Location and amount of joint involvement differentiates rheumatoid arthritis into different clinical subsets

Location and amount of joint involvement differentiates rheumatoid arthritis into different clinical subsets | 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 Location and amount of joint involvement differentiates rheumatoid arthritis into different clinical subsets Tjardo D. Maarseveen, Marc P. Maurits, Lavinia Agra Coletto, Simone Perniola, and 22 more This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-6256181/v1 This work is licensed under a CC BY 4.0 License Status: Published Journal Publication published 23 Oct, 2025 Read the published version in npj Digital Medicine → Version 1 posted 14 You are reading this latest preprint version Abstract Background Rheumatoid arthritis (RA) is a heterogeneous disease. Patients vary in symptoms, prognosis and treatment response, demonstrating the need for a more refined taxonomy. Objective To identify distinct phenotypic subsets of RA patients based on baseline clinical data, in order to advance understanding of disease etiology and treatment strategies. Methods We collected hematological, serological, and clinical data from RA-patients in the Leiden Rheumatology clinic(n = 1,387), and combined multimodal deep learning techniques with clustering to identify phenotypically distinct RA subsets. These clusters were tested for associations in clinical outcomes. Findings were replicated in clinical trial data (n = 307) and independent secondary care (9 clinics, n = 515), and further explored for histological differences in synovial tissue (n = 194). Results Four distinct RA subsets with different Joint Involvement Patterns (JIP), emerged: 1) foot-predominant arthritis, 2) seropositive oligoarticular disease, 3) seronegative hand arthritis, and 4) polyarthritis. We found high cluster stability, no physician influence, significant difference in remission rates (P = 0.007) and methotrexate failure ( P < 0.001) in initial and replication sets. The JIP-hand subgroup had significantly better outcomes. This was largest in the ACPA-positive stratum (JIP-hand versus JIP-foot (HR:0.37 (95%CI: 0.15–0.60) P < 0.001), JIP-hand versus JIP-poly HR:0.33 (95%CI: 0.15–0.72) P = 0.005). This was independent of baseline disease activity, clinical markers (RF, ACPA, Sex, Age), and symptom duration. Synovial histology showed both JIP-poly and JIP-hand had increased synovial lining and inflammatory infiltrate, with JIP-hand showing notably high stromal density. JIP-feet scored evenly across categories without standing out, while JIP-oligo had lower synovitis degree. Conclusions We identified and validated four distinct RA phenotypes characterized by joint involvement patterns, which associate with treatment outcomes and synovial histology. These findings may allow for targeted research into RA mechanisms and therapies. Health sciences/Diseases/Rheumatic diseases/Rheumatoid arthritis Biological sciences/Computational biology and bioinformatics/Classification and taxonomy Biological sciences/Computational biology and bioinformatics/Computational models Biological sciences/Computational biology and bioinformatics/Data integration Biological sciences/Computational biology and bioinformatics/Data processing Biological sciences/Computational biology and bioinformatics/Machine learning Biological sciences/Biological techniques/Bioinformatics Biological sciences/Immunology/Inflammation Biological sciences/Computational biology and bioinformatics Biological sciences/Immunology Health sciences/Medical research Figures Figure 1 Figure 2 Figure 3 Figure 4 Figure 5 INTRODUCTION Rheumatoid arthritis (RA) is a heterogeneous disease. The current classification criteria for RA were developed to approximate the decision to start early treatment and the exclusion of other diseases. At clinical presentation, patients vary in the number and pattern of joints involved, presence of extra articular manifestations and abnormalities in blood and synovial tissue [ 1 , 2 , 3 ]. The heterogeneity of RA also manifests in clinical outcomes, namely prognosis, treatment response and comorbidities. This evident diversity likely impacts the interpretation of treatment effect and etiologic factors such as genetics and downplay their importance altogether [ 4 ]. If phenotypic subsetting into more homogeneous groups is possible, it could improve research into the etiology of RA and enhance its treatment. For centuries, pattern recognition on clinical variables by doctors has been the driving force of disease identification and examination of the underlying etiologic mechanisms. Thus far clinicians have not identified the relevant (sub)patterns in RA. The presence of ACPA [ 5 , 6 , 7 , 8 ] and the age of onset [ 9 , 10 , 11 ] have been raised as possible dichotomous disease subsetting features. However, neither of these markers in isolation adequately addresses the heterogeneity and complexity of the disease. This suggests there are other factors involved. Cluster analysis combining a high number of factors has demonstrated its effectiveness in categorizing complex diseases (such as diabetes type II, asthma, osteoarthritis) into subtypes that differ in clinical outcomes or biological background [ 12 , 13 , 14 ]. In the context of RA, there is quite some focus on molecular phenotyping such as done by Lewis et al [ 15 ], who discovered patterns in synovial tissue at baseline, with the lymphoid-myeloid pathotype being a predictor for a poor outcome at disease onset [ 16 ]. Others used clinical and comorbidity information for clustering and identified four subsets, including one that exhibited a higher likelihood for biological DMARD initiation [ 17 ]. Likewise, Curtis et al [ 18 ] used clinical variables, though not exclusively at baseline, and identified five clusters that differed in disease activity, RA-duration and type of comorbidities. These outcomes are typically highly influenced by treatment decisions and events that occur independent of the specific RA type. Furthermore, detailed clinical information such as the pattern of involved joints may be relevant for disease differentiation as exemplified by psoriatic arthritis (PsA) [ 19 ], yet none of the previous studies capitalize on this information for clustering. Electronic Health Records (EHR) data provides a powerful asset for clustering as it encompasses a wide variety of data modalities (laboratory values, clinical examination, demographics) that each offer a unique perspective on the patient’s condition. The EHRs are collected as routine clinical care, and thus resemble the true patient population more closely than a study population collected with a particular hypothesis in mind. The diversity of data types does however pose a methodological challenge due to structural differences between the data modalities. The recent surge of deep learning tools [ 20 ], offers the possibility to combine different EHR-layers into a patient representation by extracting the (hidden) factors that capture most variation in the data. At present, there exist many machine learning (ML) techniques to learn the relevant (clinical) patterns, and encode patients accordingly. These embeddings can be used to detect patients' subgroups, identify patterns, build predictive models or assist in making disease classifications. The literature reports that clustering on top of these embeddings typically outperforms conventional techniques in the case of high dimensional or complicated data [ 21 , 22 , 23 ]. In this study we aimed to dissect the clinical heterogeneity of RA by using the symptoms at initial presentation, so before external factors such as treatment interfere. We hypothesize that the location of the involved joints and the inflammatory patterns observed in the blood play a role in subsetting RA, similar to their significance in distinguishing PsA from RA [ 19 ]. To achieve this we make use of advanced data-driven techniques to identify and analyze disease differentiating signatures based on initial clinical variables, and see if they relate to clinical outcomes and histological synovial features. METHODS Patients Our study comprises three different phases: a i) developmental phase where we identify and validate subtypes in a discovery set according to long term outcomes (set A), a ii) replication phase where we cluster novel patients using historic trial- (set B) and external hospital data (set C) to infer generalizability by replicating the treatment analysis and finally iii) a downstream analysis in external hospital data (set D) where we explore differences between clusters in their synovial tissue. Set A consisted of 1,387 RA-patients that visited the rheumatology outpatient clinic of the Leiden University Medical Centre (LUMC) for the first time between August 29th, 2011 till December 1st 2022. RA diagnosis was based on the physician’s diagnosis within 1 year since first visit.[ 24 , 25 ]. Set B concerned 307 RA-patients from the IMPROVED trial that were recruited between March 2007 till September 2010 [ 26 ]. This trial recruited undifferentiated arthritis and early RA with less than 2 years of symptoms. We selected only those patients who met the ACR2010 criteria within one year after inclusion. All patients received MTX at baseline and were randomized into two arms of treatment intensification if they did not reach remission after 4 months. Set C included 515 RA patients from Reumazorg Zuid West Nederland (RZWN), collected between January 2015 and December 1, 2022. These patients were from nine different hospitals across the south west of the Netherlands, with the largest groups coming from Goes (n = 157), Roosendaal (n = 153), and Vlissingen (n = 49). Herein, the diagnosis of RA was defined as having an ICD-code for RA and starting with a conventional DMARD. Set D included 262 RA-patients fulfilling the ACR2010 criteria for RA from the SYNGem Biopsy Unit cohort of the Fondazione Policlinico Universitario A. Gemelli IRCCS–Università Cattolica del Sacro Cuore from Rome, Italy, all of whom underwent minimally invasive ultrasound-guided synovial tissue biopsy at their first rheumatological evaluation within their clinical routine management. Each tissue was processed for H&E staining and synovitis was graded using the total Krenn synovitis score (KSS) [ 27 ]. Across all sets, a minimum follow-up of 1-year was required to ascertain the diagnosis of RA. Prior to conducting the study, we acquired approval from the ethics committee of the LUMC. Patients and public were not involved during the development, execution, and dissemination of the study. Preprocessing of electronic health records To construct patient phenotypic profiles, we extracted information on serology (RF and ACPA), location of joint involvement (tender- and swollen joints (TJC and SJC)), demographics, blood profiles (hemoglobin, hematocrit, leukocyte- and thrombocytes levels) and ESR at baseline (Table S1 ). Baseline was defined as the first visit to the clinic (set A&C) or the moment of inclusion in the trial (set B). Patients with missing lab or joint location variables were dropped (Fig. S1 ). We normalized the numerical data using a Yeo-Johnson transformation, except for the ESR levels where we applied a log transformation due to their log-normal distribution. For the categorical data, we implemented one-hot encoding, which created separate binary fields (yes/no) for each possible category value. Construction of patient embedding We integrated the different EHR data types to create a condensed patient representation (called a patient embedding) using a multi-modal autoencoder (MMAE). This MMAE had a narrowing structure with encoder layers of 128, 64, and finally 8 neurons. For categorical data, we used Sigmoid activation functions with Bernoulli loss, while numerical data utilized ReLU activation with Gaussian loss. To prevent overfitting, we compared performance between our training set (80% of data) and validation set (20%), using the Adam optimizer throughout. We visualized how much each type of EHR data contributed to our final embedding using a flameplot analysis (Fig. S2 (0)) After creating the patient embedding, we grouped patients into subcommunities based on clinical similarities using PhenoGraph [ 28 ]. We selected PhenoGraph rather than more common methods like K-means clustering because it performs better with our sparse, high-dimensional medical data (see our supplemental materials for additional technical details). Cluster interpretation For each cluster, we examined the characteristics and visualized the phenotype on a pictorial mannequin with an integrated heatmap to demonstrate the number of involved joints. We used a surrogate ML-technique to model the cluster assignment and subjected this model to a SHAP (SHapley Additive exPlanations) [ 29 ] analysis to retrospectively identify the most important variables per cluster. The SHAP plots show the strength and direction of impact of that variable for each patient (also those who are not assigned to that cluster). Cluster validation To confirm that our identified clusters comprised a stable and relevant partitioning, we performed a number of validation checks. We examined cluster stability by measuring how often patients co-cluster across 1000 random subsets of the data and assessed possible factors influencing the partitioning. Next, we conducted a Local Inverse Simpson's Index analysis [ 30 ] to assess whether physicians are evenly distributed across the clusters or if the patient subgroups merely reflect the reporting differences between physicians (see supplemental material). To infer the clinical relevance, clinical outcomes were evaluated using a Cox regression model, including: time to MTX-failure (defined by replacement of- or adding an additional DMARD to MTX) and remission (DAS44 < 1.6) within one year. Moreover, we evaluated the replicability on an external dataset (set B & set C), where individuals are assigned to clusters in accordance with the previously learned patient embedding (see supplemental material). Diversity in synovial characteristics In addition to the clinical validation, we compared the characteristics of the synovium across the four subgroups, by repeating the clustering analysis in set D (SYNGem cohort). Specifically, we compared the overall Krenn synovitis score (KSS) [ 27 ] and its individual components (lining layer hyperplasia, stromal density and inflammatory infiltrate). Each KSS component was measured on an ordinal scale (0: none, 1: low, 2: moderate, 3: severe). To analyze this, we used ordinal logistic regression, which estimates the odds of transitioning from one severity level to the next, thus relying on sufficient representation. However, since our study focused on treatment-naive RA patients with active synovitis, the lowest level (0: none) for each KSS subitem was rarely observed. As a result, when present, it was combined with the 'low' category (1: low). Statistical tests We used the Kruskal-Wallis test followed by Dunn's post-hoc test to compare numerical values across multiple groups. For survival analysis we used the log-rank test to examine the overall trend and a univariate Cox-regression [ 31 ] to quantify the cluster differences. We inferred the DAS-remission status during the survival analysis, carrying the last observation forward if it was missing (effectively the same as time to event). The proportional hazards assumption was verified by examining the Schoenfeld residuals [ 32 ]. We also adjusted the Cox regression model for MTX-response on covariates suspected to influence treatment response (e.g. ACPA positivity, RF positivity, SJC, TJC, Sex and Age). The statistical significance was inferred with ANOVA. For the ordinal values of the KSS components we used an ordered logit, followed up by a post-hoc Wald test to look at individual differences. Web Interface All of our scripts are publicly available online at Github [ 33 ]. Moreover, we developed an interactive web tool ( https://knevel-lab.github.io/Rheumalyze/ ) that enables users to map their cohort onto our patient embedding, allowing them to identify the JIP phenotypes present in their data. RESULTS Patients We retrieved 2,691 RA patients for training set A of whom, 1,387 were included in our study based on the availability of lab values and joint counts. For the replication, set B and set C had 364 and 1,227 RA cases, of whom 307 and 515 had complete information (Table S2 , Fig. S1 ). For the downstream analysis, we looked at biopsy data from 264 patients of set D, of whom 194 patients had synovial tissue taken from the same joint (Table S3). A workflow diagram for the different phases of the study is shown in Fig. 1 . Each dataset captured a typical early RA population [ 32 , 33 ]. In comparison to set A, patients in the replication sets exhibited higher rates of seropositivity and had fewer tender joints. On average, set B patients were younger and set C patients had less inflammation, with a median ESR of 16. We used the phenotypic variables (see methods) of set A to construct the patient embedding. Four clusters separated by joint location, serology and blood values The patient embedding showed four different clusters (Fig. S3, S4) which were not determined by any single clinical variable, as indicated by the wide dispersion of values. The primary variation across the clusters was their difference in affected joints, leading us to name them Joint Involvement Patterns (JIP). Additionally, the clusters differed in levels of inflammation, age, and