{"paper_id":"7145d5e1-1533-4476-9456-660a8d7fdc6b","body_text":"1 \nTitle: Risk-stratified monitoring for sulfasalazine toxicity: prognostic model \ndevelopment and validation. \nAuthors: A Abhishek (https://orcid.org/0000 -0003-0121-4919)1 PhD, Matthew J \nGrainge2 PhD, Tim Card 2 PhD, Hywel C Williams2 PhD, Maarten W Taal  \n(https://orcid.org/0000-0002-9065-212X)3 MD, Guruprasad P Aithal \n(https://orcid.org/0000-0003-3924-4830)4 PhD, Christopher P Fox 5 PhD, Christian D \nMallen6 PhD, Matthew D Stevenson7 PhD, Georgina Nakafero (https://orcid.org/0000-\n0002-3859-7354)1 PhD*, Richard D Riley8 PhD*.*Contributed equally. \nAddress: 1Academic Rheumatology, School of Medicine, University of Nottingham, \nNottingham NG5 1PB, UK . 2Lifespan and Population Health, School of Medicine, \nUniversity of Nottingham, Nottingham NG5 1PB, UK. 3Centre for Kidney Research and \nInnovation, Translational Medical Sciences, School of Medicine, University of \nNottingham, Derby DE22 3NE, UK . 4Nottingham Digestive Diseases Centre, \nTranslational Medical Sciences, School of Medicine, University of Nottingham, \nNottingham NG7 2UH, UK.  5Translational Medical Sciences, School of Medicine, \nUniversity of Nottingham, Nottingham, UK . 6Primary Care Centre Versus Arthritis, \nSchool of Medicine, Keele University, Keele ST5 5BJ, UK . 7School of Health and \nRelated Research, University of Sheffield, Sheffield S1 4DA, UK. 8Institute of Applied \nHealth Research, College of Medical and Dental Sciences, University of Birmingham, \nBirmingham B15 2TT, UK. \nCorresponding author: Prof. Abhishek Address for correspondence: A23, Academic \nRheumatology, Clinical Sciences Building, The University of Nottingham, Nottingham NG5 \n1PB, UK.Email: abhishek.abhishek@nottingham.ac.uk            \nWord count: 3434      Abstract word count: 250 \n . CC-BY 4.0 International licenseIt is made available under a \nperpetuity. \n is the author/funder, who has granted medRxiv a license to display the preprint in(which was not certified by peer review)preprint \nThe copyright holder for thisthis version posted December 18, 2023. ; https://doi.org/10.1101/2023.12.15.23299947doi: medRxiv preprint \nNOTE: This preprint reports new research that has not been certified by peer review and should not be used to guide clinical practice.\n\n 2 \nConflicts of interest A.A. has received Institutional research grants from AstraZeneca \nand Oxford  Immunotech; and personal fees from UpToDate (royalty), Springer \n(royalty), Cadilla Pharmaceuticals (lecture fees), NGM Bio (consulting), Limbic \n(consulting) and personal fees from Inflazome (consulting) unrelated to the work. GP \nAithal has received consulting fees from Abbott, Albereo, Amryth, AstraZeneca, \nBenevolent AI, DNDI, GlaxoSmithKline, NuCANA, Pfizer, Roche Diagnostics, Servier \nPharmaceuticals, W.L Gore & Associates paid to the University of Nottingham \nunrelated to the work . CPF has received Consultancy/Advisory board fees from \nAbbvie, GenMab, Incyte, Morphosys,  Roche, Takeda, Ono, Kite/Gilead, \nBMS/Celgene, BTG/Veriton and departmental research funding from BeiGene \nunrelated to the work. The other authors have no conflict of interest to declare. \n \n \n  \n . CC-BY 4.0 International licenseIt is made available under a \nperpetuity. \n is the author/funder, who has granted medRxiv a license to display the preprint in(which was not certified by peer review)preprint \nThe copyright holder for thisthis version posted December 18, 2023. ; https://doi.org/10.1101/2023.12.15.23299947doi: medRxiv preprint \n\n 3 \nAbstract:  \nBackground: Sulfasalazine induced cytopenia, nephrotoxicity, and hepatotoxicity is \nuncommon during long-term treatment. Some guidelines recommend three monthly \nmonitoring blood-tests indefinitely while others recommend stopping monitoring after \none year. To rationalise monitoring we developed and validated a prognostic model \nfor clinically significant blood, liver, or kidney toxicity during established sulfasalazine \ntreatment. \nDesign: Retrospective cohort study. \nSetting: UK primary-care. Data from Clinical Practice Research Datalink  Gold and \nAurum formed independent development and validation cohorts. \nParticipants: Age ≥18 years , new diagnosis of  an inflammatory condition  and \nsulfasalazine prescription. \nStudy period: 01/01/2007 to 31/12/2019. \nOutcome: Sulfasalazine discontinuation with abnormal monitoring blood-test result. \nAnalysis: Patients were followed -up from six  months after first primary-care \nprescription to the earliest of outcome , drug discontinuation, death, 5  years, or \n31/12/2019.Penalised Cox regression was performed to develop the risk equation . \nMultiple imputation handled missing predictor data. Model performance was assessed \nin terms of calibration and discrimination. \nResults: 8,936 participants were included in the development cohort (473 events, \n23,299 person-years) and 5,203 participants were included in the validation cohort \n(280 events, 12,867 person-years).Nine candidate predictors were included . The \noptimism adjusted R2D and Royston D statistic in the development data were 0.13 and \n0.79 respectively. The calibration slope (95% confidence interval (CI)) and Royston D \n . CC-BY 4.0 International licenseIt is made available under a \nperpetuity. \n is the author/funder, who has granted medRxiv a license to display the preprint in(which was not certified by peer review)preprint \nThe copyright holder for thisthis version posted December 18, 2023. ; https://doi.org/10.1101/2023.12.15.23299947doi: medRxiv preprint \n\n 4 \nstatistic (95% CI) in validation cohort was 1.19 (0.96-1.43) and 0.87 (0.67-1.07) \nrespectively. \nConclusion: This prognostic model for sulfasalazine toxicity utilises readily available \ndata and should be used to risk-stratify blood-test monitoring  during established \nsulfasalazine treatment. \n  \n . CC-BY 4.0 International licenseIt is made available under a \nperpetuity. \n is the author/funder, who has granted medRxiv a license to display the preprint in(which was not certified by peer review)preprint \nThe copyright holder for thisthis version posted December 18, 2023. ; https://doi.org/10.1101/2023.12.15.23299947doi: medRxiv preprint \n\n 5 \nEvidence before this study? \n• Hepatic, haematological, and renal toxicity from sulfasalazine occurs \nuncommonly after the first -few months of treatment . Nevertheless, the \nmanufacturers and some specialist societies e.g ., the American College of \nRheumatology recommend monitoring blood-tests at three monthly intervals \nduring established treatment. Other guidelines e.g., from the British Society \nof Rheumatology recommend no monitoring after the first two years of \ntreatment.   \n• It is not known whether hepatic, haematological, and renal  toxicities due to \nsulfasalazine can be predicted and monitoring be risk-stratified. \nAdded value of this study? \n• This study  developed a prognostic model that discriminated patients at \nvarying risk of sulfasalazine toxicity during long -term treatment . It had \nexcellent performance characteristics in an independent validation cohort. \n• The model performed well across age-groups, and in people with rheumatoid \narthritis and other inflammatory conditions. \n• Any cytopenia or liver enzyme elevation prior to start of follow-up, chronic \nkidney disease stage-3, diabetes, methotrexate prescription, leflunomide \nprescription, and age were strong predictors of sulfasalazine toxicity. \nImplications of all the available evidence. \n• This prognostic model  utilises information that can be easily ascertained \nduring clinical  visits. It can be used to inform decisions on the interval \nbetween monitoring blood-tests. \n . CC-BY 4.0 International licenseIt is made available under a \nperpetuity. \n is the author/funder, who has granted medRxiv a license to display the preprint in(which was not certified by peer review)preprint \nThe copyright holder for thisthis version posted December 18, 2023. ; https://doi.org/10.1101/2023.12.15.23299947doi: medRxiv preprint \n\n 6 \n• The results of this study ought to be considered by national and international \nRheumatology guideline writing groups to rationalise monitoring during long-\nterm sulfasalazine treatment. \n \n  \n . CC-BY 4.0 International licenseIt is made available under a \nperpetuity. \n is the author/funder, who has granted medRxiv a license to display the preprint in(which was not certified by peer review)preprint \nThe copyright holder for thisthis version posted December 18, 2023. ; https://doi.org/10.1101/2023.12.15.23299947doi: medRxiv preprint \n\n 7 \nIntroduction Sulfasalazine is commonly used  in the treatment o f inflammatory \ndiseases such as rheumatoid arthritis  (RA), psoriatic arthritis  (PsA), axial \nspondylarthritis, reactive arthritis, and infrequently in the management of inflammatory \nbowel disease (IBD) (the latter is mostly treated with 5-aminosalicylates due to a better \nsafety profile)1-3. Although effective, sulfasalazine can cause cytopenia and elevated \nliver enzymes typically in the first three to six months of treatment, although late onset \ntoxicity is reported 4-16. Sulfasalazine can also cause crystalluria  and interstitial \nnephritis, and is not recommended in those with severe renal impairment 17. Cautious \nuse is recommended in those with mild to moderate renal impairment17. \n \nThere is considerable inconsistency in guid ance on how to monitor patients on long-\nterm sulfasalazine treatment for asymptomatic bone marrow, liver and/or renal toxicity. \nThe British Society of Rheumatology (BSR) guidelines recommend two to four weekly \nblood-tests for full blood count  (FBC), liver function test  (LFT), urea electrolytes and \ncreatinine (UE&C) for the first three months of treatment followed by three-monthly \ntesting in the first year  and no  further monitoring blood-tests thereafter18. On the \ncontrary, the American College of Rheumatology (ACR) guidelines recommend close \nmonitoring for the first three months of treatment, followed by three-monthly blood-\ntesting for FBC, UE&C, and LFT during the entire duration of treatment 19. The \nsummary of product characteristics for sulfasalazine recommends monitoring with \nFBC, LFT and UE&C at three monthly intervals during long-term treatment20. However, \nwhether everyone needs a fixed monitoring schedule  once established on \nsulfasalazine treatment, or whether monitoring can be risk-stratified during long-term \ntreatment is not known. \n \n . CC-BY 4.0 International licenseIt is made available under a \nperpetuity. \n is the author/funder, who has granted medRxiv a license to display the preprint in(which was not certified by peer review)preprint \nThe copyright holder for thisthis version posted December 18, 2023. ; https://doi.org/10.1101/2023.12.15.23299947doi: medRxiv preprint \n\n 8 \nTo predict clinically significant laboratory abnormalities  during established \nsulfasalazine treatment and to inform the frequency of testing, we have developed and \nvalidated a prognostic model for clinically significant myelotoxicity, hepatotoxicity \nand/or nephrotoxicity due to sulfasalazine. \n . CC-BY 4.0 International licenseIt is made available under a \nperpetuity. \n is the author/funder, who has granted medRxiv a license to display the preprint in(which was not certified by peer review)preprint \nThe copyright holder for thisthis version posted December 18, 2023. ; https://doi.org/10.1101/2023.12.15.23299947doi: medRxiv preprint \n\n 9 \nMethods Data source:  Data from the Clinical Practice Research Datalink (CPRD) \nAurum and Gold were used for model development and validation respectively 21 22. \nCPRD is an anonymised longitudinal database of electronic health records originated \nduring clinical care in the N ational Health Service in the UK . With almost universal \ncoverage of UK residents,  participants that contributed data to the CPRD are \nrepresentative of the UK population 21. The CPRD  includes information on \ndemographic details, lifestyle factors (e.g., smoking, alcohol intake), diagnoses, \nresults of blood -tests, and details of primary-care prescriptions. CPRD Gold and \nAurum complement each other in terms of coverage of general -practices due to their \nuse of different software for data capture . Some g eneral practices that have \ncontributed data to both databases are identifiable using a bridging file provided by the \nCPRD. \n \nApprovals: Independent Scientific Advisory Committee of the MHRA (Reference: \n19_275R, 20_000236R). \nStudy design: Retrospective cohort study. \nStudy period: 1st January 2007 to 31st December 2019. \n \nStudy population:  Participants aged 18 years or older with a  new diagnosis of  \ninflammatory disease (e.g., RA, axial spondyloarthritis, PsA, IBD etc.) and prescribed \nsulfasalazine by their GP for ≥six months were eligible. Patients were required to have \n≥1-year disease-free registration in their current general practice  to be classified as \nhaving a new diagnosis 23.Additionally, patients were required to have received their \nfirst sulfasalazine prescription either after the first record of inflammatory disease in \nthe CPRD or in the 90 -days preceding . This 90-day period was allowed because \n . CC-BY 4.0 International licenseIt is made available under a \nperpetuity. \n is the author/funder, who has granted medRxiv a license to display the preprint in(which was not certified by peer review)preprint \nThe copyright holder for thisthis version posted December 18, 2023. ; https://doi.org/10.1101/2023.12.15.23299947doi: medRxiv preprint \n\n 10 \nrecording of diagnosis may lag prescriptions. These two requirements minimised the \nchance of patients on long -term sulfasalazine treatment appearing as new users of \nsulfasalazine when they moved to a different general practice . Patients with chronic \nliver disease, haematological disease, and chronic kidney disease (CKD) stage 4 or 5 \nprior to cohort entry were excluded as described in a previous manuscript24. \n \nSulfasalazine prescriptions: In the UK, sulfasalazine initiation and dose -escalation \noccur in hospital out-patient clinics. During this period prescriptions are issued by the \nhospital specialists . They also  organise monitoring blood -tests and acts on any \nabnormalities. Once a patient is established on treatment, typically approximately six \nmonths after initiating on treatment, the responsibility for prescribing and monitoring, \nincluding with periodic blood-tests is handed to the patients’ general practitioner (GP) \nas per the NHS shared-care protocols. During shared-care monitoring, the GP seeks \nadvice from the hospital specialist if there are side -effects including abnormal blood -\ntest results, and treatment changes are directed by the specialist. \n \nStart of f ollow-up: Patients were followed-up from 180 days after  their first primary-\ncare sulfasalazine prescription until the earliest  of outcome, death, transfer out of \npractice, 90-days prescription gap, last data collection from practice, 31/12/2019  or \nfive-years. \n \nOutcome: