Sclerosis, Therapy Switches, Clinical Practice Guidelines, Evidence-Based Practice","lastPublishedDoi":"10.21203/rs.3.rs-8295331/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-8295331/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003eModels for individualized treatment recommendations in relapsing\u0026ndash;remitting multiple sclerosis often select therapies based on predicted risks of relapse and/or 3-month confirmed disability progression (3m-CDP). Recommendations should be interpretable, align with guidelines, and reflect patient perspectives.\u003c/p\u003e \u003cp\u003eWe validated recently developed prognostic algorithms in eight subgroups defined by reason for therapy switch (intolerance, lack of efficacy, pregnancy desire, serious adverse events, programmed stop, personal convenience, therapy initiation, drug-holiday end). From the OFSEP registry, we analyzed 3768 therapy cycles (2017\u0026ndash;2021) with six commonly used therapies (interferon beta, glatiramer acetate, teriflunomide, dimethyl fumarate, fingolimod, natalizumab). Algorithms produced ranked therapy lists per outcome; certainty was quantified by entropy (0\u0026thinsp;=\u0026thinsp;clear; 4.78\u0026thinsp;=\u0026thinsp;maximal uncertainty).\u003c/p\u003e \u003cp\u003eRecommendations were more consistent for relapse (median entropy 0.54, IQR 0.39\u0026ndash;0.61) than for 3m-CDP (0.95, IQR 0.88\u0026ndash;1.01). Across subgroups, natalizumab most often ranked highest for relapse, whereas teriflunomide or interferon beta frequently ranked highest for 3m-CDP. Calibration varied across switcher subgroups and outcomes, while discrimination was comparable to overall set: C-index 0.776 (95% CI 0.753\u0026ndash;0.797) for 3m-CDP and 0.638 (95% CI 0.615\u0026ndash;0.661) for relapse. Guideline and relapse-based recommendations did not always align, notably in the pregnancy-desire subgroup. For 3m-CDP, guideline-based recommendations are scarce, precluding systematic comparison with model outputs. Agreement between the best-ranked therapy for relapse and 3m-CDP was zero (κ\u0026thinsp;=\u0026thinsp;0), underscoring outcome-dependent divergence.\u003c/p\u003e \u003cp\u003eAlgorithm-generated recommendations depend on the chosen outcome and may diverge from guideline-based practice or patient perspectives. Transparent communication of uncertainty and outcome trade-offs is essential for shared decision-making.\u003c/p\u003e","manuscriptTitle":"Validating therapy decisions by reasons for therapy switch in relapsing-remitting multiple sclerosis","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2026-04-07 10:20:24","doi":"10.21203/rs.3.rs-8295331/v1","editorialEvents":[{"type":"communityComments","content":0},{"type":"reviewersInvited","content":"","date":"2026-04-01T10:27:57+00:00","index":"","fulltext":""},{"type":"editorAssigned","content":"","date":"2026-03-31T10:23:15+00:00","index":"","fulltext":""},{"type":"editorInvited","content":"","date":"2025-12-18T10:10:05+00:00","index":"","fulltext":""},{"type":"checksComplete","content":"","date":"2025-12-16T19:19:55+00:00","index":"","fulltext":""},{"type":"submitted","content":"Scientific Reports","date":"2025-12-16T16:00:39+00:00","index":"","fulltext":""}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"scientific-reports","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"scirep","sideBox":"Learn more about [Scientific Reports](http://www.nature.com/srep/)","snPcode":"","submissionUrl":"","title":"Scientific Reports","twitterHandle":"","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"stoa","reportingPortfolio":"Scientific Reports","inReviewEnabled":true,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"889468ea-c48f-4969-b405-52bbc19e1d9c","owner":[],"postedDate":"April 7th, 2026","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"under-review","subjectAreas":[{"id":65692271,"name":"Health sciences/Diseases"},{"id":65692272,"name":"Health sciences/Medical research"},{"id":65692273,"name":"Health sciences/Neurology"},{"id":65692274,"name":"Biological sciences/Neuroscience"}],"tags":[],"updatedAt":"2026-04-07T10:20:24+00:00","versionOfRecord":[],"versionCreatedAt":"2026-04-07 10:20:24","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-8295331","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-8295331","identity":"rs-8295331","version":["v1"]},"buildId":"XKTyCvWXoU3ODBz1xrDgd","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}