seropositivity (Table S4, Fig. S5).: ● Cluster 1: JIP-foot moderate number of involved joints, particularly feet joints, younger patients, low leukocyte and thrombocyte levels. ● Cluster 2: JIP-oligo limited joint involvement and mostly seropositive patients. ● Cluster 3 JIP-hand elderly patients, symmetrical polyarthritis of hands, seronegative. ● Cluster 4 JIP-polyarthritis majority seronegative polyarthritis in hand and feet though with lower ESR. The clusters were stable with an average > 80% of patients grouping together in the same cluster over the 1000 iterations in the stability analysis (Fig. S6&S7). In fact, the stability was better in our combined multi-modal approach than if we take each data type (numeric/categorical) separately (Fig. S8). The clusters were not driven by treating physicians (Fig. S9) and were generalizable across different validation sets (Table 1 , Fig. S10&S11), showing similar joint involvement patterns (Fig. 2 ). Also, the identified patient clusters did not seem to represent different disease stages as the cluster with the longest symptom duration had the lowest joint count and vice versa. There were differences in cluster prevalences between the validation sets (Table S5&S6). Table 1 Baseline characteristics of the different patient clusters (set A + replication sets B & C) JIP-Foot JIP-Oligo JIP-Hand JIP-Poly N 596 761 450 402 Sex, female ɣ [n(%)] 389 (65.3) 505 (66.4) 272 (60.4) 272 (67.7) Age ɣ (SD, yr) 56.6 (14.6) 59.5 (14.6) 66.4 (13.2) 54.7 (14.7) RF ɣ [n(%)] 370 (62.1) 497 (65.3) 200 (44.4) 196 (48.8) ACPA ɣ [n(%)] 355 (59.6) 449 (59.0) 162 (36.0) 183 (45.5) ESR ɣ (IQR, mm/hr) 22 (9–36) 24 (11–38) 28 (13–48) 22 (9–40) DAS44(3) (IQR) 3.5 (3.0–4.0) 2.4 (1.9–2.8) 3.7 (3.2–4.3) 4.6 (4.0-5.5) SJC (IQR) 8 (5–12) 2 (1–5) 10 (7–15) 15 (9–22) TJC (IQR) 11 (8–15) 3 (2–5) 11 (7–15) 22 (15–30) DAS28(3) (IQR) 5.2 (4.5-6.0) 4.0 (3.3–4.7) 5.5 (4.9–6.2) 6.5 (5.5–7.2) Follow up (IQR, days) 1307 (733–2020) 1428 (760–2082) 1127 (566–1875) 1512 (1012–2246) Symptom duration * (IQR, days) 143 (56–364) 186 (70–399) 101 (48–279) 147 (56–357) Where SD, standard deviation; RF, rheumatoid factor; ACPA, anti-cyclic citrullinated peptide antibodies; ESR, erythrocyte sedimentation rate; IQR, interquartile range; DAS, three component disease activity score (either 44 or 28 joint scheme); SJC, swollen joint count; TJC, tender joint count; ɣ, clinical variables that were used for clustering (set A); *, symptom duration was only calculated for patients from set A and B. Validation on clinical outcomes beyond baseline In set A, 80% of patients received MTX as an initial drug across all clusters. The Kaplan Meier curves show a difference in MTX failure between the clusters: 27%, 23%, 16%, 30% (for cluster 1–4, P < 0.001, Fig. 3 a). The JIP-hand had clearly the best prognosis, where patients were twice as likely to stay on MTX than the most severe disease subtype JIP-poly (HR 0.48 (95% CI 0.35–0.77), P < 0.001). Additionally, the JIP-hand did better than the JIP-foot (HR:0.55 (0.37–0.82) P = 0.003). Consistent with MTX response, we observed differences in remission rates: 44.3%, 47.4%, 55.7%, 38.5% (for cluster 1–4, P = 0.007, Fig. 3 b) with the biggest difference between the JIP-hand and JIP-poly (HR 1.65 (95% CI 1.2–2.29), P = 0.002), also when corrected for baseline disease activity (Fig. S12). ACPA within the clusters Since the literature reports that ACPA is indicative of persistent disease [ 36 ], we examined whether the ACPA status was the main factor driving the difference in MTX failure. In set A we found a higher treatment failure in ACPA positive than negative patients (27.6% versus 22.0%, Fig. 4 ), though it was not significant ( P = 0.057). Moreover, the association of ACPA with MTX-failure differed within the clusters ( P < 0.001, Fig. 4 ). The difference between the JIP-hand with the foot clusters JIP-poly and JIP-foot was larger within the ACPA-positive stratum (JIP-hand vs JIP-foot (HR:0.37 (0.15–0.60) P < 0.001), JIP-hand vs JIP-poly HR:0.33 (0.15–0.72) P = 0.005) (Fig. 4 b) For remission we could not find this difference in the ACPA-positive stratum. The good response in JIP-hand raised the question whether this group overrepresented patients with parvovirus induced arthritis instead of RA, but none of our clusters were enriched for parvovirus positive patients (Fig. S13). Replication In set C, 79% of patients received MTX as the initial drug across all clusters, whereas in set B all patients were administered MTX. In both replication sets we again observed a better outcome of JIP-hand for MTX failure (global P < 0.001, and P < 0.001). Remission could only be tested in replication set B where it confirmed our previous finding ( P < 0.001). Consistent with the original finding, the difference between JIP-hand and JIP-poly was particularly strong in the ACPA positive stratum in both replication set B (OR: 0.22 (0.12–0.63), P = 0.017) and set C (HR:0.38 (0.22–0.68), P < 0.001). Within the ACPA positive stratum we also found the significant difference between JIP-hand and JIP-foot back in replication set C (HR:0.37 (0.15–0.93), P = 0.034), though not in replication set B (OR: 1.03 (0.34–3.05) P = 0.801). All the analyses remained significant when corrected for baseline DAS (Fig. S14). Even after adjusting for symptom duration in sets A and B, the association between the cluster and treatment persisted (Fig. S15) Informative value of clusters beyond known risk factors for MTX failure To ascertain that the cluster association with treatment outcome was not merely the product of already established clinical markers we adjusted the Cox regression model (Fig. S16). Here, the inclusion of other well-known contributing factors like ACPA, RF or the number of affected joints did not diminish the additive value of clustering ( P = 0.020 in set A, and P = 0.019 in set C). Diversity in synovial characteristics Next, we examined histological variations between patient clusters in Set D according to the Krenn synovitis components [ 27 ]. Biopsies were taken from the most inflamed joint, typically the knee or wrist. Due to the limited number of samples from other joints, a thorough comparison across different anatomical locations was not feasible. Therefore, we primarily focused on a single, consistent location that was involved in most clusters —specifically, the knee joints. In total, we analyzed synovial tissue from 194 patients, distributed across four clusters: 27 in JIP-foot, 49 in JIP-oligo, 86 in JIP-hand, and 32 in JIP-poly (see Fig. S17). The Krenn synovitis score [ 27 ] significantly differed between the patients from the four different clusters with a mean score of 5,4, 5 and 6 respectively ( P = 0.004) with the highest in the JIP-Poly and the lowest in JIP-Oligo clusters respectively ( Fig. 5 ) . The clusters showed significant histopathological variations between the JIP-variants in lining layer hyperplasia ( P = 0.026), stromal density ( P = 0.045), and inflammatory infiltrate (P = 0.002). These were primarily driven by the distinction between the more stereotypical RA-clusters (JIP-hand/JIP-poly) and JIP-oligo, which was characterized by low-grade synovitis. In JIP-poly 43.8% had severe synovial lining hyperplasia compared to 14.3% in JIP-oligo ( P = 0.002). Moreover, 21.9% had severe inflammatory cell infiltration compared to 6.1% in JIP-oligo ( P = 0.001) and 18.8% in stromal density versus 8.2% for JIP oligo ( P = 0.034). JIP-hand exhibited increased synovitis levels compared to JIP-oligo for lining layer hyperplasia (27.9% vs 14.3%; P = 0.015) and inflammatory infiltrate (17.4% vs 6.1%; P = 0.013). Furthermore, 29.1% had severe stromal density compared to 7.4% in JIP-foot ( P = 0.211) and 8.2% in JIP-oligo ( P = 0.013). Though, JIP-foot had a similar overall synovitis as JIP-hand, it was not characterized by any particular KSS component, rather it had moderate inflammatory activity across all elements. Since KSS was previously shown to be linked to disease activity [ 3 ], we adjusted for DAS categories to ensure the cluster difference was not merely an effect of the clinical metric. After correction, differences in lining layer hyperplasia (P = 0.035) and inflammatory infiltrate (P = 0.005) persisted between clusters, while stromal density differences became non-significant (P = 0.082). For a complete overview, we also examined whether sampling from joint areas specific to each JIP phenotype would yield different results compared to using only knee biopsies. We collected samples from lower extremities for JIP-foot patients, knees for JIP-oligo patients, fingers or wrists for JIP-hand patients, and knees or wrists for JIP-poly patients. This targeted sampling approach produced results similar to those we found when analyzing knee biopsies alone (Fig. S18). DISCUSSION Through deep learning and clustering analysis of real-world clinical data, we identified four distinct RA phenotypes at baseline: foot-dominant (JIP-foot), seropositive oligoarticular (JIP-oligo), seronegative hand-dominant (JIP-hand), and polyarticular (JIP-poly) disease. While our hypothesis-free approach enabled detection of novel non-linear clinical signatures, we mitigated the risk of spurious correlations through rigorous validation, including stability testing, clinical outcome validation, independent cohort replication, and synovial histological correlation. A key finding was the marked difference in treatment success between hand and foot clusters. Both foot-dominant clusters (JIP-foot, JIP-poly) showed higher MTX failure and lower remission rates compared to the JIP-hand, independent of baseline joint involvement, symptom duration, or treatment timing. This validates previous cross-sectional observations of poor prognosis in feet/ankle-involved disease [ 37 ], though our study uniquely demonstrates this association in treatment-naïve patients at presentation. The impact of foot involvement on treatment failure rivaled that of ACPA-positivity as an independent risk factor. This observation holds particular clinical relevance given the frequent exclusion of lower extremity joints from disease activity scores, despite their common involvement [ 37 , 38 ]. Contrary to common assumptions, we did not observe a clear ACPA dichotomy. This finding aligns with previous baseline studies which demonstrate that despite known differences in risk factors and prognosis between ACPA-positive and ACPA-negative patients, the clinical phenotype at initial diagnosis is similar for both groups [ 36 , 39 , 40 ]. ACPA prevalence was lowest in typical RA clusters (JIP-hand, JIP-poly) and highest in the oligoarticular cluster (JIP-oligo). While this pattern might partially reflect classification criteria [ 41 ], our use of one-year diagnosis validation and physician diagnosis in sets A and C minimizes misclassification bias. The high ACPA-positivity in JIP-foot corroborates recent findings of increased foot involvement in ACPA-positive patients [ 42 ]. The treatment response disparity between hand and foot clusters was most pronounced in ACPA-positive patients. Across replication sets, we consistently observed significant differences between JIP-hand and JIP-poly, with two of three datasets also showing increased response in JIP-hand versus JIP-foot. For remission, we found that the different cluster-associated outcomes vanished in ACPA-positive patients, possibly reflecting the impact of targeted therapy intensification protocols particularly for ACPA positive patients. Patients with hand-dominant joint inflammation (JIP-hand) showed surprisingly good outcomes, prompting us to explore several possible explanations. We found that these positive results could not be explained solely by how long patients had symptoms, their disease activity at baseline, or their lower prevalence of ACPA in that cluster. Further analysis of synovial tissue samples revealed distinct histological differences. The JIP-poly and JIP-hand groups both showed severe synovitis. JIP-poly patients had more lining hyperplasia and sublining leukocytic infiltration, while JIP-hand showed more severe stromal density. Compared to JIP-oligo, which had mostly mild inflammation, these clusters showed significant differences. When we adjusted for disease activity levels, the difference in stromal density disappeared, but the differences in inflammatory infiltrate and cell lining hyperplasia remained significant. The marked difference within the histological environment, between the aggressive patterns in JIP-poly and JIP-hand groups and the mild inflammation in JIP-oligo further supports the notion that these may represent fundamentally different disease subtypes. Though, more research is needed to better understand the underlying molecular and cellular mechanisms, and their potential relation to treatment. The combination of short symptom duration, widespread joint inflammation, and low seropositivity might suggest that JIP-hand could represent a self-limiting reactive arthritis [ 43 , 44 ], though we found no differences in parvovirus positivity between clusters. An alternative explanation is that the current therapeutic agents may be optimized for JIP-hand -like presentations, given clinical trials' emphasis on upper extremity outcomes. Our clusters captured previously described age-related subsets, including elderly-onset RA (EORA) in JIP-hand, characterized by higher inflammation markers, lower female prevalence, and reduced autoantibody positivity [ 10 , 11 ]. However, our analysis revealed more granular subtypes beyond the EORA/YORA dichotomy. A limitation of our study is that we defined MTX success based on changes in medication, which could include switches due to side effects, though this likely underestimates rather than overestimates the observed associations. Center-specific therapeutic approaches varied but did not affect cluster-outcome associations. While temporal cluster stability was not directly assessed, previous evidence of consistent joint involvement patterns [ 45 ] and cross-sectional associations support stability over time. Another limitation is that we solely focused on knee biopsies due to insufficient data for other joint locations. This prevented us from exploring and comparing different tissue environments even though they might be crucial for understanding the different phenotypes. Important to underline is that our identified clusters are not set in stone. Though we observed a high robustness of our clusters, patients laid on a gradient (Fig. S4) and did not segregate in clearly separable modules. The cluster structure that we identified could also be summarized into more or fewer clusters and the clusters might become clearer when more layers of information are added. Such types of information could be genetics, gene expression patterns and molecular profiles from blood [ 15 , 46 ]. Despite these limitations, our study demonstrates the value of unsupervised, data-driven approaches in uncovering hidden disease patterns, with joint involvement patterns emerging as a major axis of variation. In conclusion, our clustering analysis revealed four baseline RA phenotypes characterized by distinct hand and foot involvement patterns that predict one-year clinical outcomes, and correspond to histological differences. This data-driven approach provides greater granularity than traditional age or ACPA-based dichotomies, suggesting distinct etiologies that warrant further biological investigation. Declarations Contributors RK and TM developed the study design together with EB and MPM. TM ran the cluster analysis, while SB ran the permutation analysis. BBdK repeated the survival analysis in the replication set. SAB and CA provided the replication data of the IMPROVED trial, while KG and JVvD provided the data from Reumazorg Zuid West Nederland. SA, LAC and SP provided and processed the SYNGem cohort data for the histological analysis. All authors contributed to the interpretation of the results, and provided ideas for further downstream analysis. AHM annotated the classification criteria. RK and TM drafted the first version of the manuscript. All authors reviewed and approved the final draft of the manuscript to be submitted. Funding This project received funding from two major European Union Horizon grants for research and innovation. The first grant supported the SQUEEZE project (activity No. 101095052) and included funding for TDM, MPM, NS, TWJH, and RK. The second grant supported the SPIDeRR project (activity No. 101080711) and funded TDM, MPM, NS, TWJH, MJTR, EvdA, and RK. Additionally, the European Research Council provided financial support for the Glycan Switch project (activity No. 101071386), specifically for TWJH and RK. This study also received co-funding from the ZonMW Klinische Fellow program (No. 40-00703-97-19069) and the ZonMW Open Competitie (No. 09120012110075): for TDM, MPM and RK. The authors declare no competing interests. Competing interest The authors declare no competing interests. Patient and public involvement Patients and/or the public were not involved in the design, or conduct, or reporting, or dissemination plans of this research. Ethics approval We received ethical approval from the Medical Ethics Committee (METC) at Leiden University Medical Center according to study protocol B18.057. Data availability statement We built an interactive webtool (https://knevel-lab.github.io/Rheumalyze/) that users can use to project their patients on the Leiden data, to infer the JIP-phenotype within their own dataset. Moreover, we have made our scripts available in a public repository at GitHub [33]. Study data is available upon reasonable request. Acknowledgements We would like to express our gratitude to Nick Bos for developing a web tool to cluster patients. Our thanks also go to Samantha Jurado-Zapata and David Steeman for their assistance with extracting and processing Electronic Health Record data from the Leiden University Medical Center. Furthermore, we deeply appreciate Bas van der Wal for his efforts in preparing the Reumazorg Zuidwest Nederland data and Joy van der Pol for his work on preparing the IMPROVED trial data. References Grassi W, De Angelis R, Lamanna G, et al . The clinical features of rheumatoid arthritis. Eur J Radiol. 