Sulfasalazine-toxicity associated drug discontinuation was the outcome  of \ninterest. This was defined as a prescription gap of ≥90 days with either an abnormal \nblood-test result or a diagnostic code for abnormal blood-test result within ±60 days of \nthe last prescription date25.The blood tests were considered abnormal if any of the \n . CC-BY 4.0 International licenseIt is made available under a \nperpetuity. \n is the author/funder, who has granted medRxiv a license to display the preprint in(which was not certified by peer review)preprint \nThe copyright holder for thisthis version posted December 18, 2023. ; https://doi.org/10.1101/2023.12.15.23299947doi: medRxiv preprint \n\n 11 \nfollowing were present: total leucocyte count <3.5×109/L, neutrophil count <1.6×109/L, \nplatelet count <140×109/L, alanine transaminase and/or aspartate transaminase >100 \nIU/mL, and decline in kidney function, defined as either progression of chronic kidney \ndisease based on medical codes recorded by the GP, or >26 μmol/L increase in \ncreatinine concentration, the th reshold for consideration of acute kidney injury 18 26.In \na previous validation stud y on methotrexate discontinuation , only 5.4% of abnormal \nblood-test results in this time-window were potentially explained by an alternate illness \n25. \nA random sample  of sulfasalazine discontinuations with abnormal blood test results \nwas drawn. Data for all diagnostic codes entered during primary -care consultations \nwithin ±60 days of the abnormal blood test result were extracted . A.A. screened the \nlist to identify outcomes that could potentially be explained by an alternative condition \nor its treatment. \nPredictors: These were selected by the clinical members of the study-team based on \ntheir clinical expertise and knowledge of the published literature. Age, sex, body mass \nindex (BMI), alcohol intake, and diabetes were included as they  associate with drug \ninduced liver injury (DILI)  27 28 . Individual inflammatory disease s were considered \nseparately because sulfasalazine toxicity is reported to be less common in people with \ninflammatory bowel disease than in those with RA3. CKD stage-3 was included as it \nreduces sulfasalazine  clearance29. Statins, carbamazepine, valproate, and \nparacetamol were included as their use is associated with sulfasalazine toxicity as per \nthe British National Formulary. Methotrexate, leflunomide, thiopurines were included \nas they can cause cytopenia, elevated liver enzymes and acute kidney injury \n(AKI).Either cytopenia (neutrophil count <2 x 109/l, total leucocyte count <4 x 109/l, or \nplatelet count <150 109/l) or elevated transaminase (ALT and/or AST >35 IU/l) during \n . CC-BY 4.0 International licenseIt is made available under a \nperpetuity. \n is the author/funder, who has granted medRxiv a license to display the preprint in(which was not certified by peer review)preprint \nThe copyright holder for thisthis version posted December 18, 2023. ; https://doi.org/10.1101/2023.12.15.23299947doi: medRxiv preprint \n\n 12 \nthe first six months of primary-care prescription were included  as they  predicted \ncytopenia and/or transaminitis in other studies 30 31. \n \nThe latest record of demographic and lifestyle factors,  diseases recorded within two \nyears prior to start of follow-up, and latest primary-care prescriptions within six-month \nprior to start of follow-up were used to define predictors except for CKD stage-3 that \nwas defined using both GP records and/or eGFR 30-59 ml/min. GPs typically review \npatients with long -term conditions  annually. A two -year look -back was utilised to \nminimise the risk of missing data from those that did not attend in a year. \n \nPatient and public involvement (PPI)  PPI members were involved in selecting and \nprioritising the research question . They advised to use readily available datasets for \nthe study rather than conduct an expensive and time-consuming clinical trial.  \n \nSample size:  In a previously published cohort of 1 ,321 RA patients, 85 stopped \nsulfasalazine with neutropenia, thrombocytopenia, or elevated liver enzymes during a \nmean follow -up of 2.39 years16. Assuming a similar incidence of treatment \ndiscontinuation for model development, the minimum sample size needed to minimise \nmodel overfitting (a target shrinkage factor of 0.9) and ensure precise estimation of \noverall risk was 1 ,748 participants ( 113 outcomes) based on a maximum of 25 \nparameters, Cox-Snell R2 value of 0.12,  outcome rate of 0.0 27/person-year16, a 5 -\nyear time horizon, and a mean follow -up period of 2.39 years using the formulae of \nRiley et al.32. The sample size for external model validation was much larger than the \ntypically recommended minimum sample size of 200 events 33. \n \n . CC-BY 4.0 International licenseIt is made available under a \nperpetuity. \n is the author/funder, who has granted medRxiv a license to display the preprint in(which was not certified by peer review)preprint \nThe copyright holder for thisthis version posted December 18, 2023. ; https://doi.org/10.1101/2023.12.15.23299947doi: medRxiv preprint \n\n 13 \nStatistical analysis: Multiple imputation handled missing data on BMI, alcohol intake, \nand sulfasalazine dose using chained equations 34. We carried out 10 imputations in \nthe development dataset and five imputations in the validation dataset - a pragmatic \napproach considering the larger size of CPRD Aurum. The imputation model included \nall candidate predictors, Nelson -Aalen cumulative hazard function and outcome \nvariable. The data analysis was undertaken using the Stata command “mi estimate” in \na combined dataset that included all imputations. \n \nModel development: Fraction polynomial regression (first degree) analysis was used \nto model non-linear risk relationships with continuous predictors, but these were not \nbetter than the linear terms (p > 0.05), hence were not transformed. All 12 candidate \npredictors (19 parameters) were included in the Cox model and coefficients of each \nparameter estimated and combined using Rubin’s rule across the imputed datasets . \nThe risk equation for predicting an individual’s risk of sulfasalazine discontinuation with \nabnormal blood-test results by five-years follow-up was formulated in the development \ndata. The baseline survival function at t=5 years, a non-parametric estimate of survival \nfunction when all predictor values are set to zero, which is equivalent to the Kaplan -\nMeier product -limit estimate, was estimated along with the estimated regression \ncoefficients (β) and the individual’s predictor values (X).This led to the equation for the \npredicted absolute risk over time 35:  \nPredicted risk of sulfasalazine-toxicity associated drug discontinuation at 5-years =1 – \nS0(t=5)exp(Xβ)  where S0(t=5) is the baseline survival function at 5-years of follow-up and \nβX is the linear predictor, β1x1+ β2x2+ … + βpxp. \n \n . CC-BY 4.0 International licenseIt is made available under a \nperpetuity. \n is the author/funder, who has granted medRxiv a license to display the preprint in(which was not certified by peer review)preprint \nThe copyright holder for thisthis version posted December 18, 2023. ; https://doi.org/10.1101/2023.12.15.23299947doi: medRxiv preprint \n\n 14 \nModel internal validation and shrinkage: The performance of the model in terms of \ncalibration (where 1.00 is the ideal) was assessed by plotting agreement between \npredicted and observed outcomes. Internal validation was performed to correct \nperformance estimates for optimism due to overfitting by bootstrapping with \nreplacement 500 samples of the development data . The full model was fitted in each \nbootstrap sample and then its performance was quantified in the  bootstrap sample \n(apparent performance) and the original sample (test model performance ), and the  \noptimism calculated (difference in test performance and apparent performance) .A \nuniform shrinkage factor was estimated as the average of calibration slopes from the \nbootstrap samples. This process was repeated for all 10 imputed datasets, and the \nfinal uniform shrinkage calculated by averaging across the estimated shrinkage \nestimates from each imputation. Optimism-adjusted estimates of performance for the \noriginal model were then calculated, as the original apparent performance minus the \noptimism. \n \nTo account for overfitting during model development process, the original β \ncoefficients were multiplied by the final uniform shrinkage factor and the baseline \nhazards re-estimated conditional on the shrunken β coefficients to ensure that overall \ncalibration was maintained, producing a final model . The D statistic, a measure of \ndiscrimination, interpreted as a log hazard ratio (HR), the exponential of which gives \nthe HR comparing two groups defined by above/below the median of the linear \npredictor was calculated 36 37.R2, a measure of variation explained by the model was \ncalculated. \n \n . CC-BY 4.0 International licenseIt is made available under a \nperpetuity. \n is the author/funder, who has granted medRxiv a license to display the preprint in(which was not certified by peer review)preprint \nThe copyright holder for thisthis version posted December 18, 2023. ; https://doi.org/10.1101/2023.12.15.23299947doi: medRxiv preprint \n\n 15 \nModel external validation: External validation of the final model was performed using \ndata from CPRD Gold. The final developed model equation was applied to the \nvalidation dataset, and calibration and discrimination were examined using the same \nmeasures as above  36 37 .Calibration of 5-year risks was examined by plotting \nagreement between estimated risk from the model and observed outcome risks. In the \ncalibration plot , predicted and observed risks were divided into 10 equally sized \ngroups. Additionally, pseudo-observations were used to construct smooth calibration \ncurves across all individuals via a running non-parametric smoother. Separate graphs \nwere plotted for each imputation of the validation cohort and an example of one plot is \nshown in the results . Subgroup analyses considered age -group and inflammatory \ndisease type (RA vs. others). Stata-MP version 16 was used for all statistical analyses. \nThis study was reported in line with the transparent reporting of a multivariate \nprediction model for individual prediction or diagnosis guidelines 38. \n \n  \n . CC-BY 4.0 International licenseIt is made available under a \nperpetuity. \n is the author/funder, who has granted medRxiv a license to display the preprint in(which was not certified by peer review)preprint \nThe copyright holder for thisthis version posted December 18, 2023. ; https://doi.org/10.1101/2023.12.15.23299947doi: medRxiv preprint \n\n 16 \nResults Study participants: Data for 8,936 and 5,203 participants contributing 23,299 \nand 12,867 person -years follow -up were included in the derivation and validation \ncohorts, respectively (Supplementary figure S1 and S2) . Most participants in both \ncohorts were diagnosed with RA, were female, and had similar prevalence of lifestyle \nfactors, comorbidities and drug treatments (Table 1) . Nine candidate predictors ( 21 \nparameters) were included in the model (Table 2). \n \nModel development: In the derivation dataset, 473  outcome events occurred during \nthe follow-up period at a rate (95% CI) of 20.30 (18.55 - 22.22) per 1,000 person -\nyears. Of these, 256 , 131, and 113  patients respectively stopped treatment due to \ncytopenia, renal function decline , and elevated liver enzymes . Outcome validation \nexercise in 178 outcomes revealed that only 4.5% outcomes (n=8) could potentially \nbe explained by another contemporaneous illness or its treatments, with a positive \npredictive value of 95.5% (Table S1). \nThese events occurred throughout the 5-year follow-up period when the entire cohort \nwas considered (Figure S 3) and when patients co -prescribed either methotrexate or \nleflunomide or thiopurine with sulfasalazine were excluded (Figure S4). CKD-stage 3, \ndiabetes (either type 1 or 2) , co-prescription of methotrexate, co-prescription of \nleflunomide, and either cytopenia or elevated liver enzymes during first six months of  \nsulfasalazine prescription were strong predictors of drug discontinuation with adjusted \nHR hazard ratio (95% CI) 1.96 (1.47 -2.62), 1.34 (1.01 -1.78), 1.39 (1.15 -1.68), 2.05 \n(1.09-3.86) and 2.80 (2.29-3.42) respectively (Table 2). From the bootstrap, a uniform \nshrinkage factor of 0.84 was obtained and used to shrink predictor coefficients in the \nfinal model for optimism and after re-estimation, the final model’s cumulative baseline \nsurvival function (S0) was 0.940 at 5-years of follow-up (Box 1). \n . CC-BY 4.0 International licenseIt is made available under a \nperpetuity. \n is the author/funder, who has granted medRxiv a license to display the preprint in(which was not certified by peer review)preprint \nThe copyright holder for thisthis version posted December 18, 2023. ; https://doi.org/10.1101/2023.12.15.23299947doi: medRxiv preprint \n\n 17 \nModel performance in the d evelopment cohort: As expected, the calibration slope \n(95% CI) in the development data was1.00 (95% CI 0.85-1.15). Calibration plot of the \nfinal (i.e., after shrinkage) model at 5-years showed that the average model predictions \nmatched the average observed outcome  probabilities across 10 groups of patients, \nwith confidence intervals overlapping the 45 -degree line (perfect prediction line)  \n(Figure 1). As most patients had a low risk of outcome (Figure S5), most of the deciles \nclustered at the bottom left of the calibration plot (Figure S6). The smoothed calibration \ncurve at 5 -years showed alignment of observed risk to the predicted risk with wide \nconfidence intervals at high -risk probabilities (Figure 1). The Royston D statistic was \n0.91 (95% CI 0.77 – 1.05), corresponding to a HR (95% CI) of 2.48 (2.16 -2.86) \ncomparing the risk of participants who were above the median of linear predictor to \nthat below the median . The optimism adjusted Royston D statistic was 0.79 , \ncorresponding to a HR of 2.20 (Table 3).     \n \nModel performance in the validation cohort: There were 280 outcomes at a rate (95% \nCI) of 21.76 (19.36-24.47)/1000 person-years in the validation cohort. The calibration \nslope (95% CI) across the 5-year follow-up period was 1.19 (0.96-1.43) (Figure 2). The \ncalibration plot showed reasonable correspondence between observed and predicted \nrisk at 5-years across the tenths of risk (Figure S7). Most of the deciles clustered at \nthe bottom left of the calibration plot  due to a low risk of outcome for most patients \n(Figures S7, S8). When individual risks were plotted, the smoothed calibration curve \nshowed alignment of the predicted risk to the observed risk  at low risk  and wide \nconfidence intervals overlapping the perfect prediction line at high -risk probabilities \n(Figure 2). Model performance was also tested at years 1, 2, 3 and 4 (Figure S9-S12) \nand showed a similar pattern except for over-prediction of risk at 1 year. The Royston \n . CC-BY 4.0 International licenseIt is made available under a \nperpetuity. \n is the author/funder, who has granted medRxiv a license to display the preprint in(which was not certified by peer review)preprint \nThe copyright holder for thisthis version posted December 18, 2023. ; https://doi.org/10.1101/2023.12.15.23299947doi: medRxiv preprint \n\n 18 \nD statistic in the validation data was 0.87 (0.67,1.07), corresponding to a HR (95% CI) \nof 2.39 (1.95 -2.92). Model discrimination in the derivation and validation data was \nbroadly similar (Table 3). The model performed well in those younger or older than 60 \nyears, in those with RA or other conditions (Figure S13, S14). \nWorked examples: Ten anonymised patient profiles, one from the middle of each of \nthe 10 groups defined by deciles of predicted risk were selected from the development \ncohort, the higher the decile group the higher the risk, and the risk equation was \napplied to each . The cumulative probability of outcome over five years ranged from \n5.3% in the middle of the first group to 9.3% in the middle of the seventh group, and \n19.0% in the middle of the 10th group (Table S2). \n  \n . CC-BY 4.0 International licenseIt is made available under a \nperpetuity. \n is the author/funder, who has granted medRxiv a license to display the preprint in(which was not certified by peer review)preprint \nThe copyright holder for thisthis version posted December 18, 2023. ; https://doi.org/10.1101/2023.12.15.23299947doi: medRxiv preprint \n\n 19 \nDiscussion We have developed and externally validated a prognostic model for \nsulfasalazine discontinuation due to abnormal blood -test results. To the best of our \nknowledge this is the first such risk-prediction model. It performed well in predicting \noutcomes by five  years and in clinically relevant subgroups  defined by age and \ninflammatory condition . Previous studies have variably reported  NAT-2 acetylator \nstatus to be associated  with sulfasalazine toxicity 15 39 . However, these studies \nevaluated all side -effects and did not separately assess  either myelotoxicity, \nhepatotoxicity, or nephrotoxicity as evaluated in the current study. \n \nOur findings suggest that a one size fits all approach to monitoring for blood, liver, or \nrenal toxicity using three monthly blood-tests during long-term sulfasalazine treatment \nas recommended in the SmPC and the ACR guidelines, and not monitoring for these \nafter the first year of treatment as recommended in the BSR guidelines are  both \ninappropriate because there is a large interindividual variation in the risk of developing \nthese side-effects. The large variation in risk implies that it may be reasonable to not \nmonitor some patien ts after the first year of sulfasalazine treatment, while others at \nhigher risk of side-effects are monitored frequently e.g., three-monthly. It is important \nto realise that DILI can be idiosyncratic and annual testing is unlikely to detect them \nearly enough to improve patient outcome. It is beyond our remit to propose threshold \nat which the frequency of monitoring blood -tests should be altered . These decisions \nare best taken by guideline writing groups. Thus, our findings ought to be considered \nby guideline writing groups. \n \nIt is important that the results of this study are not used to risk -stratify monitoring in \npatients newly started on sulfasalazine because our prognosis model used data from \n . CC-BY 4.0 International licenseIt is made available under a \nperpetuity. \n is the author/funder, who has granted medRxiv a license to display the preprint in(which was not certified by peer review)preprint \nThe copyright holder for thisthis version posted December 18, 2023. ; https://doi.org/10.1101/2023.12.15.23299947doi: medRxiv preprint \n\n 20 \npatients prescribed sulfasalazine by their GP for six months after initiating treatment \nand dose-escalation in a hospital outpatient . It typically takes three to six months to \nstabilise a patients’ sulfasalazine dose before prescription and monitoring is handed \nover to the GP. In healthcare systems where such shared care arrangements do not \nexist, this strategy may be applied after one year of sulfasalazine treatment. Although \ngenerally perceived to be safe, sulfasalazine use carries a risk of myelotoxicity and \nhepatotoxicity comparable to that observed with methotrexate in people with RA40. \n \nCKD stage-3, diabetes, and concomitant methotrexate or leflunomide therapy were \nstrong independent predictors of sulfasalazine discontinuation with abnormal \nmonitoring blood-test results in this study. These associations may be due to reduced \nsulfasalazine clearance in CKD and DILI being associated with diabetes41. Abnormal \nblood-test results during the first six -months of therapy were strong independent \npredictors of discontinuing sulfasalazine with abnormal monitoring blood-test results, \nlike findings for methotrexate and leflunomide 24 42 . Elevated liver enzymes and \ncytopenia before starting treatment have previously been associated with abnormal \nblood-test results in patients treated with methotrexate and biologics respectively43-49. \n \nThere are several strengths of this study . First, we used a large real -world and \nnationally representative dataset for model development and a  similar independent \ndataset for external validation. Second, the study population included patients with a \nrange of diseases and the results have broad generalisability . Third, the prognostic \nfactors were selected by an expert multidisciplinary team based on clinical experience. \nFourth, our outcome required the abnormal blood -test result to be associated with \nsulfasalazine discontinuation, thus, allowing the model to predict clinically relevant \n . CC-BY 4.0 International licenseIt is made available under a \nperpetuity. \n is the author/funder, who has granted medRxiv a license to display the preprint in(which was not certified by peer review)preprint \nThe copyright holder for thisthis version posted December 18, 2023. ; https://doi.org/10.1101/2023.12.15.23299947doi: medRxiv preprint \n\n 21 \noutcomes. Fifth, the prognostic model is easy to use in practice , and can be easily  \nbuilt into GP electronic health records. \n \nHowever, several limitations of this study ought to be considered . First, we did not \nhave access to the date when the patient was first prescribed sulfasalazine in the \nhospital clinic. Second, we did not have data on concurrent use of biologics  as these \nare hospital prescribed . However, there is no evidence to suggest that biologics \nincrease sulfasalazine toxicity. Third, we did not have data on disease activity as these \nare not recorded in the CPRD . Fourth, the abnormal blood test could be due to a \ndifferent illness and not due to sulfasalazine. However, in our previous validation \nstudies on methotrexate, only 5.4% of abnormal blood-test results could be explained \nby an alternative illness 25. Fifth, although the external validation dataset was distinct \nfrom the model development dataset, it also originated from UK general practice. We \nrecommend therefore that our model be validated in a dataset from another country. \nSixth, there were 31 (0.3%)  patients in the highest three risk groups defined according \nto tenths of risk, resulting in uncertainty regarding predictors for these groups . \nSeventh, we did not perform competing risk regression. However, this does not limit \nthe validity of our findings as there were few deaths (28 [0.3%])  