1998;27 Suppl 1:S18-24. doi: 10.1016/s0720-048x(98)00038-2 . PMID: 9652497. Heidari B. Rheumatoid Arthritis: Early diagnosis and treatment outcomes. Caspian J Intern Med. 2011 Winter;2(1):161 – 70. doi: 10.1016/s0720-048x(98)00038-2 . PMID: 24024009; Alivernini S, Tolusso B, Gessi M, et al. Inclusion of Synovial Tissue-Derived Characteristics in a Nomogram for the Prediction of Treatment Response in Treatment-Naive Rheumatoid Arthritis Patients. Arthritis Rheumatol. 2021;73(9):1601–1613. doi: 10.1002/art.41726 . Epub 2021 Aug 9. PMID: 33750008; PMCID: PMC8457106. Padyukov L. Genetics of rheumatoid arthritis. Semin Immunopathol. 2022;44(1):47–62. doi: 10.1007/s00281-022-00912-0 . PMID: 35088123; Syversen SW, Goll GL, van der Heijde D, et al . Prediction of radiographic progression in rheumatoid arthritis and the role of antibodies against mutated citrullinated vimentin: results from a 10-year prospective study. Annals of the Rheumatic Diseases 2010;69:345-351.5. doi: 10.1136/ard.2009.113092 . PMID: 19648126. Matthijssen XME, Niemantsverdriet E, Huizinga TWJ, et al . Enhanced treatment strategies and distinct disease outcomes among autoantibody-positive and -negative rheumatoid arthritis patients over 25 years: A longitudinal cohort study in the Netherlands. PLOS Medicine 2020;17(9): e1003296. doi: 10.1371/journal.pmed.1003296 Pedersen M, Jacobsen S, Klarlund M, et al . Environmental risk factors differ between rheumatoid arthritis with and without auto-antibodies against cyclic citrullinated peptides. Arthritis Res Ther. 2006;8(4):R133. doi: 10.1186/ar2022 . PMID: 16872514; Nordberg LB, Lillegraven S, Lie E and the ARCTIC working group, et al. Patients with seronegative RA have more inflammatory activity compared with patients with seropositive RA in an inception cohort of DMARD-naïve patients classified according to the 2010 ACR/EULAR criteria. Annals of the Rheumatic Diseases 2017;76:341–345. doi: 10.1136/annrheumdis-2015-208873 Sugihara T, Harigai M. Targeting Low Disease Activity in Elderly-Onset Rheumatoid Arthritis: Current and Future Roles of Biological Disease-Modifying Antirheumatic Drugs. Drugs Aging. 2016;33(2):97–107. doi: 10.1007/s40266-015-0341-2 . PMID: 26833350; Deal CL, Meenan RF, Goldenberg DL, et al. The clinical features of elderly-onset rheumatoid arthritis. A comparison with younger-onset disease of similar duration. Arthritis Rheum. 1985;28(9):987 – 94. doi: 10.1002/art.1780280905 . PMID: 4038365. Sugihara T, Harigai M. Targeting Low Disease Activity in Elderly-Onset Rheumatoid Arthritis: Current and Future Roles of Biological Disease-Modifying Antirheumatic Drugs. Drugs Aging. 2016;33(2):97–107. doi: 10.1007/s40266-015-0341-2 . PMID: 26833350; Ahlqvist E, Storm P, Käräjämäki A, et al . Novel subgroups of adult-onset diabetes and their association with outcomes: a data-driven cluster analysis of six variables. Lancet Diabetes Endocrinol. 2018;6(5):361–369. doi: 10.1016/S2213-8587(18)30051-2 . PMID: 29503172. Howrylak JA, Fuhlbrigge AL, Strunk RC, et al . Childhood Asthma Management Program Research Group. Classification of childhood asthma phenotypes and long-term clinical responses to inhaled anti-inflammatory medications. J Allergy Clin Immunol. 2014;133(5):1289 – 300, 1300.e1-12. doi: 10.1016/j.jaci.2014.02.006 . PMID: 24892144; Angelini F, Widera P, Mobasheri A, et al . Osteoarthritis endotype discovery via clustering of biochemical marker data. Ann Rheum Dis. 2022;81(5):666–675. doi: 10.1136/annrheumdis-2021-221763 . PMID: 35246457. Lewis MJ, Barnes MR, Blighe K, et al . Molecular Portraits of Early Rheumatoid Arthritis Identify Clinical and Treatment Response Phenotypes. Cell Rep. 2019;28(9):2455–2470.e5. doi: 10.1016/j.celrep.2019.07.091 . PMID: 31461658; Lliso-Ribera G, Humby F, Lewis M, et al . Synovial tissue signatures enhance clinical classification and prognostic/treatment response algorithms in early inflammatory arthritis and predict requirement for subsequent biological therapy: results from the pathobiology of early arthritis cohort (PEAC). Ann Rheum Dis. 2019;78(12):1642–1652. doi: 10.1136/annrheumdis-2019-215751 . PMID: 31582377; Jung SM, Park KS, Kim KJ. Clinical phenotype with high risk for initiation of biologic therapy in rheumatoid arthritis: a data-driven cluster analysis. Clin Exp Rheumatol. 2021 Nov-Dec;39(6):1282–1290. doi: 10.55563/clinexprheumatol/24zas6 . PMID: 33635214. Curtis JR, Weinblatt M, Saag K, et al . Data-Driven Patient Clustering and Differential Clinical Outcomes in the Brigham and Women's Rheumatoid Arthritis Sequential Study Registry. Arthritis Care Res (Hoboken). 2021;73(4):471–480. doi: 10.1002/acr.24471 . Merola JF, Espinoza LR, Fleischmann R. Distinguishing rheumatoid arthritis from psoriatic arthritis. RMD Open. 2018;4(2):e000656. doi: 10.1136/rmdopen-2018-000656 . PMID: 30167326; Onken A, Yague JG, Sakata S. Deep multimodal autoencoders for identifying latent representations of spike counts and local field potentials. Bernstein Conference 2018. doi: 10.12751/nncn.bc2018.0223 Steinbach M, Ertöz L, Kumar V. The Challenges of Clustering High Dimensional Data. In: Wille LT, editor. New Directions in Statistical Physics: Econophysics, Bioinformatics, and Pattern Recognition. Berlin, Heidelberg: Springer Berlin Heidelberg; 2004. p. 273–309. Available from: https://doi.org/10.1007/978-3-662-08968-2_16 Cai J, Fan J, Gu W, et al . Efficient deep embedded subspace clustering. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition 2022 (pp. 1–10). Castela Forte J, Yeshmagambetova G, van der Grinten ML, et al . Identifying and characterizing high-risk clusters in a heterogeneous ICU population with deep embedded clustering. Sci Rep. 2021;11(1):12109. doi: 10.1038/s41598-021-91297-x . PMID: 34103544; Maarseveen TD, Meinderink T, Reinders MJT, et al . Machine Learning Electronic Health Record Identification of Patients with Rheumatoid Arthritis: Algorithm Pipeline Development and Validation Study. JMIR Med Inform. 2020;8(11):e23930. doi: 10.2196/23930 . PMID: 33252349; Maarseveen TD, Maurits MP, Niemantsverdriet E, et al . Handwork vs machine: a comparison of rheumatoid arthritis patient populations as identified from EHR free-text by diagnosis extraction through machine-learning or traditional criteria-based chart review. Arthritis Res Ther. 2021;23(1):174. doi: 10.1186/s13075-021-02553-4 . PMID: 34158089; Heimans L, Wevers-de Boer KV, Visser K, et al . A two-step treatment strategy trial in patients with early arthritis aimed at achieving remission: the IMPROVED study. Ann Rheum Dis. 2014;73(7):1356-61. doi: 10.1136/annrheumdis-2013-203243 . PMID: 23716067. Krenn V, Morawietz L, Burmester GR, Kinne RW, Mueller-Ladner U, Muller B & Haupl T. Synovitis score: discrimination between chronic low‐grade and high‐grade synovitis. Histopathology, 2006; 49(4), 358–364. doi: 10.1111/j.1365-2559.2006.02508.x . PMID: 16978198. Levine JH, Simonds EF, Bendall SC, et al . Data-Driven Phenotypic Dissection of AML Reveals Progenitor-like Cells that Correlate with Prognosis. Cell. 2015;162(1):184 – 97. doi: 10.1016/j.cell.2015.05.047 . PMID: 26095251; Roth, AE. The Shapley Value: Essays in Honor of Lloyd S. Shapley. Cambridge: Cambridge University Press. 1988. doi: 10.1017/CBO9780511528446 Korsunsky I, Millard N, Fan J, et al . Fast, sensitive and accurate integration of single-cell data with Harmony. Nat Methods. 2019;16(12):1289–1296. doi: 10.1038/s41592-019-0619-0 . Epub 2019 Nov 18. PMID: 31740819; Cox DR. Regression models and life-tables. Journal of the Royal Statistical Society: Series B (Methodological), 1972; 34(2), 187–202. Schoenfeld, D. Chi-squared goodness-of-fit tests for the proportional hazards regression model. Biometrika, 1980; 67(1), 145–153. Maarseveen TD. EHR Clustering RA. GitHub. Published: August 8, 2022. Link: https://github.com/levrex/EHR-Clustering-RA Nishimura K, Sugiyama D, Kogata Y, et al . Meta-analysis: Diagnostic Accuracy of Anti–Cyclic Citrullinated Peptide Antibody and Rheumatoid Factor for Rheumatoid Arthritis. Ann Intern Med.2007;146:797–808. doi: 10.7326/0003-4819-146-11-200706050-00008 van Nies JAB, Tsonaka R, Gaujoux-Viala C, et al . Evaluating relationships between symptom duration and persistence of rheumatoid arthritis: does a window of opportunity exist? Results on the Leiden Early Arthritis Clinic and ESPOIR cohorts. Annals of the Rheumatic Diseases 2015;74:806–812. Visser H, le Cessie S, Vos K, et al . How to diagnose rheumatoid arthritis early: a prediction model for persistent (erosive) arthritis. Arthritis Rheum. 2002;46(2):357 – 65. doi: 10.1002/art.10117 . PMID: 11840437. Lee SW, Kim SY, Chang SH. Prevalence of feet and ankle arthritis and their impact on clinical indices in patients with rheumatoid arthritis: a cross-sectional study. BMC Musculoskelet Disord 20, 420 (2019). https://doi.org/10.1186/s12891-019-2773-z Hoque A, Steultjens M, Dickson DM, et al . Patients' and clinicians' perspectives on the clinical utility of the Rheumatoid Arthritis Foot Disease Activity Index. Rheumatol Int. 2022;42(10):1807–1817. doi: 10.1007/s00296-022-05147-8 . Epub 2022 May 27. PMID: 35622086; van der Helm-van Mil AHM, Verpoort KN, Breedveld FC, et al . Antibodies to citrullinated proteins and differences in clinical progression of rheumatoid arthritis. Arthritis Res Ther. 2005;7(5):R949-58. doi: 10.1186/ar1767 . PMID: 16207336; Bugatti S, Manzo A, Montecucco C et al. The Clinical Value of Autoantibodies in Rheumatoid Arthritis. Front Med (Lausanne) 2018;5:339. Aletaha D, Neogi T, Silman AJ, et al . 2010 Rheumatoid arthritis classification criteria: an American College of Rheumatology/European League Against Rheumatism collaborative initiative. Arthritis Rheum. 2010;62(9):2569-81. doi: 10.1002/art.27584 . PMID: 20872595. Burgers LE, van Steenbergen HW, Ten Brinck RM, et al . Differences in the symptomatic phase preceding ACPA-positive and ACPA-negative RA: a longitudinal study in arthralgia during progression to clinical arthritis. Ann Rheum Dis. 2017;76(10):1751–1754. doi: 10.1136/annrheumdis-2017-211325 . PMID: 28606964. Takahashi Y, Murai C, Shibata S, et al . Human parvovirus B19 as a causative agent for rheumatoid arthritis. Proc Natl Acad Sci U S A. 1998;95(14):8227–8232. doi: 10.1073/pnas.95.14.8227 Naides SJ, Scharosch LL, Foto F, et al . Rheumatologic manifestations of human parvovirus B19 infection in adults. Initial two-year clinical experience. Arthritis Rheum. 1990;33(9):1297 – 309. doi: 10.1002/art.1780330901 . PMID: 2169746. Heckert SL, Bergstra SA, Matthijssen XME, et al . Joint inflammation tends to recur in the same joints during the rheumatoid arthritis disease course. Ann Rheum Dis. 2022;81(2):169–174. doi: 10.1136/annrheumdis-2021-220882 . PMID: 34462262. Zhang F, Jonsson AH, Nathan A, et al . Cellular deconstruction of inflamed synovium defines diverse inflammatory phenotypes in rheumatoid arthritis. bioRxiv. 2022;2022.02.25.481990. FOOTNOTES Contributors RK and TM developed the study design together with EB and MPM. TM ran the cluster analysis, while SB ran the permutation analysis. BBdK repeated the survival analysis in the replication set. SAB and CA provided the replication data of the IMPROVED trial, while KG and JVvD provided the data from Reumazorg Zuid West Nederland. SA, LAC and SP provided and processed the SYNGem cohort data for the histological analysis. All authors contributed to the interpretation of the results, and provided ideas for further downstream analysis. AHM annotated the classification criteria. RK and TM drafted the first version of the manuscript. All authors reviewed and approved the final draft of the manuscript to be submitted. Additional Declarations No competing interests reported. Supplementary Files 20250318SupplementaryInformation.pdf Cite Share Download PDF Status: Published Journal Publication published 23 Oct, 2025 Read the published version in npj Digital Medicine → Version 1 posted Editorial decision: Revision requested 04 Jun, 2025 Reviews received at journal 29 May, 2025 Reviewers agreed at journal 18 May, 2025 Reviewers agreed at journal 17 May, 2025 Reviewers agreed at journal 16 May, 2025 Reviews received at journal 30 Apr, 2025 Reviews received at journal 07 Apr, 2025 Reviewers agreed at journal 03 Apr, 2025 Reviewers agreed at journal 31 Mar, 2025 Reviewers agreed at journal 29 Mar, 2025 Reviewers invited by journal 21 Mar, 2025 Editor assigned by journal 20 Mar, 2025 Submission checks completed at journal 20 Mar, 2025 First submitted to journal 18 Mar, 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-6256181","acceptedTermsAndConditions":true,"allowDirectSubmit":false,"archivedVersions":[],"articleType":"Article","associatedPublications":[],"authors":[{"id":435668038,"identity":"ac5f2565-a020-44f9-8fef-e2457e67455a","order_by":0,"name":"Tjardo D. Maarseveen","email":"","orcid":"","institution":"Leiden University Medical Center","correspondingAuthor":false,"prefix":"","firstName":"Tjardo","middleName":"D.","lastName":"Maarseveen","suffix":""},{"id":435668041,"identity":"4643b276-85cd-4fa4-a7aa-93d126d7ee6b","order_by":1,"name":"Marc P. Maurits","email":"","orcid":"","institution":"Leiden University Medical Center","correspondingAuthor":false,"prefix":"","firstName":"Marc","middleName":"P.","lastName":"Maurits","suffix":""},{"id":435668044,"identity":"f1a4aa26-2cf0-4ee0-8ee5-9e765de5ab4b","order_by":2,"name":"Lavinia Agra Coletto","email":"","orcid":"","institution":"Fondazione Policlinico Universitario A. Gemelli IRCCS","correspondingAuthor":false,"prefix":"","firstName":"Lavinia","middleName":"Agra","lastName":"Coletto","suffix":""},{"id":435668047,"identity":"af7b6dfe-09b0-4c3e-a116-c0be1f484345","order_by":3,"name":"Simone Perniola","email":"","orcid":"","institution":"Fondazione Policlinico Universitario A. Gemelli IRCCS","correspondingAuthor":false,"prefix":"","firstName":"Simone","middleName":"","lastName":"Perniola","suffix":""},{"id":435668050,"identity":"c1c92864-9d8b-4918-83dc-568fabf3c7c3","order_by":4,"name":"Stefan Böhringer","email":"","orcid":"","institution":"Leiden University Medical Center","correspondingAuthor":false,"prefix":"","firstName":"Stefan","middleName":"","lastName":"Böhringer","suffix":""},{"id":435668051,"identity":"7606580d-ec08-4142-a0ef-b3323ca0b264","order_by":5,"name":"Nils Steinz","email":"","orcid":"","institution":"Leiden University Medical Center","correspondingAuthor":false,"prefix":"","firstName":"Nils","middleName":"","lastName":"Steinz","suffix":""},{"id":435668053,"identity":"037be904-72fe-4574-a71f-f0c758f0a414","order_by":6,"name":"Sytske Anne Bergstra","email":"","orcid":"","institution":"Leiden University Medical Center","correspondingAuthor":false,"prefix":"","firstName":"Sytske","middleName":"Anne","lastName":"Bergstra","suffix":""},{"id":435668056,"identity":"0081589a-2bf2-4e62-8f88-623b1e862fe5","order_by":7,"name":"Dario Bruno","email":"","orcid":"","institution":"Fondazione Policlinico Universitario A. Gemelli IRCCS","correspondingAuthor":false,"prefix":"","firstName":"Dario","middleName":"","lastName":"Bruno","suffix":""},{"id":435668057,"identity":"24c2c649-f4b9-42eb-88c8-5050b5996061","order_by":8,"name":"Maria Rita Gigante","email":"","orcid":"","institution":"Fondazione Policlinico Universitario A. Gemelli IRCCS","correspondingAuthor":false,"prefix":"","firstName":"Maria","middleName":"Rita","lastName":"Gigante","suffix":""},{"id":435668058,"identity":"758ce52b-3c22-416a-a23c-ac59e8713be9","order_by":9,"name":"Viviana A. Pacucci","email":"","orcid":"","institution":"Fondazione Policlinico Universitario A. Gemelli IRCCS","correspondingAuthor":false,"prefix":"","firstName":"Viviana","middleName":"A.","lastName":"Pacucci","suffix":""},{"id":435668063,"identity":"3d5634be-1355-4b1f-9174-08441837e9f8","order_by":10,"name":"Luca Petricca","email":"","orcid":"","institution":"Fondazione Policlinico Universitario A. Gemelli IRCCS","correspondingAuthor":false,"prefix":"","firstName":"Luca","middleName":"","lastName":"Petricca","suffix":""},{"id":435668064,"identity":"382700d7-830d-4085-a793-0f5abe907c41","order_by":11,"name":"Bianca Boxma-de Klerk","email":"","orcid":"","institution":"Leiden University Medical Center","correspondingAuthor":false,"prefix":"","firstName":"Bianca","middleName":"Boxma-de","lastName":"Klerk","suffix":""},{"id":435668065,"identity":"f0b7dd92-e5c7-40df-b318-174881cc8d6c","order_by":12,"name":"Herman Kasper Glas","email":"","orcid":"","institution":"Reumazorg Zuid West Nederland","correspondingAuthor":false,"prefix":"","firstName":"Herman","middleName":"Kasper","lastName":"Glas","suffix":""},{"id":435668071,"identity":"433e245a-1efd-4a24-abb6-7ae689f6c810","order_by":13,"name":"Clara Di Mario","email":"","orcid":"","institution":"Immunology Research Core Facility – Gemelli Science and Technology Park (GSTeP) - Fondazione Policlinico Universitario A. Gemelli IRCCS","correspondingAuthor":false,"prefix":"","firstName":"Clara","middleName":"Di","lastName":"Mario","suffix":""},{"id":435668075,"identity":"a497658f-f709-43a9-845d-d1722bfd7419","order_by":14,"name":"Denise Campobasso","email":"","orcid":"","institution":"Immunology Research Core Facility – Gemelli Science and Technology Park (GSTeP) - Fondazione Policlinico Universitario A. Gemelli IRCCS","correspondingAuthor":false,"prefix":"","firstName":"Denise","middleName":"","lastName":"Campobasso","suffix":""},{"id":435668078,"identity":"c938b1c4-d456-43b8-b1a6-36bcc8ec5fd6","order_by":15,"name":"Barbara Tolusso","email":"","orcid":"","institution":"Immunology Research Core Facility – Gemelli Science and Technology Park (GSTeP) - Fondazione Policlinico Universitario A. Gemelli IRCCS","correspondingAuthor":false,"prefix":"","firstName":"Barbara","middleName":"","lastName":"Tolusso","suffix":""},{"id":435668082,"identity":"41c6a6eb-4549-4148-b553-f1d99401d167","order_by":16,"name":"Josien Veris-van Dieren","email":"","orcid":"","institution":"Reumazorg Zuid West Nederland","correspondingAuthor":false,"prefix":"","firstName":"Josien","middleName":"Veris-van","lastName":"Dieren","suffix":""},{"id":435668083,"identity":"53e36500-e9db-4f56-9576-e78c44f6c3e4","order_by":17,"name":"Annette H. M. Helm-van Mil","email":"","orcid":"","institution":"Leiden University Medical Center","correspondingAuthor":false,"prefix":"","firstName":"Annette","middleName":"H. M. Helm-van","lastName":"Mil","suffix":""},{"id":435668084,"identity":"6bbdfaf1-cfe5-4a1e-9b52-d525d2176fd8","order_by":18,"name":"Elisa Gremese","email":"","orcid":"","institution":"Fondazione Policlinico Universitario A. Gemelli IRCCS","correspondingAuthor":false,"prefix":"","firstName":"Elisa","middleName":"","lastName":"Gremese","suffix":""},{"id":435668086,"identity":"f44f1ba7-bf27-473a-82fe-625770d14ea5","order_by":19,"name":"Maria Antonietta D’Agostino","email":"","orcid":"","institution":"Fondazione Policlinico Universitario A. Gemelli IRCCS","correspondingAuthor":false,"prefix":"","firstName":"Maria","middleName":"Antonietta","lastName":"D’Agostino","suffix":""},{"id":435668090,"identity":"e5c07556-d5ab-44d9-8bca-26274d1a7995","order_by":20,"name":"Marcel J T. Reinders","email":"","orcid":"","institution":"Leiden University Medical Center","correspondingAuthor":false,"prefix":"","firstName":"Marcel","middleName":"J T.","lastName":"Reinders","suffix":""},{"id":435668091,"identity":"a715d3a1-bcd5-492f-9be2-0aae3ccb483f","order_by":21,"name":"Marco Gessi","email":"","orcid":"","institution":"Fondazione Policlinico Universitario A. Gemelli IRCCS","correspondingAuthor":false,"prefix":"","firstName":"Marco","middleName":"","lastName":"Gessi","suffix":""},{"id":435668092,"identity":"9a62ad95-a63f-4c52-8832-af22aff3ca45","order_by":22,"name":"Tom W J Huizinga","email":"","orcid":"","institution":"Leiden University Medical Center","correspondingAuthor":false,"prefix":"","firstName":"Tom","middleName":"W J","lastName":"Huizinga","suffix":""},{"id":435668093,"identity":"335d7168-6a4c-49be-a8a6-a6f40123949f","order_by":23,"name":"Stefano Alivernini","email":"","orcid":"","institution":"Fondazione Policlinico Universitario A. Gemelli IRCCS","correspondingAuthor":false,"prefix":"","firstName":"Stefano","middleName":"","lastName":"Alivernini","suffix":""},{"id":435668094,"identity":"0e9cc23d-ca04-4b2e-80be-e4861b88fcc9","order_by":24,"name":"Erik B. Akker","email":"","orcid":"","institution":"Leiden University Medical Center","correspondingAuthor":false,"prefix":"","firstName":"Erik","middleName":"B.","lastName":"Akker","suffix":""},{"id":435668095,"identity":"97df728c-6d81-42b3-bb9a-2f7849e0b2e7","order_by":25,"name":"Rachel Knevel","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAAqklEQVRIiWNgGAWjYFACNjApw88O4coQrYVHshnC5SFei8FhYrWYN7ClPfjYZsNjfJjHdMMPhjuEtcgcYDtuOLMtjcfsMI/ZzR6GZ4S1SDCwt0nznDkM1nKbgeEwkVr+nPnPY9xMvBa2Y9IMFQd4DJiJ1sLMlibZU5HMI3GYrexmjwExWtjbzCR+GNjJ8bc3b7vxo+KwHEEtDMwoPAPCGkbBKBgFo2AUEAEAfkIt+e4pFBcAAAAASUVORK5CYII=","orcid":"","institution":"Leiden University Medical Center","correspondingAuthor":true,"prefix":"","firstName":"Rachel","middleName":"","lastName":"Knevel","suffix":""}],"badges":[],"createdAt":"2025-03-18 20:53:18","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-6256181/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-6256181/v1","draftVersion":[],"editorialEvents":[{"content":"https://doi.org/10.1038/s41746-025-01997-1","type":"published","date":"2025-10-23T16:16:51+00:00"}],"editorialNote":"","failedWorkflow":false,"files":[{"id":79657723,"identity":"021e415e-fa9d-496d-acca-7d999982f580","added_by":"auto","created_at":"2025-04-01 09:09:25","extension":"png","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":2053263,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cem\u003e\u003cstrong\u003eWorkflow illustrating the study outline. \u003c/strong\u003e\u003c/em\u003e\u003cem\u003eThis study aims to clarify the complexity of rheumatoid arthritis disease through data-driven analysis, providing insights into its etiology and predicting future clinical outcomes. Our cluster analysis identified four distinct subgroups at baseline, which differed in their response to MTX treatment and remission rates within one year. These clusters and their treatment effects were successfully replicated in other centers. Further analysis on a separate cohort revealed that these clusters also corresponded to differences in the synovium.\u003c/em\u003e\u003c/p\u003e","description":"","filename":"fig1Workflow.png","url":"https://assets-eu.researchsquare.com/files/rs-6256181/v1/c2f1daf49d940efbd7ae2d31.png"},{"id":79657714,"identity":"87c70adc-7473-44db-a41a-9e867494b13b","added_by":"auto","created_at":"2025-04-01 09:09:24","extension":"png","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":5519530,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cem\u003e\u003cstrong\u003ePictorial mannequins for replication set B and C and their original counterpart (set A) to show the affected joints for each cluster with color and size to depict prevalence.\u003c/strong\u003e\u003c/em\u003e\u003cem\u003eFrequency is colored on a gradient from red (=100%) to yellow (=0%). When there is no colored dot, it signifies the absence of both swelling and pain at baseline for these patients.\u003c/em\u003e\u003c/p\u003e","description":"","filename":"fig2PictorialMannequinReplication.png","url":"https://assets-eu.researchsquare.com/files/rs-6256181/v1/a11556b4ca9d07e022ee4785.png"},{"id":79657712,"identity":"89a71839-d403-45f8-a7d2-2d6b8f5d7fb0","added_by":"auto","created_at":"2025-04-01 09:09:24","extension":"png","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":3846736,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cem\u003e\u003cstrong\u003eDownstream analysis depicting association of baseline clusters with methotrexate failure (A, B, C) - and remission (D, E) after 1 year.\u003c/strong\u003e\u003c/em\u003e\u003cem\u003e Here we generated Kaplan Meier survival curves for MTX-switch - and remission rates (DAS44\u0026lt;1.6) for survival time data or cross tabs for binary outcomes (i.e. set B had MTX switch integrated in protocol). Follow-up survival data is shown only for MTX-starters in set A (n=1,084; A), set B (n=273, B), set C (n=406; C) or patients with remission information in set A (n=676; D) and set B (n=295; E). Global trend was inferred with the log rank test for survival curves or chi-squared test for the cross tabs.\u003c/em\u003e\u003c/p\u003e","description":"","filename":"fig3KaplanMeier.png","url":"https://assets-eu.researchsquare.com/files/rs-6256181/v1/87169d8875cd9d13c34d6786.png"},{"id":79657710,"identity":"b2646965-4c1a-4c52-b2ec-d3788c66b655","added_by":"auto","created_at":"2025-04-01 09:09:23","extension":"png","order_by":4,"title":"Figure 4","display":"","copyAsset":false,"role":"figure","size":1372701,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cem\u003e\u003cstrong\u003eKaplan Meier depicting the A) ACPA effect on MTX switch rates in general and stratified by cluster, B) local cluster differences between JIP-hand and foot clusters (JIP-foot \u0026amp; JIP-poly) within the ACPA positive stratum.\u003c/strong\u003e\u003c/em\u003e\u003cem\u003e Here the dashed lines indicate the seropositive patients. Global trend was inferred with the log rank test for survival curves or chi-squared test for the cross tabs.\u003c/em\u003e\u003c/p\u003e","description":"","filename":"fig4ACPAstratification.png","url":"https://assets-eu.researchsquare.com/files/rs-6256181/v1/d46008bd8cc193226a52ad30.png"},{"id":79658649,"identity":"60a7b997-e739-4394-a6b7-dd92d691af26","added_by":"auto","created_at":"2025-04-01 09:17:28","extension":"png","order_by":5,"title":"Figure 5","display":"","copyAsset":false,"role":"figure","size":905310,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eS\u003c/strong\u003e\u003cem\u003e\u003cstrong\u003eummary of synovitis across various clusters, presented as: a) the total Krenn synovitis score for each cluster, and the individual components: b) lining layer hyperplasia, c) stromal density, and d) inflammatory infiltrate. \u003c/strong\u003e\u003c/em\u003e\u003cem\u003eGlobal trends were assessed using the Kruskal-Wallis test for the Krenn synovitis score and an ordered logit model for each subitem grade. These analyses were followed by a post hoc Wald test, with significance levels indicated as *p \u0026lt; 0.05, **p \u0026lt; 0.01, and ***p \u0026lt; 0.001.\u003c/em\u003e\u003c/p\u003e","description":"","filename":"fig5SynovialBiopsies.png","url":"https://assets-eu.researchsquare.com/files/rs-6256181/v1/d2e776eaafaad19de46fb826.png"},{"id":94490728,"identity":"b7295992-e9ce-43f1-b362-008cfe4d94d4","added_by":"auto","created_at":"2025-10-27 17:14:19","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":15037462,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-6256181/v1/388ed8c7-315d-48a6-8597-ef7e111cbe35.pdf"},{"id":79657730,"identity":"727331f9-9a4b-44d9-a4a1-91ade5ea716f","added_by":"auto","created_at":"2025-04-01 09:09:26","extension":"pdf","order_by":2,"title":"","display":"","copyAsset":false,"role":"supplement","size":5545308,"visible":true,"origin":"","legend":"","description":"","filename":"20250318SupplementaryInformation.pdf","url":"https://assets-eu.researchsquare.com/files/rs-6256181/v1/d80b815571d639d9b62d63b2.pdf"}],"financialInterests":"No competing interests reported.","formattedTitle":"Location and amount of joint involvement differentiates rheumatoid arthritis into different clinical subsets","fulltext":[{"header":"INTRODUCTION","content":"\u003cp\u003eRheumatoid arthritis (RA) is a heterogeneous disease. The current classification criteria for RA were developed to approximate the decision to start early treatment and the exclusion of other diseases. At clinical presentation, patients vary in the number and pattern of joints involved, presence of extra articular manifestations and abnormalities in blood and synovial tissue [\u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e, \u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2\u003c/span\u003e, \u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e]. The heterogeneity of RA also manifests in clinical outcomes, namely prognosis, treatment response and comorbidities. This evident diversity likely impacts the interpretation of treatment effect and etiologic factors such as genetics and downplay their importance altogether [\u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e4\u003c/span\u003e]. If phenotypic subsetting into more homogeneous groups is possible, it could improve research into the etiology of RA and enhance its treatment.\u003c/p\u003e \u003cp\u003eFor centuries, pattern recognition on clinical variables by doctors has been the driving force of disease identification and examination of the underlying etiologic mechanisms. Thus far clinicians have not identified the relevant (sub)patterns in RA. The presence of ACPA [\u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e5\u003c/span\u003e, \u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e6\u003c/span\u003e, \u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e7\u003c/span\u003e, \u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e8\u003c/span\u003e] and the age of onset [\u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e9\u003c/span\u003e, \u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e10\u003c/span\u003e, \u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e11\u003c/span\u003e] have been raised as possible dichotomous disease subsetting features. However, neither of these markers in isolation adequately addresses the heterogeneity and complexity of the disease. This suggests there are other factors involved.\u003c/p\u003e \u003cp\u003eCluster analysis combining a high number of factors has demonstrated its effectiveness in categorizing complex diseases (such as diabetes type II, asthma, osteoarthritis) into subtypes that differ in clinical outcomes or biological background [\u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e12\u003c/span\u003e, \u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e13\u003c/span\u003e, \u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e14\u003c/span\u003e]. In the context of RA, there is quite some focus on molecular phenotyping such as done by Lewis et al [\u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e15\u003c/span\u003e], who discovered patterns in synovial tissue at baseline, with the lymphoid-myeloid pathotype being a predictor for a poor outcome at disease onset [\u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e16\u003c/span\u003e]. Others used clinical and comorbidity information for clustering and identified four subsets, including one that exhibited a higher likelihood for biological DMARD initiation [\u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e17\u003c/span\u003e]. Likewise, Curtis et al [\u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e18\u003c/span\u003e] used clinical variables, though not exclusively at baseline, and identified five clusters that differed in disease activity, RA-duration and type of comorbidities. These outcomes are typically highly influenced by treatment decisions and events that occur independent of the specific RA type. Furthermore, detailed clinical information such as the pattern of involved joints may be relevant for disease differentiation as exemplified by psoriatic arthritis (PsA) [\u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e19\u003c/span\u003e], yet none of the previous studies capitalize on this information for clustering.\u003c/p\u003e \u003cp\u003eElectronic Health Records (EHR) data provides a powerful asset for clustering as it encompasses a wide variety of data modalities (laboratory values, clinical examination, demographics) that each offer a unique perspective on the patient\u0026rsquo;s condition. The EHRs are collected as routine clinical care, and thus resemble the true patient population more closely than a study population collected with a particular hypothesis in mind. The diversity of data types does however pose a methodological challenge due to structural differences between the data modalities. The recent surge of deep learning tools [\u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e20\u003c/span\u003e], offers the possibility to combine different EHR-layers into a patient representation by extracting the (hidden) factors that capture most variation in the data. At present, there exist many machine learning (ML) techniques to learn the relevant (clinical) patterns, and encode patients accordingly. These embeddings can be used to detect patients' subgroups, identify patterns, build predictive models or assist in making disease classifications. The literature reports that clustering on top of these embeddings typically outperforms conventional techniques in the case of high dimensional or complicated data [\u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e21\u003c/span\u003e, \u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e22\u003c/span\u003e, \u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e23\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eIn this study we aimed to dissect the clinical heterogeneity of RA by using the symptoms at initial presentation, so before external factors such as treatment interfere. We hypothesize that the location of the involved joints and the inflammatory patterns observed in the blood play a role in subsetting RA, similar to their significance in distinguishing PsA from RA [\u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e19\u003c/span\u003e]. To achieve this we make use of advanced data-driven techniques to identify and analyze disease differentiating signatures based on initial clinical variables, and see if they relate to clinical outcomes and histological synovial features.\u003c/p\u003e"},{"header":"METHODS","content":"\u003cdiv id=\"Sec3\" class=\"Section2\"\u003e \u003ch2\u003ePatients\u003c/h2\u003e \u003cp\u003eOur study comprises three different phases: a i) developmental phase where we identify and validate subtypes in a discovery set according to long term outcomes (set A), a ii) replication phase where we cluster novel patients using historic trial- (set B) and external hospital data (set C) to infer generalizability by replicating the treatment analysis and finally iii) a downstream analysis in external hospital data (set D) where we explore differences between clusters in their synovial tissue.