in the derivation cohort \nand 8 (0.2%) deaths in validation cohort up to 5-year follow-up period. \n \nIn conclusion, we have developed and externally validated a prognostic model for \nsulfasalazine discontinuation with abnormal monitoring blood -test results . These \nfindings need to be considered by national and international specialist societies’ \nguideline writing groups to decide upon risk-stratified frequency of monitoring blood -\ntests during long-term sulfasalazine treatment. \n . CC-BY 4.0 International licenseIt is made available under a \nperpetuity. \n is the author/funder, who has granted medRxiv a license to display the preprint in(which was not certified by peer review)preprint \nThe copyright holder for thisthis version posted December 18, 2023. ; https://doi.org/10.1101/2023.12.15.23299947doi: medRxiv preprint \n\n 22 \nEthics statements \nEthical approval \nNot required as this study is based on secondary data analysis. \nData availability statement \nData used in the study are from the Clinical Practice Research Datalink.Study protocol \nis available from www.cprd.com . \nFootnotes \nFunding This research was funded by National Institute for Health and Care Research \n(NIHR) grants NIHR130580 .The funders had no role in conducting and/or reporting \nthis study. \nContributor and guarantor information : GN, MJG, HCW, TC, MWT, GPA, CPF, \nCDM, MDS, RDR, and AA designed the study .GN analysed the data supervised by \nMJG, RDR and AA.GN, MJG, HCW, TC, MWT, GPA, CPF, CDM, MDS, RDR, and AA \ninterpreted the data .AA drafted the manuscript .All authors critically evaluated and \nrevised the manuscript.The corresponding author attests that all listed authors meet \nauthorship criteria and that no others meeting the criteria have been omitted.AA is the \nguarantor. \n \n \n \n . CC-BY 4.0 International licenseIt is made available under a \nperpetuity. \n is the author/funder, who has granted medRxiv a license to display the preprint in(which was not certified by peer review)preprint \nThe copyright holder for thisthis version posted December 18, 2023. ; https://doi.org/10.1101/2023.12.15.23299947doi: medRxiv preprint \n\n 23 \nTable 1: Distribution of candidate predictors in development and validation cohorts \nPredictor1  Development cohort \n(CPRD Aurum) \nn=8,936 \nValidation cohort \n(CPRD Gold) \nn=5,203 \nAge, mean (SD) year 55.3 (14.8) 55.5 (14.8) \nFemale sex 5,535 (61.9) 3,240 (62.3) \nBody Mass Index    \n<18.5 kg/m2 138 (1.5) 88 (1.7) \n18.5-24.9 kg/m2 2,441 (27.3) 1,428 (27.5) \n25.0-29.9 kg/m2 2,840 (31.8) 1,678 (32.3) \n≥30 kg/m2 2,714 (30.4) 1,626 (31.3) \nMissing  803 (9.0) 383 (7.4) \nAlcohol use    \nNon-user 1,705 (19.1) 805 (15.5) \nLow (1-14 units/week) 3,854 (43.1) 2,859 (55.0) \nModerate (15-21 units/week) 535 (6.0) 251 (4.8) \nHazardous (>21 units/week) 667 (7.5) 273 (5.3) \nEx-user 996 (11.2) 359 ((6.9) \nMissing 1,179 (13.2) 656 (12.6) \nInflammatory conditions   \n  Rheumatoid arthritis  6,945 (77.7) 4,067 (78.2) \nPsoriatic arthritis 1,354 (15.2) 773 (14.9) \nInflammatory bowel disease 319 (3.6) 173 (3.3) \nAnkylosing spondylitis/reactive \narthritis \n318 (3.6) 190 (3.7) \nComorbidities    \nDiabetes  982 (11.0) 519 (10.0) \nChronic kidney disease stage-3  613 (6.9) 333 (6.4) \nImmunosuppressive drugs    \nMethotrexate 2,999 (33.6) 1,785 (34.3) \nLeflunomide 109 (1.2) 78 (1.5) \nAzathioprine/mercaptopurine 73 (0.8) 41 (0.8) \nOther drugs    \nStatins  2,088 (23.4) 1,130 (21.7) \nCarbamazepine/valproate 103 (1.2) 37 (0.7) \nParacetamol  1,445 (16.2) 884 (17.0) \nAt least mild cytopenia or liver \nenzyme elevation in six-months \npreceding start of follow-up \n1,264 (14.2) 753 (14.5) \n1Values are numbers (percentage) unless stated otherwise. \n  \n . CC-BY 4.0 International licenseIt is made available under a \nperpetuity. \n is the author/funder, who has granted medRxiv a license to display the preprint in(which was not certified by peer review)preprint \nThe copyright holder for thisthis version posted December 18, 2023. ; https://doi.org/10.1101/2023.12.15.23299947doi: medRxiv preprint \n\n 24 \n \nTable 2: Final model hazard ratios and β-coefficients \nc Adjusted HR \n(95% CI) \nCoefficients \nAge, mean (SD) year 1.01 (1.00,1.02) .0076439  \nFemale sex 1.08 (0.88, 1.31) .0741336 \nBody Mass Index  0.98 (0.97,1.00) -.0168035 \nAlcohol use    \nNon-user Reference   \nLow (1-14 units/week) 1.02 (0.80, 1.29) .0182851 \nModerate (15-21 units/week) 0.64 (0.38,1.06) -.4507257 \nHazardous (>21 units/week) 0.87 (0.58,1.33) -.133557 \nEx-user 0.94 (0.67,1.32) -.0651469 \nInflammatory conditions   \n  Rheumatoid arthritis Reference   \nPsoriatic arthritis 1.03 (0.78,1.36) .0316689 \nInflammatory bowel disease 0.74 (0.38,1.44) -.305206 \nAnkylosing spondylitis/reactive arthritis 1.25 (0.74,2.12) .2214547 \nComorbidities    \nDiabetes  1.34 (1.01, 1.78) .2909969 \nChronic kidney disease stage-3  1.96 (1.47,2.62) .671859 \nImmunosuppressive drugs    \nMethotrexate 1.39 (1.15,1.68) .3315573  \nLeflunomide 2.05 (1.09,3.86) .7164324 \nAzathioprine/mercaptopurine 1.24 (0.37,4.17) .2189764 \nOther drugs    \nStatins  0.98 (0.78,1.24) -.0181917  \nCarbamazepine/valproate 0.74 (0.28,2.00) -.2949835  \nParacetamol  1.14 (0.90,1.43) .1272515  \nBlood-test abnormalities   \nAt least mild cytopenia or liver enzyme \nelevation in six-months preceding start \nof follow-up \n2.80 (2.29,3.42) 1.029245  \n \n1HR: hazard ratio, CI: confidence interval.The reported values are before shrinkage. \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n . CC-BY 4.0 International licenseIt is made available under a \nperpetuity. \n is the author/funder, who has granted medRxiv a license to display the preprint in(which was not certified by peer review)preprint \nThe copyright holder for thisthis version posted December 18, 2023. ; https://doi.org/10.1101/2023.12.15.23299947doi: medRxiv preprint \n\n 25 \nTable 3: Model diagnostics \nMeasure Apparent \nperformance* \nTest \nperformance§ \nAverage \noptimism¥ \nOptimism \ncorrected \nperformance† \nExternal \nvalidation \n(CPRD \nAurum)‡ \nOverall calibration \nslope  \n1.00 \n(0.85,1.15) \n0.84 \n(0.70,0.98) \n0.16 0.84 \n(0.69,0.99) \n1.19 \n(0.96,1.43) \nR2D  0.17 \n(0.12,0.21) \n0.15 \n(0.11,0.19) \n0.04 0.13 \n(0.08,0.17) \n0.15 \n(0.10,0.21) \nRoyston D statistic 0.91 \n(0.77,1.05) \n0.85 \n(0.72,0.99) \n0.12 0.79 \n(0.65,0.93) \n0.87 \n(0.67,1.07) \n \n*Refers to performance (95% CI) estimated directly from the data that was used to \ndevelop the model. \n§ Determined by executing full model in each bootstrap sample (500 samples with \nreplacement), calculating bootstrap performance, and applying same model in \noriginal sample. \n¥ Average difference between model performance in bootstrap data and test \nperformance in original dataset \n†Subtracting average optimism from apparent performance.   \n‡ Penalised model was externally validated (Penalised calibration slope:1.19; 95% CI \n1.01, 1.37)  \n CPRD: Clinical Practice Research Datalink  \n \nBox 1: Equation to predict the risk of sulfasalazine discontinuation after six months \nof primary care prescription and within the next 5-years. \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \nAll variables are code 0, and 1 if absent or present respectively, except for BMI and \nage that were continuous variables.0.940 is the baseline survival function at 5-years, \n0.84 is the shrinkage factor and the other numbers are the estimated regression \nRisk score = 1 – 0.940 exp(0.84βX), where βX=(.0076439 * Age in years at first primary-care \nprescription  + .0741336*female-sex - .0168035*BMI + .0182851*low alcohol intake  - \n.4507257*moderate alcohol intake - .1335573*hazardous alcohol intake - .0651469*Ex-alcohol \nintake + .0316689*Psoriasis - .305206*IBD + .2214547*ankylosing spondylitis/reactive arthritis \n+ .2909969*diabetes + .671859*CKD  + .3315573*MTX + .7164324* LEF + .2189764*AZA or \n6-MP - .0181917*statins - .2949835 *Carbamazepine/valproate +  .1272515*paracetamol  + \n1.029245* At-least mild cytopenia or liver enzyme elevation within six-months of primary care \nAZA/6-MP prescription. \n . CC-BY 4.0 International licenseIt is made available under a \nperpetuity. \n is the author/funder, who has granted medRxiv a license to display the preprint in(which was not certified by peer review)preprint \nThe copyright holder for thisthis version posted December 18, 2023. ; https://doi.org/10.1101/2023.12.15.23299947doi: medRxiv preprint \n\n 26 \ncoefficients for the predictors, which indicate their mutually adjusted relative \ncontribution to the outcome risk. \n \nFigure 1: Calibration of a prognostic model for SSZ discontinuation with abnormal \nmonitoring blood-test results at 5 years in the development cohort. \n \nData from a single imputed dataset was used; So(t=5) 0.940 \n \n \n \n \n \n \n \n \n \n . CC-BY 4.0 International licenseIt is made available under a \nperpetuity. \n is the author/funder, who has granted medRxiv a license to display the preprint in(which was not certified by peer review)preprint \nThe copyright holder for thisthis version posted December 18, 2023. ; https://doi.org/10.1101/2023.12.15.23299947doi: medRxiv preprint \n\n 27 \nFigure 2: Calibration of a prognostic model for SSZ discontinuation with abnormal \nmonitoring blood-test results at 5 years in the validation cohort. \n \nData from a single imputed dataset was used; So(t=5) 0.940 \n \n \n \n  \n . CC-BY 4.0 International licenseIt is made available under a \nperpetuity. \n is the author/funder, who has granted medRxiv a license to display the preprint in(which was not certified by peer review)preprint \nThe copyright holder for thisthis version posted December 18, 2023. ; https://doi.org/10.1101/2023.12.15.23299947doi: medRxiv preprint \n\n 28 \nReferences  \n1.Ogdie A, Coates LC, Gladman DD.Treatment guidelines in psoriatic \narthritis.Rheumatology (Oxford) 2020;59(Suppl 1):i37-i46.doi: \n10.1093/rheumatology/kez383 [published Online First: 2020/03/12] \n2.Singh JA.Treatment Guidelines in Rheumatoid Arthritis.Rheum Dis Clin North Am \n2022;48(3):679-89.doi: 10.1016/j.rdc.2022.03.005 [published Online First: \n2022/08/12] \n3.Ransford RA, Langman MJ.Sulphasalazine and mesalazine: serious adverse reactions \nre-evaluated on the basis of suspected adverse reaction reports to the Committee \non Safety of Medicines.Gut 2002;51(4):536-9.doi: 10.1136/gut.51.4.536 \n[published Online First: 2002/09/18] \n4.Farr M, Tunn EJ, Symmons DP, et al.Sulphasalazine in rheumatoid arthritis: \nhaematological problems and changes in haematological indices associated with \ntherapy.Br J Rheumatol 1989;28(2):134-8.doi: 10.1093/rheumatology/28.2.134 \n[published Online First: 1989/04/01] \n5.MCCONKEY B.Ten Years of Sulphasalazine Use in Rheumatoid Arthritis.Rheumatology \n1989;28(2):175-76.doi: 10.1093/rheumatology/28.2.175 \n6.van Riel PLCM, van Gestel AM, van de Putte LBA.Long-Term Usage and Side-Effect \nProfile of Sulphasalazine in Rheumatoid Arthritis.Rheumatology \n1995;XXXIV(suppl_4):40-42.doi: 10.1093/rheumatology/XXXIV.suppl_4.40 \n7.de Abajo FJ, Montero D, Madurga M, et al.Acute and clinically relevant drug-induced \nliver injury: a population based case-control study.Br J Clin Pharmacol \n2004;58(1):71-80.doi: 10.1111/j.1365-2125.2004.02133.x [published Online \nFirst: 2004/06/23] \n8.McConkey B, Amos RS, Durham S, et al.Sulphasalazine in rheumatoid arthritis.Br Med \nJ 1980;280(6212):442-4.doi: 10.1136/bmj.280.6212.442 [published Online First: \n1980/02/16] \n9.Farr M, Scott DGI, Bacon PA.Side Effect Profile of 200 Patients with Inflammatory \nArthritides Treated with Sulphasalazine.Drugs 1986;32(1):49-53.doi: \n10.2165/00003495-198600321-00010 \n10.Amos RS, Pullar T, Bax DE, et al.Sulphasalazine for rheumatoid arthritis: toxicity in \n774 patients monitored for one to 11 years.Br Med J (Clin Res Ed) \n1986;293(6544):420-3.doi: 10.1136/bmj.293.6544.420 [published Online First: \n1986/08/16] \n11.MacGilchrist AJ, Hunter JA.Sulphasalazine hepatotoxicity: lack of a hypersensitivity \nresponse.Ann Rheum Dis 1986;45(11):967-8.doi: 10.1136/ard.45.11.967 \n[published Online First: 1986/11/01] \n12.Marabani M, Madhok R, Capell HA, et al.Leucopenia during sulphasalazine treatment \nfor rheumatoid arthritis.Ann Rheum Dis 1989;48(6):505-7.doi: \n10.1136/ard.48.6.505 [published Online First: 1989/06/01] \n13.Jobanputra P, Amarasena R, Maggs F, et al.Hepatotoxicity associated with \nsulfasalazine in inflammatory arthritis: A case series from a local surveillance of \nserious adverse events.BMC Musculoskelet Disord 2008;9:48.doi: 10.1186/1471-\n2474-9-48 [published Online First: 2008/04/15] \n14.Keisu M, Ekman E.Sulfasalazine associated agranulocytosis in Sweden 1972 –\n1989.European Journal of Clinical Pharmacology 1992;43(3):215-18.doi: \n10.1007/BF02333012 \n15.Wiese MD, Alotaibi N, O'Doherty C, et al.Pharmacogenomics of NAT2 and ABCG2 \ninfluence the toxicity and efficacy of sulphasalazine containing DMARD regimens \nin early rheumatoid arthritis.The Pharmacogenomics Journal 2014;14(4):350-\n55.doi: 10.1038/tpj.2013.45 \n16.Grove ML, Hassell AB, Hay EM, et al.Adverse reactions to disease-modifying anti-\nrheumatic drugs in clinical practice.Qjm 2001;94(6):309-19.doi: \n10.1093/qjmed/94.6.309 [published Online First: 2001/06/08] \n17.Committee JF.British National Formulary.[85].London: BMJ Group and Pharmaceutical \nPress; [2023]. \n . CC-BY 4.0 International licenseIt is made available under a \nperpetuity. \n is the author/funder, who has granted medRxiv a license to display the preprint in(which was not certified by peer review)preprint \nThe copyright holder for thisthis version posted December 18, 2023. ; https://doi.org/10.1101/2023.12.15.23299947doi: medRxiv preprint \n\n 29 \n18.Ledingham J, Gullick N, Irving K, et al.BSR and BHPR guideline for the prescription \nand monitoring of non-biologic disease-modifying anti-rheumatic \ndrugs.Rheumatology (Oxford) 2017;56(6):865-68.doi: \n10.1093/rheumatology/kew479 [published Online First: 2017/03/25] \n19.Singh JA, Saag KG, Bridges SL, Jr., et al.2015 American College of Rheumatology \nGuideline for the Treatment of Rheumatoid Arthritis.Arthritis Care Res (Hoboken) \n2016;68(1):1-25.doi: 10.1002/acr.22783 [published Online First: 2015/11/08] \n20.Pfizer.Salazopyrin tablets.Healthcare Professionals (SmPC) 2021 [updated \n24/12/2021.2021:[Available from: \nhttps://www.medicines.org.uk/emc/product/3838/smpc#gref accessed \n23/07/2023 2023. \n21.Herrett E, Gallagher AM, Bhaskaran K, et al.Data Resource Profile: Clinical Practice \nResearch Datalink (CPRD).International journal of epidemiology 2015;44(3):827-\n36.doi: 10.1093/ije/dyv098 [published Online First: 2015/06/08] \n22.Wolf A, Dedman D, Campbell J, et al.Data resource profile: Clinical Practice Research \nDatalink (CPRD) Aurum.International journal of epidemiology 2019;48(6):1740-\n40g.doi: 10.1093/ije/dyz034 [published Online First: 2019/03/13] \n23.Abhishek A, Doherty M, Kuo CF, et al.Rheumatoid arthritis is getting less frequent-\nresults of a nationwide population-based cohort study.Rheumatology (Oxford) \n2017;56(5):736-44.doi: 10.1093/rheumatology/kew468 [published Online