\u003c/p\u003e \u003cp\u003eSet A consisted of 1,387 RA-patients that visited the rheumatology outpatient clinic of the Leiden University Medical Centre (LUMC) for the first time between August 29th, 2011 till December 1st 2022. RA diagnosis was based on the physician\u0026rsquo;s diagnosis within 1 year since first visit.[\u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e24\u003c/span\u003e, \u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e25\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eSet B concerned 307 RA-patients from the IMPROVED trial that were recruited between March 2007 till September 2010 [\u003cspan citationid=\"CR26\" class=\"CitationRef\"\u003e26\u003c/span\u003e]. This trial recruited undifferentiated arthritis and early RA with less than 2 years of symptoms. We selected only those patients who met the ACR2010 criteria within one year after inclusion. All patients received MTX at baseline and were randomized into two arms of treatment intensification if they did not reach remission after 4 months.\u003c/p\u003e \u003cp\u003eSet C included 515 RA patients from Reumazorg Zuid West Nederland (RZWN), collected between January 2015 and December 1, 2022. These patients were from nine different hospitals across the south west of the Netherlands, with the largest groups coming from Goes (n\u0026thinsp;=\u0026thinsp;157), Roosendaal (n\u0026thinsp;=\u0026thinsp;153), and Vlissingen (n\u0026thinsp;=\u0026thinsp;49). Herein, the diagnosis of RA was defined as having an ICD-code for RA and starting with a conventional DMARD.\u003c/p\u003e \u003cp\u003e Set D included 262 RA-patients fulfilling the ACR2010 criteria for RA from the SYNGem Biopsy Unit cohort of the Fondazione Policlinico Universitario A. Gemelli IRCCS\u0026ndash;Universit\u0026agrave; Cattolica del Sacro Cuore from Rome, Italy, all of whom underwent minimally invasive ultrasound-guided synovial tissue biopsy at their first rheumatological evaluation within their clinical routine management. Each tissue was processed for H\u0026amp;E staining and synovitis was graded using the total Krenn synovitis score (KSS) [\u003cspan citationid=\"CR27\" class=\"CitationRef\"\u003e27\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eAcross all sets, a minimum follow-up of 1-year was required to ascertain the diagnosis of RA. Prior to conducting the study, we acquired approval from the ethics committee of the LUMC. Patients and public were not involved during the development, execution, and dissemination of the study.\u003c/p\u003e \u003c/div\u003e\n\u003ch3\u003ePreprocessing of electronic health records\u003c/h3\u003e\n\u003cp\u003eTo construct patient phenotypic profiles, we extracted information on serology (RF and ACPA), location of joint involvement (tender- and swollen joints (TJC and SJC)), demographics, blood profiles (hemoglobin, hematocrit, leukocyte- and thrombocytes levels) and ESR at baseline (Table \u003cspan refid=\"MOESM1\" class=\"InternalRef\"\u003eS1\u003c/span\u003e). Baseline was defined as the first visit to the clinic (set A\u0026amp;C) or the moment of inclusion in the trial (set B). Patients with missing lab or joint location variables were dropped (Fig. \u003cspan refid=\"MOESM1\" class=\"InternalRef\"\u003eS1\u003c/span\u003e).\u003c/p\u003e \u003cp\u003eWe normalized the numerical data using a Yeo-Johnson transformation, except for the ESR levels where we applied a log transformation due to their log-normal distribution. For the categorical data, we implemented one-hot encoding, which created separate binary fields (yes/no) for each possible category value.\u003c/p\u003e\n\u003ch3\u003eConstruction of patient embedding\u003c/h3\u003e\n\u003cp\u003eWe integrated the different EHR data types to create a condensed patient representation (called a patient embedding) using a multi-modal autoencoder (MMAE). This MMAE had a narrowing structure with encoder layers of 128, 64, and finally 8 neurons. For categorical data, we used Sigmoid activation functions with Bernoulli loss, while numerical data utilized ReLU activation with Gaussian loss. To prevent overfitting, we compared performance between our training set (80% of data) and validation set (20%), using the Adam optimizer throughout. We visualized how much each type of EHR data contributed to our final embedding using a flameplot analysis (Fig. \u003cspan refid=\"MOESM2\" class=\"InternalRef\"\u003eS2\u003c/span\u003e(0))\u003c/p\u003e \u003cp\u003eAfter creating the patient embedding, we grouped patients into subcommunities based on clinical similarities using PhenoGraph [\u003cspan citationid=\"CR28\" class=\"CitationRef\"\u003e28\u003c/span\u003e]. We selected PhenoGraph rather than more common methods like K-means clustering because it performs better with our sparse, high-dimensional medical data (see our supplemental materials for additional technical details).\u003c/p\u003e\n\u003ch3\u003eCluster interpretation\u003c/h3\u003e\n\u003cp\u003eFor each cluster, we examined the characteristics and visualized the phenotype on a pictorial mannequin with an integrated heatmap to demonstrate the number of involved joints. We used a surrogate ML-technique to model the cluster assignment and subjected this model to a SHAP (SHapley Additive exPlanations) [\u003cspan citationid=\"CR29\" class=\"CitationRef\"\u003e29\u003c/span\u003e] analysis to retrospectively identify the most important variables per cluster. The SHAP plots show the strength and direction of impact of that variable for each patient (also those who are not assigned to that cluster).\u003c/p\u003e\n\u003ch3\u003eCluster validation\u003c/h3\u003e\n\u003cp\u003eTo confirm that our identified clusters comprised a stable and relevant partitioning, we performed a number of validation checks. We examined cluster stability by measuring how often patients co-cluster across 1000 random subsets of the data and assessed possible factors influencing the partitioning. Next, we conducted a Local Inverse Simpson's Index analysis [\u003cspan citationid=\"CR30\" class=\"CitationRef\"\u003e30\u003c/span\u003e] to assess whether physicians are evenly distributed across the clusters or if the patient subgroups merely reflect the reporting differences between physicians (see supplemental material).\u003c/p\u003e \u003cp\u003eTo infer the clinical relevance, clinical outcomes were evaluated using a Cox regression model, including: time to MTX-failure (defined by replacement of- or adding an additional DMARD to MTX) and remission (DAS44\u0026thinsp;\u0026lt;\u0026thinsp;1.6) within one year. Moreover, we evaluated the replicability on an external dataset (set B \u0026amp; set C), where individuals are assigned to clusters in accordance with the previously learned patient embedding (see supplemental material).\u003c/p\u003e \u003cdiv id=\"Sec8\" class=\"Section2\"\u003e \u003ch2\u003eDiversity in synovial characteristics\u003c/h2\u003e \u003cp\u003eIn addition to the clinical validation, we compared the characteristics of the synovium across the four subgroups, by repeating the clustering analysis in set D (SYNGem cohort). Specifically, we compared the overall Krenn synovitis score (KSS) [\u003cspan citationid=\"CR27\" class=\"CitationRef\"\u003e27\u003c/span\u003e] and its individual components (lining layer hyperplasia, stromal density and inflammatory infiltrate). Each KSS component was measured on an ordinal scale (0: none, 1: low, 2: moderate, 3: severe). To analyze this, we used ordinal logistic regression, which estimates the odds of transitioning from one severity level to the next, thus relying on sufficient representation. However, since our study focused on treatment-naive RA patients with active synovitis, the lowest level (0: none) for each KSS subitem was rarely observed. As a result, when present, it was combined with the 'low' category (1: low).\u003c/p\u003e \u003c/div\u003e\n\u003ch3\u003eStatistical tests\u003c/h3\u003e\n\u003cp\u003eWe used the Kruskal-Wallis test followed by Dunn's post-hoc test to compare numerical values across multiple groups. For survival analysis we used the log-rank test to examine the overall trend and a univariate Cox-regression [\u003cspan citationid=\"CR31\" class=\"CitationRef\"\u003e31\u003c/span\u003e] to quantify the cluster differences. We inferred the DAS-remission status during the survival analysis, carrying the last observation forward if it was missing (effectively the same as time to event). The proportional hazards assumption was verified by examining the Schoenfeld residuals [\u003cspan citationid=\"CR32\" class=\"CitationRef\"\u003e32\u003c/span\u003e]. We also adjusted the Cox regression model for MTX-response on covariates suspected to influence treatment response (e.g. ACPA positivity, RF positivity, SJC, TJC, Sex and Age). The statistical significance was inferred with ANOVA. For the ordinal values of the KSS components we used an ordered logit, followed up by a post-hoc Wald test to look at individual differences.\u003c/p\u003e\n\u003ch3\u003eWeb Interface\u003c/h3\u003e\n\u003cp\u003eAll of our scripts are publicly available online at Github [\u003cspan citationid=\"CR33\" class=\"CitationRef\"\u003e33\u003c/span\u003e]. Moreover, we developed an interactive web tool (\u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://knevel-lab.github.io/Rheumalyze/\u003c/span\u003e\u003cspan address=\"https://knevel-lab.github.io/Rheumalyze/\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003cspan type=\"Underline\" class=\"Underline\" name=\"Emphasis\"\u003e)\u003c/span\u003e that enables users to map their cohort onto our patient embedding, allowing them to identify the JIP phenotypes present in their data.\u003c/p\u003e"},{"header":"RESULTS","content":"\u003cdiv id=\"Sec12\" class=\"Section2\"\u003e \u003ch2\u003ePatients\u003c/h2\u003e \u003cp\u003eWe retrieved 2,691 RA patients for training set A of whom, 1,387 were included in our study based on the availability of lab values and joint counts. For the replication, set B and set C had 364 and 1,227 RA cases, of whom 307 and 515 had complete information (Table \u003cspan refid=\"MOESM2\" class=\"InternalRef\"\u003eS2\u003c/span\u003e, Fig. \u003cspan refid=\"MOESM1\" class=\"InternalRef\"\u003eS1\u003c/span\u003e). For the downstream analysis, we looked at biopsy data from 264 patients of set D, of whom 194 patients had synovial tissue taken from the same joint (Table S3). A workflow diagram for the different phases of the study is shown in Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003e.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003eEach dataset captured a typical early RA population [\u003cspan citationid=\"CR32\" class=\"CitationRef\"\u003e32\u003c/span\u003e, \u003cspan citationid=\"CR33\" class=\"CitationRef\"\u003e33\u003c/span\u003e]. In comparison to set A, patients in the replication sets exhibited higher rates of seropositivity and had fewer tender joints. On average, set B patients were younger and set C patients had less inflammation, with a median ESR of 16. We used the phenotypic variables (see methods) of set A to construct the patient embedding.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec13\" class=\"Section2\"\u003e \u003ch2\u003eFour clusters separated by joint location, serology and blood values\u003c/h2\u003e \u003cp\u003eThe patient embedding showed four different clusters (Fig. S3, S4) which were not determined by any single clinical variable, as indicated by the wide dispersion of values. The primary variation across the clusters was their difference in affected joints, leading us to name them Joint Involvement Patterns (JIP). Additionally, the clusters differed in levels of inflammation, age, and seropositivity (Table S4, Fig. S5).:\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec14\" class=\"Section2\"\u003e \u003ch2\u003e● Cluster 1: JIP-foot\u003c/h2\u003e \u003cp\u003e \u003cdiv class=\"BlockQuote\"\u003e \u003cp\u003emoderate number of involved joints, particularly feet joints, younger patients, low leukocyte and thrombocyte levels.\u003c/p\u003e \u003c/div\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec15\" class=\"Section2\"\u003e \u003ch2\u003e● Cluster 2: JIP-oligo\u003c/h2\u003e \u003cp\u003e \u003cdiv class=\"BlockQuote\"\u003e \u003cp\u003elimited joint involvement and mostly seropositive patients.\u003c/p\u003e \u003c/div\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec16\" class=\"Section2\"\u003e \u003ch2\u003e● Cluster 3 JIP-hand\u003c/h2\u003e \u003cp\u003e \u003cdiv class=\"BlockQuote\"\u003e \u003cp\u003eelderly patients, symmetrical polyarthritis of hands, seronegative.\u003c/p\u003e \u003c/div\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec17\" class=\"Section2\"\u003e \u003ch2\u003e● Cluster 4 JIP-polyarthritis\u003c/h2\u003e \u003cp\u003e \u003cdiv class=\"BlockQuote\"\u003e \u003cp\u003emajority seronegative polyarthritis in hand and feet though with lower ESR.\u003c/p\u003e \u003c/div\u003e \u003c/p\u003e \u003cp\u003eThe clusters were stable with an average\u0026thinsp;\u0026gt;\u0026thinsp;80% of patients grouping together in the same cluster over the 1000 iterations in the stability analysis (Fig. S6\u0026amp;S7). In fact, the stability was better in our combined multi-modal approach than if we take each data type (numeric/categorical) separately (Fig. S8). The clusters were not driven by treating physicians (Fig. S9) and were generalizable across different validation sets (Table\u0026nbsp;\u003cspan refid=\"Tab1\" class=\"InternalRef\"\u003e1\u003c/span\u003e, Fig. S10\u0026amp;S11), showing similar joint involvement patterns (Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003e). Also, the identified patient clusters did not seem to represent different disease stages as the cluster with the longest symptom duration had the lowest joint count and vice versa. There were differences in cluster prevalences between the validation sets (Table S5\u0026amp;S6).\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\u003eBaseline characteristics of the different patient clusters (set A\u0026thinsp;+\u0026thinsp;replication sets B \u0026amp; C)\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\u0026nbsp;\u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eJIP-Foot\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003eJIP-Oligo\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003eJIP-Hand\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c5\"\u003e \u003cp\u003eJIP-Poly\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\u003eN\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e596\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e761\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e450\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e402\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eSex, female\u003c/b\u003e \u003csup\u003e\u003cb\u003eɣ\u003c/b\u003e\u003c/sup\u003e \u003cb\u003e[n(%)]\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e389 (65.3)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e505 (66.4)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e272 (60.4)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e272 (67.7)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eAge\u003c/b\u003e \u003csup\u003e\u003cb\u003eɣ\u003c/b\u003e\u003c/sup\u003e \u003cb\u003e(SD, yr)\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e56.6 (14.6)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e59.5 (14.6)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e66.4 (13.2)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e54.7 (14.7)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eRF\u003c/b\u003e \u003csup\u003e\u003cb\u003eɣ\u003c/b\u003e\u003c/sup\u003e \u003cb\u003e[n(%)]\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e370 (62.1)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e497 (65.3)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e200 (44.4)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e196 (48.8)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eACPA\u003c/b\u003e \u003csup\u003e\u003cb\u003eɣ\u003c/b\u003e\u003c/sup\u003e \u003cb\u003e[n(%)]\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e355 (59.6)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e449 (59.0)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e162 (36.0)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e183 (45.5)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eESR\u003c/b\u003e \u003csup\u003e\u003cb\u003eɣ\u003c/b\u003e\u003c/sup\u003e \u003cb\u003e(IQR, mm/hr)\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e22 (9\u0026ndash;36)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e24 (11\u0026ndash;38)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e28 (13\u0026ndash;48)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e22 (9\u0026ndash;40)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eDAS44(3) (IQR)\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e3.5 (3.0\u0026ndash;4.0)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e2.4 (1.9\u0026ndash;2.8)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e3.7 (3.2\u0026ndash;4.3)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e4.6 (4.0-5.5)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eSJC (IQR)\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e8 (5\u0026ndash;12)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e2 (1\u0026ndash;5)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e10 (7\u0026ndash;15)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e15 (9\u0026ndash;22)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eTJC (IQR)\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e11 (8\u0026ndash;15)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e3 (2\u0026ndash;5)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e11 (7\u0026ndash;15)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e22 (15\u0026ndash;30)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eDAS28(3) (IQR)\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e5.2 (4.5-6.0)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e4.0 (3.3\u0026ndash;4.7)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e5.5 (4.9\u0026ndash;6.2)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e6.5 (5.5\u0026ndash;7.2)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eFollow up (IQR, days)\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e1307 (733\u0026ndash;2020)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e1428 (760\u0026ndash;2082)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e1127 (566\u0026ndash;1875)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e1512 (1012\u0026ndash;2246)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eSymptom duration * (IQR, days)\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e143 (56\u0026ndash;364)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e186 (70\u0026ndash;399)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e101 (48\u0026ndash;279)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e147 (56\u0026ndash;357)\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\u003e \u003cem\u003eWhere SD, standard deviation; RF, rheumatoid factor; ACPA, anti-cyclic citrullinated peptide antibodies; ESR, erythrocyte sedimentation rate; IQR, interquartile range; DAS, three component disease activity score (either 44 or 28 joint scheme); SJC, swollen joint count; TJC, tender joint count; ɣ, clinical variables that were used for clustering (set A); *, symptom duration was only calculated for patients from set A and B.\u003c/em\u003e \u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec18\" class=\"Section2\"\u003e \u003ch2\u003eValidation on clinical outcomes beyond baseline\u003c/h2\u003e \u003cp\u003eIn set A, 80% of patients received MTX as an initial drug across all clusters. The Kaplan Meier curves show a difference in MTX failure between the clusters: 27%, 23%, 16%, 30% (for cluster 1\u0026ndash;4, \u003cem\u003eP\u003c/em\u003e\u0026thinsp;\u0026lt;\u0026thinsp;0.001, Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003ea). The JIP-hand had clearly the best prognosis, where patients were twice as likely to stay on MTX than the most severe disease subtype JIP-poly (HR 0.48 (95% CI 0.35\u0026ndash;0.77), \u003cem\u003eP\u003c/em\u003e\u0026thinsp;\u0026lt;\u0026thinsp;0.001). Additionally, the JIP-hand did better than the JIP-foot (HR:0.55 (0.37\u0026ndash;0.82) \u003cem\u003eP\u003c/em\u003e\u0026thinsp;=\u0026thinsp;0.003).\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003eConsistent with MTX response, we observed differences in remission rates: 44.3%, 47.4%, 55.7%, 38.5% (for cluster 1\u0026ndash;4, \u003cem\u003eP\u003c/em\u003e\u0026thinsp;=\u0026thinsp;0.007, Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003eb) with the biggest difference between the JIP-hand and JIP-poly (HR 1.65 (95% CI 1.2\u0026ndash;2.29), \u003cem\u003eP\u003c/em\u003e\u0026thinsp;=\u0026thinsp;0.002), also when corrected for baseline disease activity (Fig. S12).\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec19\" class=\"Section2\"\u003e \u003ch2\u003eACPA within the clusters\u003c/h2\u003e \u003cp\u003eSince the literature reports that ACPA is indicative of persistent disease [\u003cspan citationid=\"CR36\" class=\"CitationRef\"\u003e36\u003c/span\u003e], we examined whether the ACPA status was the main factor driving the difference in MTX failure. In set A we found a higher treatment failure in ACPA positive than negative patients (27.6% versus 22.0%, Fig.\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003e), though it was not significant (\u003cem\u003eP\u003c/em\u003e\u0026thinsp;=\u0026thinsp;0.057). Moreover, the association of ACPA with MTX-failure differed within the clusters (\u003cem\u003eP\u003c/em\u003e\u0026thinsp;\u0026lt;\u0026thinsp;0.001, Fig.\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003e).\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003eThe difference between the JIP-hand with the foot clusters JIP-poly and JIP-foot was larger within the ACPA-positive stratum (JIP-hand vs JIP-foot (HR:0.37 (0.15\u0026ndash;0.60) \u003cem\u003eP\u003c/em\u003e\u0026thinsp;\u0026lt;\u0026thinsp;0.001), JIP-hand vs JIP-poly HR:0.33 (0.15\u0026ndash;0.72) \u003cem\u003eP\u003c/em\u003e\u0026thinsp;=\u0026thinsp;0.005) (Fig.\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003eb) For remission we could not find this difference in the ACPA-positive stratum.\u003c/p\u003e \u003cp\u003eThe good response in JIP-hand raised the question whether this group overrepresented patients with parvovirus induced arthritis instead of RA, but none of our clusters were enriched for parvovirus positive patients (Fig. S13).\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec20\" class=\"Section2\"\u003e \u003ch2\u003eReplication\u003c/h2\u003e \u003cp\u003eIn set C, 79% of patients received MTX as the initial drug across all clusters, whereas in set B all patients were administered MTX. In both replication sets we again observed a better outcome of JIP-hand for MTX failure (global \u003cem\u003eP\u003c/em\u003e\u0026thinsp;\u0026lt;\u0026thinsp;0.001, and \u003cem\u003eP\u003c/em\u003e\u0026thinsp;\u0026lt;\u0026thinsp;0.001). Remission could only be tested in replication set B where it confirmed our previous finding (\u003cem\u003eP\u003c/em\u003e\u0026thinsp;\u0026lt;\u0026thinsp;0.001). Consistent with the original finding, the difference between JIP-hand and JIP-poly was particularly strong in the ACPA positive stratum in both replication set B (OR: 0.22 (0.12\u0026ndash;0.63), \u003cem\u003eP\u003c/em\u003e\u0026thinsp;=\u0026thinsp;0.017) and set C (HR:0.38 (0.22\u0026ndash;0.68), \u003cem\u003eP\u003c/em\u003e\u0026thinsp;\u0026lt;\u0026thinsp;0.001). Within the ACPA positive stratum we also found the significant difference between JIP-hand and JIP-foot back in replication set C (HR:0.37 (0.15\u0026ndash;0.93), \u003cem\u003eP\u003c/em\u003e\u0026thinsp;=\u0026thinsp;0.034), though not in replication set B (OR: 1.03 (0.34\u0026ndash;3.05) \u003cem\u003eP\u003c/em\u003e\u0026thinsp;=\u0026thinsp;0.801). All the analyses remained significant when corrected for baseline DAS (Fig. S14). Even after adjusting for symptom duration in sets A and B, the association between the cluster and treatment persisted (Fig. S15)\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec21\" class=\"Section2\"\u003e \u003ch2\u003eInformative value of clusters beyond known risk factors for MTX failure\u003c/h2\u003e \u003cp\u003eTo ascertain that the cluster association with treatment outcome was not merely the product of already established clinical markers we adjusted the Cox regression model (Fig. S16). Here, the inclusion of other well-known contributing factors like ACPA, RF or the number of affected joints did not diminish the additive value of clustering (\u003cem\u003eP\u003c/em\u003e\u0026thinsp;=\u0026thinsp;0.020 in set A, and \u003cem\u003eP\u0026thinsp;=\u003c/em\u003e\u0026thinsp;0.019 in set C).\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec22\" class=\"Section2\"\u003e \u003ch2\u003eDiversity in synovial characteristics\u003c/h2\u003e \u003cp\u003eNext, we examined histological variations between patient clusters in Set D according to the Krenn synovitis components [\u003cspan citationid=\"CR27\" class=\"CitationRef\"\u003e27\u003c/span\u003e]. Biopsies were taken from the most inflamed joint, typically the knee or wrist. Due to the limited number of samples from other joints, a thorough comparison across different anatomical locations was not feasible. Therefore, we primarily focused on a single, consistent location that was involved in most clusters \u0026mdash;specifically, the knee joints.\u003c/p\u003e \u003cp\u003eIn total, we analyzed synovial tissue from 194 patients, distributed across four clusters: 27 in JIP-foot, 49 in JIP-oligo, 86 in JIP-hand, and 32 in JIP-poly (see Fig. S17). The Krenn synovitis score [\u003cspan citationid=\"CR27\" class=\"CitationRef\"\u003e27\u003c/span\u003e] significantly differed between the patients from the four different clusters with a mean score of 5,4, 5 and 6 respectively (\u003cem\u003eP\u0026thinsp;=\u0026thinsp;0.004)\u003c/em\u003e with the highest in the JIP-Poly and the lowest in JIP-Oligo clusters respectively \u003cem\u003e(\u003c/em\u003eFig.\u0026nbsp;\u003cspan refid=\"Fig5\" class=\"InternalRef\"\u003e5\u003c/span\u003e\u003cem\u003e)\u003c/em\u003e.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003eThe clusters showed significant histopathological variations between the JIP-variants in lining layer hyperplasia (\u003cem\u003eP\u003c/em\u003e\u0026thinsp;=\u0026thinsp;0.026), stromal density (\u003cem\u003eP\u003c/em\u003e\u0026thinsp;=\u0026thinsp;0.045), and inflammatory infiltrate (P\u0026thinsp;=\u0026thinsp;0.002). These were primarily driven by the distinction between the more stereotypical RA-clusters (JIP-hand/JIP-poly) and JIP-oligo, which was characterized by low-grade synovitis. In JIP-poly 43.8% had severe synovial lining hyperplasia compared to 14.3% in JIP-oligo (\u003cem\u003eP\u003c/em\u003e\u0026thinsp;=\u0026thinsp;0.002). Moreover, 21.9% had severe inflammatory cell infiltration compared to 6.1% in JIP-oligo (\u003cem\u003eP\u003c/em\u003e\u0026thinsp;=\u0026thinsp;0.001) and 18.8% in stromal density versus 8.2% for JIP oligo (\u003cem\u003eP\u003c/em\u003e\u0026thinsp;=\u0026thinsp;0.034). JIP-hand exhibited increased synovitis levels compared to JIP-oligo for lining layer hyperplasia (27.9% vs 14.3%; \u003cem\u003eP\u003c/em\u003e\u0026thinsp;=\u0026thinsp;0.015) and inflammatory infiltrate (17.4% vs 6.1%; \u003cem\u003eP\u003c/em\u003e\u0026thinsp;=\u0026thinsp;0.013). Furthermore, 29.1% had severe stromal density compared to 7.4% in JIP-foot (\u003cem\u003eP\u003c/em\u003e\u0026thinsp;=\u0026thinsp;0.211) and 8.2% in JIP-oligo (\u003cem\u003eP\u003c/em\u003e\u0026thinsp;=\u0026thinsp;0.013). Though, JIP-foot had a similar overall synovitis as JIP-hand, it was not characterized by any particular KSS component, rather it had moderate inflammatory activity across all elements.\u003c/p\u003e \u003cp\u003eSince KSS was previously shown to be linked to disease activity [\u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e], we adjusted for DAS categories to ensure the cluster difference was not merely an effect of the clinical metric. After correction, differences in lining layer hyperplasia (P\u0026thinsp;=\u0026thinsp;0.035) and inflammatory infiltrate (P\u0026thinsp;=\u0026thinsp;0.005) persisted between clusters, while stromal density differences became non-significant (P\u0026thinsp;=\u0026thinsp;0.082).\u003c/p\u003e \u003cp\u003eFor a complete overview, we also examined whether sampling from joint areas specific to each JIP phenotype would yield different results compared to using only knee biopsies. We collected samples from lower extremities for JIP-foot patients, knees for JIP-oligo patients, fingers or wrists for JIP-hand patients, and knees or wrists for JIP-poly patients. This targeted sampling approach produced results similar to those we found when analyzing knee biopsies alone (Fig. S18).\u003c/p\u003e \u003c/div\u003e"},{"header":"DISCUSSION","content":"\u003cp\u003eThrough deep learning and clustering analysis of real-world clinical data, we identified four distinct RA phenotypes at baseline: foot-dominant (JIP-foot), seropositive oligoarticular (JIP-oligo), seronegative hand-dominant (JIP-hand), and polyarticular (JIP-poly) disease. While our hypothesis-free approach enabled detection of novel non-linear clinical signatures, we mitigated the risk of spurious correlations through rigorous validation, including stability testing, clinical outcome validation, independent cohort replication, and synovial histological correlation.\u003c/p\u003e \u003cp\u003eA key finding was the marked difference in treatment success between hand and foot clusters. Both foot-dominant clusters (JIP-foot, JIP-poly) showed higher MTX failure and lower remission rates compared to the JIP-hand, independent of baseline joint involvement, symptom duration, or treatment timing. This validates previous cross-sectional observations of poor prognosis in feet/ankle-involved disease [\u003cspan citationid=\"CR37\" class=\"CitationRef\"\u003e37\u003c/span\u003e], though our study uniquely demonstrates this association in treatment-na\u0026iuml;ve patients at presentation. The impact of foot involvement on treatment failure rivaled that of ACPA-positivity as an independent risk factor. This observation holds particular clinical relevance given the frequent exclusion of lower extremity joints from disease activity scores, despite their common involvement [\u003cspan citationid=\"CR37\" class=\"CitationRef\"\u003e37\u003c/span\u003e, \u003cspan citationid=\"CR38\" class=\"CitationRef\"\u003e38\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eContrary to common assumptions, we did not observe a clear ACPA dichotomy. This finding aligns with previous baseline studies which demonstrate that despite known differences in risk factors and prognosis between ACPA-positive and ACPA-negative patients, the clinical phenotype at initial diagnosis is similar for both groups [\u003cspan citationid=\"CR36\" class=\"CitationRef\"\u003e36\u003c/span\u003e, \u003cspan citationid=\"CR39\" class=\"CitationRef\"\u003e39\u003c/span\u003e, \u003cspan citationid=\"CR40\" class=\"CitationRef\"\u003e40\u003c/span\u003e]. ACPA prevalence was lowest in typical RA clusters (JIP-hand, JIP-poly) and highest in the oligoarticular cluster (JIP-oligo). While this pattern might partially reflect classification criteria [\u003cspan citationid=\"CR41\" class=\"CitationRef\"\u003e41\u003c/span\u003e], our use of one-year diagnosis validation and physician diagnosis in sets A and C minimizes misclassification bias. The high ACPA-positivity in JIP-foot corroborates recent findings of increased foot involvement in ACPA-positive patients [\u003cspan citationid=\"CR42\" class=\"CitationRef\"\u003e42\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eThe treatment response disparity between hand and foot clusters was most pronounced in ACPA-positive patients. Across replication sets, we consistently observed significant differences between JIP-hand and JIP-poly, with two of three datasets also showing increased response in JIP-hand versus JIP-foot. For remission, we found that the different cluster-associated outcomes vanished in ACPA-positive patients, possibly reflecting the impact of targeted therapy intensification protocols particularly for ACPA positive patients.\u003c/p\u003e \u003cp\u003ePatients with hand-dominant joint inflammation (JIP-hand) showed surprisingly good outcomes, prompting us to explore several possible explanations. We found that these positive results could not be explained solely by how long patients had symptoms, their disease activity at baseline, or their lower prevalence of ACPA in that cluster.\u003c/p\u003e \u003cp\u003eFurther analysis of synovial tissue samples revealed distinct histological differences. The JIP-poly and JIP-hand groups both showed severe synovitis. JIP-poly patients had more lining hyperplasia and sublining leukocytic infiltration, while JIP-hand showed more severe stromal density. Compared to JIP-oligo, which had mostly mild inflammation, these clusters showed significant differences. When we adjusted for disease activity levels, the difference in stromal density disappeared, but the differences in inflammatory infiltrate and cell lining hyperplasia remained significant.\u003c/p\u003e \u003cp\u003eThe marked difference within the histological environment, between the aggressive patterns in JIP-poly and JIP-hand groups and the mild inflammation in JIP-oligo further supports the notion that these may represent fundamentally different disease subtypes. Though, more research is needed to better understand the underlying molecular and cellular mechanisms, and their potential relation to treatment.\u003c/p\u003e \u003cp\u003eThe combination of short symptom duration, widespread joint inflammation, and low seropositivity might suggest that JIP-hand could represent a self-limiting reactive arthritis [\u003cspan citationid=\"CR43\" class=\"CitationRef\"\u003e43\u003c/span\u003e, \u003cspan citationid=\"CR44\" class=\"CitationRef\"\u003e44\u003c/span\u003e], though we found no differences in parvovirus positivity between clusters. An alternative explanation is that the current therapeutic agents may be optimized for JIP-hand -like presentations, given clinical trials' emphasis on upper extremity outcomes.\u003c/p\u003e \u003cp\u003eOur clusters captured previously described age-related subsets, including elderly-onset RA (EORA) in JIP-hand, characterized by higher inflammation markers, lower female prevalence, and reduced autoantibody positivity [\u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e10\u003c/span\u003e, \u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e11\u003c/span\u003e]. However, our analysis revealed more granular subtypes beyond the EORA/YORA dichotomy.