First: \n2017/01/09] \n24.Nakafero G, Grainge MJ, Williams HC, et al.Risk stratified monitoring for \nmethotrexate toxicity in immune mediated inflammatory diseases: prognostic \nmodel development and validation using primary care data from the UK.Bmj \n2023;381:e074678.doi: 10.1136/bmj-2022-074678 [published Online First: \n2023/05/31] \n25.Nakafero G, Grainge MJ, Card T, et al.What is the incidence of methotrexate or \nleflunomide discontinuation related to cytopenia, liver enzyme elevation or kidney \nfunction decline? Rheumatology 2021;60(12):5785-94.doi: \n10.1093/rheumatology/keab254 \n26.Khwaja A.KDIGO clinical practice guidelines for acute kidney injury.Nephron Clin Pract \n2012;120(4):c179-84.doi: 10.1159/000339789 [published Online First: \n2012/08/15] \n27.Chalasani N, Björnsson E.Risk factors for idiosyncratic drug-induced liver \ninjury.Gastroenterology 2010;138(7):2246-59.doi: 10.1053/j.gastro.2010.04.001 \n[published Online First: 2010/04/17] \n28.Safy-Khan M, de Hair MJH, Welsing PMJ, et al.Current Smoking Negatively Affects the \nResponse to Methotrexate in Rheumatoid Arthritis in a Dose-responsive Way, \nIndependently of Concomitant Prednisone Use.J Rheumatol 2021;48(10):1504-\n07.doi: 10.3899/jrheum.200213 [published Online First: 2021/02/03] \n29.Klotz U.Clinical Pharmacokinetics of Sulphasalazine, Its Metabolites and Other \nProdrugs of 5-Aminosalicylic Acid.Clinical Pharmacokinetics 1985;10(4):285-\n302.doi: 10.2165/00003088-198510040-00001 \n30.Meijer B, Wilhelm AJ, Mulder CJJ, et al.Pharmacology of Thiopurine Therapy in \nInflammatory Bowel Disease and Complete Blood Cell Count Outcomes: A 5-Year \nDatabase Study.Ther Drug Monit 2017;39(4):399-405.doi: \n10.1097/ftd.0000000000000414 [published Online First: 2017/05/11] \n31.Dirven L, Klarenbeek NB, van den Broek M, et al.Risk of alanine transferase (ALT) \nelevation in patients with rheumatoid arthritis treated with methotrexate in a \nDAS-steered strategy.Clin Rheumatol 2013;32(5):585-90.doi: 10.1007/s10067-\n012-2136-8 [published Online First: 2012/12/12] \n32.Riley RD, Ensor J, Snell KIE, et al.Calculating the sample size required for developing \na clinical prediction model.BMJ 2020;368:m441.doi: 10.1136/bmj.m441 \n33.Riley RD, Debray TPA, Collins GS, et al.Minimum sample size for external validation of \na clinical prediction model with a binary outcome.Statistics in medicine \n2021;40(19):4230-51.doi: https://doi.org/10.1002/sim.9025 \n . CC-BY 4.0 International licenseIt is made available under a \nperpetuity. \n is the author/funder, who has granted medRxiv a license to display the preprint in(which was not certified by peer review)preprint \nThe copyright holder for thisthis version posted December 18, 2023. ; https://doi.org/10.1101/2023.12.15.23299947doi: medRxiv preprint \n\n 30 \n34.Schafer JL.Multiple imputation: a primer.Stat Methods Med Res 1999;8(1):3-15.doi: \n10.1177/096228029900800102 [published Online First: 1999/05/29] \n35.Steyerberg EW.Clinical Prediction Models : A Practical Approach to Development, \nValidation, and Updating.Cham, SWITZERLAND: Springer International Publishing \nAG 2019. \n36.Royston P, Altman DG.External validation of a Cox prognostic model: principles and \nmethods.BMC Medical Research Methodology 2013;13(1):33.doi: 10.1186/1471-\n2288-13-33 \n37.Cox DR.Note on Grouping.Journal of the American Statistical Association \n1957;52(280):543-47.doi: 10.1080/01621459.1957.10501411 \n38.Moons KG, Altman DG, Reitsma JB, et al.Transparent Reporting of a multivariable \nprediction model for Individual Prognosis or Diagnosis (TRIPOD): explanation and \nelaboration.Annals of internal medicine 2015;162(1):W1-73.doi: 10.7326/m14-\n0698 [published Online First: 2015/01/07] \n39.Ricart E, Taylor WR, Loftus EV, et al.N-acetyltransferase 1 and 2 genotypes do not \npredict response or toxicity to treatment with mesalamine and sulfasalazine in \npatients with ulcerative colitis.Am J Gastroenterol 2002;97(7):1763-8.doi: \n10.1111/j.1572-0241.2002.05838.x [published Online First: 2002/07/24] \n40.Mielnik P, Sexton J, Fagerli KM, et al.Discontinuation rate of sulfasalazine, leflunomide \nand methotrexate due to adverse events in a real-life setting (NOR-\nDMARD).Rheumatol Adv Pract 2023;7(2):rkad053.doi: 10.1093/rap/rkad053 \n[published Online First: 2023/07/11] \n41.Mori S, Arima N, Ito M, et al.Non-alcoholic steatohepatitis-like pattern in liver biopsy \nof rheumatoid arthritis patients with persistent transaminitis during low-dose \nmethotrexate treatment.PLoS One 2018;13(8):e0203084.doi: \n10.1371/journal.pone.0203084 [published Online First: 2018/08/25] \n42.Nakafero G, Grainge MJ, Card T, et al.Development and validation of a prognostic \nmodel for leflunomide discontinuation with abnormal blood tests during long -term \ntreatment: cohort study using data from the Clinical Practice Research Datalink \nGold and Aurum.Rheumatology (Oxford) 2022;61(7):2783-91.doi: \n10.1093/rheumatology/keab790 [published Online First: 2021/11/01] \n43.Dirven L, Klarenbeek NB, van den Broek M, et al.Risk of alanine transferase (ALT) \nelevation in patients with rheumatoid arthritis treated with methotrexate in a \nDAS-steered strategy.Clinical rheumatology 2013;32(5):585-90.doi: \nhttps://dx.doi.org/10.1007/s10067-012-2136-8 \n44.Cavalli M, Eriksson N, Sundbaum JK, et al.Genome-wide association study of liver \nenzyme elevation in an extended cohort of rheumatoid arthritis patients starting \nlow-dose methotrexate.Pharmacogenomics 2022;23(15):813-20.doi: \n10.2217/pgs-2022-0074 \n45.Sherbini AA, Gwinnutt JM, Hyrich KL, et al.Rates and predictors of methotrexate-\nrelated adverse events in patients with early rheumatoid arthritis: results from a \nnationwide UK study.Rheumatology (Oxford, England) 2022;61(10):3930-38. \n46.Suzuki Y, Hirose T, Sugiyama N, et al.Post-marketing surveillance of high-dose \nmethotrexate (>8 mg/week) in Japanese patients with rheumatoid arthritis: A \npost hoc sub-analysis of patients according to duration of prior methotrexate \nuse.Modern Rheumatology 2021;31(3):575-86. \n47.Verstappen SM, Bakker MF, Heurkens AH, et al.Adverse events and factors associated \nwith toxicity in patients with early rheumatoid arthritis treated with methotrexate \ntight control therapy: the CAMERA study.Annals of the rheumatic diseases \n2010;69(6):1044‐48.doi: 10.1136/ard.2008.106617 \n48.Schmajuk G, Miao Y, Yazdany J, et al.Identification of risk factors for elevated \ntransaminases in methotrexate users through an electronic health record.Arthritis \nCare and Research 2014;66(8):1159-66.doi: \nhttp://dx.doi.org/10.1002/acr.22294 \n49.Hastings R, Ding T, Butt S, et al.Neutropenia in patients receiving anti-tumor necrosis \nfactor therapy.Arthritis Care Res (Hoboken) 2010;62(6):764-9.doi: \n10.1002/acr.20037 [published Online First: 2010/06/11] \n . CC-BY 4.0 International licenseIt is made available under a \nperpetuity. \n is the author/funder, who has granted medRxiv a license to display the preprint in(which was not certified by peer review)preprint \nThe copyright holder for thisthis version posted December 18, 2023. ; https://doi.org/10.1101/2023.12.15.23299947doi: medRxiv preprint \n\n 31 \n \n . CC-BY 4.0 International licenseIt is made available under a \nperpetuity. \n is the author/funder, who has granted medRxiv a license to display the preprint in(which was not certified by peer review)preprint \nThe copyright holder for thisthis version posted December 18, 2023. ; https://doi.org/10.1101/2023.12.15.23299947doi: medRxiv preprint","source_license":"CC-BY-4.0","license_restricted":false}