\u003c/p\u003e \u003cp\u003eA limitation of our study is that we defined MTX success based on changes in medication, which could include switches due to side effects, though this likely underestimates rather than overestimates the observed associations. Center-specific therapeutic approaches varied but did not affect cluster-outcome associations. While temporal cluster stability was not directly assessed, previous evidence of consistent joint involvement patterns [\u003cspan citationid=\"CR45\" class=\"CitationRef\"\u003e45\u003c/span\u003e] and cross-sectional associations support stability over time.\u003c/p\u003e \u003cp\u003eAnother limitation is that we solely focused on knee biopsies due to insufficient data for other joint locations. This prevented us from exploring and comparing different tissue environments even though they might be crucial for understanding the different phenotypes.\u003c/p\u003e \u003cp\u003eImportant to underline is that our identified clusters are not set in stone. Though we observed a high robustness of our clusters, patients laid on a gradient (Fig. S4) and did not segregate in clearly separable modules. The cluster structure that we identified could also be summarized into more or fewer clusters and the clusters might become clearer when more layers of information are added. Such types of information could be genetics, gene expression patterns and molecular profiles from blood [\u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e15\u003c/span\u003e, \u003cspan citationid=\"CR46\" class=\"CitationRef\"\u003e46\u003c/span\u003e]. Despite these limitations, our study demonstrates the value of unsupervised, data-driven approaches in uncovering hidden disease patterns, with joint involvement patterns emerging as a major axis of variation.\u003c/p\u003e \u003cp\u003eIn conclusion, our clustering analysis revealed four baseline RA phenotypes characterized by distinct hand and foot involvement patterns that predict one-year clinical outcomes, and correspond to histological differences. This data-driven approach provides greater granularity than traditional age or ACPA-based dichotomies, suggesting distinct etiologies that warrant further biological investigation.\u003c/p\u003e"},{"header":"Declarations","content":"\u003ch3\u003eContributors\u003c/h3\u003e\n\u003cp\u003eRK and TM developed the study design together with EB and MPM. TM ran the cluster analysis, while SB ran the permutation analysis. BBdK repeated the survival analysis in the replication set. SAB and CA provided the replication data of the IMPROVED trial, while KG and JVvD provided the data from Reumazorg Zuid West Nederland. SA, LAC and SP provided and processed the SYNGem cohort data for the histological analysis. All authors contributed to the interpretation of the results, and provided ideas for further downstream analysis. AHM annotated the classification criteria. RK and TM drafted the first version of the manuscript. All authors reviewed and approved the final draft of the manuscript to be submitted.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eFunding\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThis project received funding from two major European Union Horizon grants for research and innovation. The first grant supported the SQUEEZE project (activity No. 101095052) and included funding for TDM, MPM, NS, TWJH, and RK. The second grant supported the SPIDeRR project (activity No. 101080711) and funded TDM, MPM, NS, TWJH, MJTR, EvdA, and RK. Additionally, the European Research Council provided financial support for the Glycan Switch project (activity No. 101071386), specifically for TWJH and RK. This study also received co-funding from the ZonMW Klinische Fellow program (No. 40-00703-97-19069) and the ZonMW Open Competitie (No. 09120012110075): for TDM, MPM and RK. The authors declare no competing interests.\u0026nbsp;\u003c/p\u003e\n\u003ch3\u003eCompeting interest\u003c/h3\u003e\n\u003cp\u003eThe authors declare no competing interests.\u0026nbsp;\u003c/p\u003e\n\u003ch3\u003ePatient and public involvement\u003c/h3\u003e\n\u003cp\u003ePatients and/or the public were not involved in the design, or conduct, or reporting, or dissemination plans of this research.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eEthics approval\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eWe received ethical approval from the Medical Ethics Committee (METC) at Leiden University Medical Center according to study protocol B18.057.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eData availability statement\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eWe built an interactive webtool (https://knevel-lab.github.io/Rheumalyze/) that users can use to project their patients on the Leiden data, to infer the JIP-phenotype within their own dataset. Moreover, we have made our scripts available in a public repository at GitHub [33]. Study data is available upon reasonable request.\u0026nbsp;\u003c/p\u003e\n\u003ch2\u003eAcknowledgements\u003c/h2\u003e\n\u003cp\u003eWe would like to express our gratitude to Nick Bos for developing a web tool to cluster patients. Our thanks also go to Samantha Jurado-Zapata and David Steeman for their assistance with extracting and processing Electronic Health Record data from the Leiden University Medical Center. Furthermore, we deeply appreciate Bas van der Wal for his efforts in preparing the Reumazorg Zuidwest Nederland data and Joy van der Pol for his work on preparing the IMPROVED trial data.\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\u003cli\u003e\u003cspan\u003eGrassi W, De Angelis R, Lamanna G, \u003cem\u003eet al\u003c/em\u003e. The clinical features of rheumatoid arthritis. Eur J Radiol. 1998;27 Suppl 1:S18-24. doi: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1016/s0720-048x(98)00038-2\u003c/span\u003e\u003cspan address=\"10.1016/s0720-048x(98)00038-2\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e. PMID: 9652497.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHeidari B. Rheumatoid Arthritis: Early diagnosis and treatment outcomes. Caspian J Intern Med. 2011 Winter;2(1):161\u0026thinsp;\u0026ndash;\u0026thinsp;70. doi: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1016/s0720-048x(98)00038-2\u003c/span\u003e\u003cspan address=\"10.1016/s0720-048x(98)00038-2\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e. PMID: 24024009;\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eAlivernini S, Tolusso B, Gessi M, et al. Inclusion of Synovial Tissue-Derived Characteristics in a Nomogram for the Prediction of Treatment Response in Treatment-Naive Rheumatoid Arthritis Patients. Arthritis Rheumatol. 2021;73(9):1601\u0026ndash;1613. doi: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1002/art.41726\u003c/span\u003e\u003cspan address=\"10.1002/art.41726\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e. Epub 2021 Aug 9. PMID: 33750008; PMCID: PMC8457106.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003ePadyukov L. Genetics of rheumatoid arthritis. Semin Immunopathol. 2022;44(1):47\u0026ndash;62. doi: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1007/s00281-022-00912-0\u003c/span\u003e\u003cspan address=\"10.1007/s00281-022-00912-0\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e. PMID: 35088123;\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSyversen SW, Goll GL, van der Heijde D, \u003cem\u003eet al\u003c/em\u003e. Prediction of radiographic progression in rheumatoid arthritis and the role of antibodies against mutated citrullinated vimentin: results from a 10-year prospective study. Annals of the Rheumatic Diseases 2010;69:345-351.5. doi: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1136/ard.2009.113092\u003c/span\u003e\u003cspan address=\"10.1136/ard.2009.113092\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e. PMID: 19648126.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMatthijssen XME, Niemantsverdriet E, Huizinga TWJ, \u003cem\u003eet al\u003c/em\u003e. Enhanced treatment strategies and distinct disease outcomes among autoantibody-positive and -negative rheumatoid arthritis patients over 25 years: A longitudinal cohort study in the Netherlands. PLOS Medicine 2020;17(9): e1003296. doi: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1371/journal.pmed.1003296\u003c/span\u003e\u003cspan address=\"10.1371/journal.pmed.1003296\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003ePedersen M, Jacobsen S, Klarlund M, \u003cem\u003eet al\u003c/em\u003e. Environmental risk factors differ between rheumatoid arthritis with and without auto-antibodies against cyclic citrullinated peptides. Arthritis Res Ther. 2006;8(4):R133. doi: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1186/ar2022\u003c/span\u003e\u003cspan address=\"10.1186/ar2022\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e. PMID: 16872514;\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eNordberg LB, Lillegraven S, Lie E and the ARCTIC working group, et al. Patients with seronegative RA have more inflammatory activity compared with patients with seropositive RA in an inception cohort of DMARD-na\u0026iuml;ve patients classified according to the 2010 ACR/EULAR criteria. Annals of the Rheumatic Diseases 2017;76:341\u0026ndash;345. doi: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1136/annrheumdis-2015-208873\u003c/span\u003e\u003cspan address=\"10.1136/annrheumdis-2015-208873\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSugihara T, Harigai M. Targeting Low Disease Activity in Elderly-Onset Rheumatoid Arthritis: Current and Future Roles of Biological Disease-Modifying Antirheumatic Drugs. Drugs Aging. 2016;33(2):97\u0026ndash;107. doi: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1007/s40266-015-0341-2\u003c/span\u003e\u003cspan address=\"10.1007/s40266-015-0341-2\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e. PMID: 26833350;\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eDeal CL, Meenan RF, Goldenberg DL, et al. The clinical features of elderly-onset rheumatoid arthritis. A comparison with younger-onset disease of similar duration. Arthritis Rheum. 1985;28(9):987\u0026thinsp;\u0026ndash;\u0026thinsp;94. doi: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1002/art.1780280905\u003c/span\u003e\u003cspan address=\"10.1002/art.1780280905\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e. PMID: 4038365.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSugihara T, Harigai M. Targeting Low Disease Activity in Elderly-Onset Rheumatoid Arthritis: Current and Future Roles of Biological Disease-Modifying Antirheumatic Drugs. Drugs Aging. 2016;33(2):97\u0026ndash;107. doi: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1007/s40266-015-0341-2\u003c/span\u003e\u003cspan address=\"10.1007/s40266-015-0341-2\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e. PMID: 26833350;\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eAhlqvist E, Storm P, K\u0026auml;r\u0026auml;j\u0026auml;m\u0026auml;ki A, \u003cem\u003eet al\u003c/em\u003e. Novel subgroups of adult-onset diabetes and their association with outcomes: a data-driven cluster analysis of six variables. Lancet Diabetes Endocrinol. 2018;6(5):361\u0026ndash;369. doi: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1016/S2213-8587(18)30051-2\u003c/span\u003e\u003cspan address=\"10.1016/S2213-8587(18)30051-2\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e. PMID: 29503172.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHowrylak JA, Fuhlbrigge AL, Strunk RC, \u003cem\u003eet al\u003c/em\u003e. Childhood Asthma Management Program Research Group. Classification of childhood asthma phenotypes and long-term clinical responses to inhaled anti-inflammatory medications. J Allergy Clin Immunol. 2014;133(5):1289\u0026thinsp;\u0026ndash;\u0026thinsp;300, 1300.e1-12. doi: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1016/j.jaci.2014.02.006\u003c/span\u003e\u003cspan address=\"10.1016/j.jaci.2014.02.006\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e. PMID: 24892144;\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eAngelini F, Widera P, Mobasheri A, \u003cem\u003eet al\u003c/em\u003e. Osteoarthritis endotype discovery via clustering of biochemical marker data. Ann Rheum Dis. 2022;81(5):666\u0026ndash;675. doi: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1136/annrheumdis-2021-221763\u003c/span\u003e\u003cspan address=\"10.1136/annrheumdis-2021-221763\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e. PMID: 35246457.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLewis MJ, Barnes MR, Blighe K, \u003cem\u003eet al\u003c/em\u003e. Molecular Portraits of Early Rheumatoid Arthritis Identify Clinical and Treatment Response Phenotypes. Cell Rep. 2019;28(9):2455\u0026ndash;2470.e5. doi: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1016/j.celrep.2019.07.091\u003c/span\u003e\u003cspan address=\"10.1016/j.celrep.2019.07.091\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e. PMID: 31461658;\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLliso-Ribera G, Humby F, Lewis M, \u003cem\u003eet al\u003c/em\u003e. Synovial tissue signatures enhance clinical classification and prognostic/treatment response algorithms in early inflammatory arthritis and predict requirement for subsequent biological therapy: results from the pathobiology of early arthritis cohort (PEAC). Ann Rheum Dis. 2019;78(12):1642\u0026ndash;1652. doi: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1136/annrheumdis-2019-215751\u003c/span\u003e\u003cspan address=\"10.1136/annrheumdis-2019-215751\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e. PMID: 31582377;\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eJung SM, Park KS, Kim KJ. Clinical phenotype with high risk for initiation of biologic therapy in rheumatoid arthritis: a data-driven cluster analysis. Clin Exp Rheumatol. 2021 Nov-Dec;39(6):1282\u0026ndash;1290. doi: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.55563/clinexprheumatol/24zas6\u003c/span\u003e\u003cspan address=\"10.55563/clinexprheumatol/24zas6\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e. PMID: 33635214.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eCurtis JR, Weinblatt M, Saag K, \u003cem\u003eet al\u003c/em\u003e. Data-Driven Patient Clustering and Differential Clinical Outcomes in the Brigham and Women's Rheumatoid Arthritis Sequential Study Registry. Arthritis Care Res (Hoboken). 2021;73(4):471\u0026ndash;480. doi: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1002/acr.24471\u003c/span\u003e\u003cspan address=\"10.1002/acr.24471\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMerola JF, Espinoza LR, Fleischmann R. Distinguishing rheumatoid arthritis from psoriatic arthritis. RMD Open. 2018;4(2):e000656. doi: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1136/rmdopen-2018-000656\u003c/span\u003e\u003cspan address=\"10.1136/rmdopen-2018-000656\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e. PMID: 30167326;\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eOnken A, Yague JG, Sakata S. Deep multimodal autoencoders for identifying latent representations of spike counts and local field potentials. Bernstein Conference 2018. doi: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.12751/nncn.bc2018.0223\u003c/span\u003e\u003cspan address=\"10.12751/nncn.bc2018.0223\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSteinbach M, Ert\u0026ouml;z L, Kumar V. The Challenges of Clustering High Dimensional Data. In: Wille LT, editor. New Directions in Statistical Physics: Econophysics, Bioinformatics, and Pattern Recognition. Berlin, Heidelberg: Springer Berlin Heidelberg; 2004. p. 273\u0026ndash;309. Available from: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1007/978-3-662-08968-2_16\u003c/span\u003e\u003cspan address=\"10.1007/978-3-662-08968-2_16\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eCai J, Fan J, Gu W, \u003cem\u003eet al\u003c/em\u003e. Efficient deep embedded subspace clustering. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition 2022 (pp. 1\u0026ndash;10).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eCastela Forte J, Yeshmagambetova G, van der Grinten ML, \u003cem\u003eet al\u003c/em\u003e. Identifying and characterizing high-risk clusters in a heterogeneous ICU population with deep embedded clustering. Sci Rep. 2021;11(1):12109. doi: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1038/s41598-021-91297-x\u003c/span\u003e\u003cspan address=\"10.1038/s41598-021-91297-x\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e. PMID: 34103544;\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMaarseveen TD, Meinderink T, Reinders MJT, \u003cem\u003eet al\u003c/em\u003e. Machine Learning Electronic Health Record Identification of Patients with Rheumatoid Arthritis: Algorithm Pipeline Development and Validation Study. JMIR Med Inform. 2020;8(11):e23930. doi: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.2196/23930\u003c/span\u003e\u003cspan address=\"10.2196/23930\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e. PMID: 33252349;\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMaarseveen TD, Maurits MP, Niemantsverdriet E, \u003cem\u003eet al\u003c/em\u003e. Handwork vs machine: a comparison of rheumatoid arthritis patient populations as identified from EHR free-text by diagnosis extraction through machine-learning or traditional criteria-based chart review. Arthritis Res Ther. 2021;23(1):174. doi: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1186/s13075-021-02553-4\u003c/span\u003e\u003cspan address=\"10.1186/s13075-021-02553-4\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e. PMID: 34158089;\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHeimans L, Wevers-de Boer KV, Visser K, \u003cem\u003eet al\u003c/em\u003e. A two-step treatment strategy trial in patients with early arthritis aimed at achieving remission: the IMPROVED study. Ann Rheum Dis. 2014;73(7):1356-61. doi: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1136/annrheumdis-2013-203243\u003c/span\u003e\u003cspan address=\"10.1136/annrheumdis-2013-203243\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e. PMID: 23716067.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eKrenn V, Morawietz L, Burmester GR, Kinne RW, Mueller-Ladner U, Muller B \u0026amp; Haupl T. Synovitis score: discrimination between chronic low‐grade and high‐grade synovitis. Histopathology, 2006; 49(4), 358\u0026ndash;364. doi: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1111/j.1365-2559.2006.02508.x\u003c/span\u003e\u003cspan address=\"10.1111/j.1365-2559.2006.02508.x\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e. PMID: 16978198.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLevine JH, Simonds EF, Bendall SC, \u003cem\u003eet al\u003c/em\u003e. Data-Driven Phenotypic Dissection of AML Reveals Progenitor-like Cells that Correlate with Prognosis. Cell. 2015;162(1):184\u0026thinsp;\u0026ndash;\u0026thinsp;97. doi: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1016/j.cell.2015.05.047\u003c/span\u003e\u003cspan address=\"10.1016/j.cell.2015.05.047\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e. PMID: 26095251;\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eRoth, AE. The Shapley Value: Essays in Honor of Lloyd S. Shapley. Cambridge: Cambridge University Press. 1988. doi:\u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1017/CBO9780511528446\u003c/span\u003e\u003cspan address=\"10.1017/CBO9780511528446\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eKorsunsky I, Millard N, Fan J, \u003cem\u003eet al\u003c/em\u003e. Fast, sensitive and accurate integration of single-cell data with Harmony. Nat Methods. 2019;16(12):1289\u0026ndash;1296. doi: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1038/s41592-019-0619-0\u003c/span\u003e\u003cspan address=\"10.1038/s41592-019-0619-0\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e. Epub 2019 Nov 18. PMID: 31740819;\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eCox DR. Regression models and life-tables. Journal of the Royal Statistical Society: Series B (Methodological), 1972; 34(2), 187\u0026ndash;202.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSchoenfeld, D. Chi-squared goodness-of-fit tests for the proportional hazards regression model. Biometrika, 1980; 67(1), 145\u0026ndash;153.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMaarseveen TD. EHR Clustering RA. GitHub. Published: August 8, 2022. Link: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://github.com/levrex/EHR-Clustering-RA\u003c/span\u003e\u003cspan address=\"https://github.com/levrex/EHR-Clustering-RA\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eNishimura K, Sugiyama D, Kogata Y, \u003cem\u003eet al\u003c/em\u003e. Meta-analysis: Diagnostic Accuracy of Anti\u0026ndash;Cyclic Citrullinated Peptide Antibody and Rheumatoid Factor for Rheumatoid Arthritis. Ann Intern Med.2007;146:797\u0026ndash;808. doi:\u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.7326/0003-4819-146-11-200706050-00008\u003c/span\u003e\u003cspan address=\"10.7326/0003-4819-146-11-200706050-00008\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003evan Nies JAB, Tsonaka R, Gaujoux-Viala C, \u003cem\u003eet al\u003c/em\u003e. Evaluating relationships between symptom duration and persistence of rheumatoid arthritis: does a window of opportunity exist? Results on the Leiden Early Arthritis Clinic and ESPOIR cohorts. Annals of the Rheumatic Diseases 2015;74:806\u0026ndash;812.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eVisser H, le Cessie S, Vos K, \u003cem\u003eet al\u003c/em\u003e. How to diagnose rheumatoid arthritis early: a prediction model for persistent (erosive) arthritis. Arthritis Rheum. 2002;46(2):357\u0026thinsp;\u0026ndash;\u0026thinsp;65. doi: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1002/art.10117\u003c/span\u003e\u003cspan address=\"10.1002/art.10117\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e. PMID: 11840437.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLee SW, Kim SY, Chang SH. Prevalence of feet and ankle arthritis and their impact on clinical indices in patients with rheumatoid arthritis: a cross-sectional study. BMC Musculoskelet Disord 20, 420 (2019). \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1186/s12891-019-2773-z\u003c/span\u003e\u003cspan address=\"10.1186/s12891-019-2773-z\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHoque A, Steultjens M, Dickson DM, \u003cem\u003eet al\u003c/em\u003e. Patients' and clinicians' perspectives on the clinical utility of the Rheumatoid Arthritis Foot Disease Activity Index. Rheumatol Int. 2022;42(10):1807\u0026ndash;1817. doi: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1007/s00296-022-05147-8\u003c/span\u003e\u003cspan address=\"10.1007/s00296-022-05147-8\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e. Epub 2022 May 27. PMID: 35622086;\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003evan der Helm-van Mil AHM, Verpoort KN, Breedveld FC, \u003cem\u003eet al\u003c/em\u003e. Antibodies to citrullinated proteins and differences in clinical progression of rheumatoid arthritis. Arthritis Res Ther. 2005;7(5):R949-58. doi: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1186/ar1767\u003c/span\u003e\u003cspan address=\"10.1186/ar1767\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e. PMID: 16207336;\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eBugatti S, Manzo A, Montecucco C \u003cem\u003eet al.\u003c/em\u003e The Clinical Value of Autoantibodies in Rheumatoid Arthritis. Front Med (Lausanne) 2018;5:339.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eAletaha D, Neogi T, Silman AJ, \u003cem\u003eet al\u003c/em\u003e. 2010 Rheumatoid arthritis classification criteria: an American College of Rheumatology/European League Against Rheumatism collaborative initiative. Arthritis Rheum. 2010;62(9):2569-81. doi: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1002/art.27584\u003c/span\u003e\u003cspan address=\"10.1002/art.27584\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e. PMID: 20872595.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eBurgers LE, van Steenbergen HW, Ten Brinck RM, \u003cem\u003eet al\u003c/em\u003e. Differences in the symptomatic phase preceding ACPA-positive and ACPA-negative RA: a longitudinal study in arthralgia during progression to clinical arthritis. Ann Rheum Dis. 2017;76(10):1751\u0026ndash;1754. doi: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1136/annrheumdis-2017-211325\u003c/span\u003e\u003cspan address=\"10.1136/annrheumdis-2017-211325\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e. PMID: 28606964.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eTakahashi Y, Murai C, Shibata S, \u003cem\u003eet al\u003c/em\u003e. Human parvovirus B19 as a causative agent for rheumatoid arthritis. Proc Natl Acad Sci U S A. 1998;95(14):8227\u0026ndash;8232. doi:\u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1073/pnas.95.14.8227\u003c/span\u003e\u003cspan address=\"10.1073/pnas.95.14.8227\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eNaides SJ, Scharosch LL, Foto F, \u003cem\u003eet al\u003c/em\u003e. Rheumatologic manifestations of human parvovirus B19 infection in adults. Initial two-year clinical experience. Arthritis Rheum. 1990;33(9):1297\u0026thinsp;\u0026ndash;\u0026thinsp;309. doi: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1002/art.1780330901\u003c/span\u003e\u003cspan address=\"10.1002/art.1780330901\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e. PMID: 2169746.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHeckert SL, Bergstra SA, Matthijssen XME, \u003cem\u003eet al\u003c/em\u003e. Joint inflammation tends to recur in the same joints during the rheumatoid arthritis disease course. Ann Rheum Dis. 2022;81(2):169\u0026ndash;174. doi: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1136/annrheumdis-2021-220882\u003c/span\u003e\u003cspan address=\"10.1136/annrheumdis-2021-220882\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e. PMID: 34462262.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eZhang F, Jonsson AH, Nathan A, \u003cem\u003eet al\u003c/em\u003e. Cellular deconstruction of inflamed synovium defines diverse inflammatory phenotypes in rheumatoid arthritis. bioRxiv. 2022;2022.02.25.481990.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eFOOTNOTES\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eContributors\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eRK and TM developed the study design together with EB and MPM. TM ran the cluster analysis, while SB ran the permutation analysis. BBdK repeated the survival analysis in the replication set. SAB and CA provided the replication data of the IMPROVED trial, while KG and JVvD provided the data from Reumazorg Zuid West Nederland. SA, LAC and SP provided and processed the SYNGem cohort data for the histological analysis. All authors contributed to the interpretation of the results, and provided ideas for further downstream analysis. AHM annotated the classification criteria. RK and TM drafted the first version of the manuscript. All authors reviewed and approved the final draft of the manuscript to be submitted.\u003c/span\u003e\u003c/li\u003e\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":false,"hideJournal":false,"highlight":"","institution":"","isAcceptedByJournal":true,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"npj-digital-medicine","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"npjdigitalmed","sideBox":"Learn more about [npj Digital Medicine](http://www.nature.com/npjdigitalmed/)","snPcode":"41746","submissionUrl":"https://submission.springernature.com/new-submission/41746/3","title":"npj Digital Medicine","twitterHandle":"","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"stoa","reportingPortfolio":"NPJ","inReviewEnabled":true,"inReviewRevisionsEnabled":true},"keywords":"","lastPublishedDoi":"10.21203/rs.3.rs-6256181/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-6256181/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003ch2\u003eBackground\u003c/h2\u003e \u003cp\u003eRheumatoid arthritis (RA) is a heterogeneous disease. Patients vary in symptoms, prognosis and treatment response, demonstrating the need for a more refined taxonomy.\u003c/p\u003e\u003ch2\u003eObjective\u003c/h2\u003e \u003cp\u003eTo identify distinct phenotypic subsets of RA patients based on baseline clinical data, in order to advance understanding of disease etiology and treatment strategies.\u003c/p\u003e\u003ch2\u003eMethods\u003c/h2\u003e \u003cp\u003eWe collected hematological, serological, and clinical data from RA-patients in the Leiden Rheumatology clinic(n\u0026thinsp;=\u0026thinsp;1,387), and combined multimodal deep learning techniques with clustering to identify phenotypically distinct RA subsets. These clusters were tested for associations in clinical outcomes. Findings were replicated in clinical trial data (n\u0026thinsp;=\u0026thinsp;307) and independent secondary care (9 clinics, n\u0026thinsp;=\u0026thinsp;515), and further explored for histological differences in synovial tissue (n\u0026thinsp;=\u0026thinsp;194).\u003c/p\u003e\u003ch2\u003eResults\u003c/h2\u003e \u003cp\u003eFour distinct RA subsets with different Joint Involvement Patterns (JIP), emerged: 1) foot-predominant arthritis, 2) seropositive oligoarticular disease, 3) seronegative hand arthritis, and 4) polyarthritis. We found high cluster stability, no physician influence, significant difference in remission rates \u003cem\u003e(P\u003c/em\u003e\u0026thinsp;=\u0026thinsp;0.007) and methotrexate failure (\u003cem\u003eP\u003c/em\u003e\u0026thinsp;\u0026lt;\u0026thinsp;0.001) in initial and replication sets. The JIP-hand subgroup had significantly better outcomes. This was largest in the ACPA-positive stratum (JIP-hand versus JIP-foot (HR:0.37 (95%CI: 0.15\u0026ndash;0.60) \u003cem\u003eP\u003c/em\u003e\u0026thinsp;\u0026lt;\u0026thinsp;0.001), JIP-hand versus JIP-poly HR:0.33 (95%CI: 0.15\u0026ndash;0.72) \u003cem\u003eP\u003c/em\u003e\u0026thinsp;=\u0026thinsp;0.005). This was independent of baseline disease activity, clinical markers (RF, ACPA, Sex, Age), and symptom duration. Synovial histology showed both JIP-poly and JIP-hand had increased synovial lining and inflammatory infiltrate, with JIP-hand showing notably high stromal density. JIP-feet scored evenly across categories without standing out, while JIP-oligo had lower synovitis degree.\u003c/p\u003e\u003ch2\u003eConclusions\u003c/h2\u003e \u003cp\u003eWe identified and validated four distinct RA phenotypes characterized by joint involvement patterns, which associate with treatment outcomes and synovial histology. These findings may allow for targeted research into RA mechanisms and therapies.\u003c/p\u003e","manuscriptTitle":"Location and amount of joint involvement differentiates rheumatoid arthritis into different clinical subsets","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2025-04-01 09:09:06","doi":"10.21203/rs.3.rs-6256181/v1","editorialEvents":[{"type":"communityComments","content":1},{"type":"decision","content":"Revision requested","date":"2025-06-04T14:07:18+00:00","index":"","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2025-05-29T08:51:49+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"275198810063424656481840933594106017556","date":"2025-05-18T18:24:32+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"202210993640333774941050752831445238141","date":"2025-05-18T03:20:16+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"17479079521753617997913977112756859917","date":"2025-05-16T06:04:40+00:00","index":"hide","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2025-04-30T23:01:04+00:00","index":"hide","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2025-04-07T08:51:21+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"256660090695209226438807116002826586491","date":"2025-04-03T17:18:32+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"122546038296964736848804304648088020610","date":"2025-03-31T07:16:20+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"38027876096892478324895694524064580755","date":"2025-03-29T09:29:10+00:00","index":"hide","fulltext":""},{"type":"reviewersInvited","content":"","date":"2025-03-21T08:51:30+00:00","index":"","fulltext":""},{"type":"editorAssigned","content":"","date":"2025-03-21T01:20:23+00:00","index":"","fulltext":""},{"type":"checksComplete","content":"","date":"2025-03-20T05:50:39+00:00","index":"","fulltext":""},{"type":"submitted","content":"npj Digital Medicine","date":"2025-03-18T20:51:31+00:00","index":"","fulltext":""}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"npj-digital-medicine","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"npjdigitalmed","sideBox":"Learn more about [npj Digital Medicine](http://www.nature.com/npjdigitalmed/)","snPcode":"41746","submissionUrl":"https://submission.springernature.com/new-submission/41746/3","title":"npj Digital Medicine","twitterHandle":"","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"stoa","reportingPortfolio":"NPJ","inReviewEnabled":true,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"36d23590-99e1-4625-b002-9d8e3597c035","owner":[],"postedDate":"April 1st, 2025","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"published-in-journal","subjectAreas":[{"id":46392305,"name":"Health sciences/Diseases/Rheumatic diseases/Rheumatoid arthritis"},{"id":46392306,"name":"Biological sciences/Computational biology and bioinformatics/Classification and taxonomy"},{"id":46392307,"name":"Biological sciences/Computational biology and bioinformatics/Computational models"},{"id":46392308,"name":"Biological sciences/Computational biology and bioinformatics/Data integration"},{"id":46392309,"name":"Biological sciences/Computational biology and bioinformatics/Data processing"},{"id":46392310,"name":"Biological sciences/Computational biology and bioinformatics/Machine learning"},{"id":46392311,"name":"Biological sciences/Biological techniques/Bioinformatics"},{"id":46392312,"name":"Biological sciences/Immunology/Inflammation"},{"id":46392313,"name":"Biological sciences/Computational biology and bioinformatics"},{"id":46392314,"name":"Biological sciences/Immunology"},{"id":46392315,"name":"Health sciences/Medical research"}],"tags":[],"updatedAt":"2025-10-27T16:38:15+00:00","versionOfRecord":{"articleIdentity":"rs-6256181","link":"https://doi.org/10.1038/s41746-025-01997-1","journal":{"identity":"npj-digital-medicine","isVorOnly":false,"title":"npj Digital Medicine"},"publishedOn":"2025-10-23 16:16:51","publishedOnDateReadable":"October 23rd, 2025"},"versionCreatedAt":"2025-04-01 09:09:06","video":"","vorDoi":"10.1038/s41746-025-01997-1","vorDoiUrl":"https://doi.org/10.1038/s41746-025-01997-1","workflowStages":[]},"version":"v1","identity":"rs-6256181","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-6256181","identity":"rs-6256181","version":["v1"]},"buildId":"XKTyCvWXoU3ODBz1xrDgd","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}