==== Front JAMA Netw Open JAMA Netw Open JAMA Network Open 2574-3805 American Medical Association 37382956 10.1001/jamanetworkopen.2023.20851 zoi230618 Research Original Investigation Online Only Rheumatology Comparative Effectiveness of Adalimumab vs Tofacitinib in Patients With Rheumatoid Arthritis in Australia Adalimumab vs Tofacitinib in Patients With Rheumatoid Arthritis Adalimumab vs Tofacitinib in Patients With Rheumatoid Arthritis Deakin Claire T. PhD 1 2 3 De Stavola Bianca L. PhD 4 Littlejohn Geoffrey MBBS 1 5 Griffiths Hedley MBBS 1 6 Ciciriello Sabina MBBS PhD 1 7 Youssef Peter MBBS PhD 1 8 9 Mathers David MBBS 1 10 Bird Paul MBBS PhD 1 11 Smith Tegan PhD 1 O’Sullivan Catherine PhD 1 Freeman Tim MSc 12 Segelov Dana MSc 12 Hoffman David MBBS 12 Seaman Shaun R. PhD 13 for the OPAL Rheumatology Network 1 OPAL Rheumatology Ltd, Sydney, New South Wales, Australia 2 Centre for Adolescent Rheumatology Versus Arthritis at University College London, University College London Hospitals, Great Ormond Street Hospital and University College London, London, United Kingdom 3 National Institute of Health Research Biomedical Centre at Great Ormond Street Hospital, London, United Kingdom 4 Population, Policy and Practice Research and Teaching Department, UCL Great Ormond Street Institute of Child Health, London, United Kingdom 5 Department of Medicine, Monash University, Clayton, Victoria, Australia 6 Barwon Rheumatology Service, Geelong, Victoria, Australia 7 Royal Melbourne Hospital, Melbourne, Victoria, Australia 8 Royal Prince Alfred Hospital, Sydney, New South Wales, Australia 9 University of Sydney, Sydney, New South Wales, Australia 10 Georgetown Arthritis, Newcastle, New South Wales, Australia 11 University of New South Wales, Kensington, New South Wales, Australia 12 Software for Specialists Pty Ltd, Sydney, New South Wales, Australia 13 MRC Biostatistics Unit, University of Cambridge, Cambridge, United Kingdom Article Information Accepted for Publication: May 10, 2023. Published: June 29, 2023. doi:10.1001/jamanetworkopen.2023.20851 Open Access: This is an open access article distributed under the terms of the CC-BY License. © 2023 Deakin CT et al. JAMA Network Open. Corresponding Author: Claire T. Deakin, PhD, OPAL Rheumatology Ltd (claire.deakin@opalrheumatology.com.au). Author Contributions: Dr Deakin had full access to all of the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis. Concept and design: Deakin, De Stavola, Ciciriello, Youssef, Bird, Smith, O’Sullivan, Segelov, Hoffman, Seaman. Acquisition, analysis, or interpretation of data: Deakin, Littlejohn, Griffiths, Youssef, Mathers, Bird, Smith, O’Sullivan, Freeman, Segelov. Drafting of the manuscript: Deakin, De Stavola, Bird, O’Sullivan, Freeman, Segelov, Seaman. Critical revision of the manuscript for important intellectual content: Deakin, De Stavola, Littlejohn, Griffiths, Ciciriello, Youssef, Mathers, Bird, Smith, O’Sullivan, Segelov, Hoffman, Seaman. Statistical analysis: Deakin, De Stavola, Freeman, Seaman. Obtained funding: Smith, Segelov. Administrative, technical, or material support: Mathers, Bird, Smith, O’Sullivan, Freeman, Hoffman. Supervision: Littlejohn, Griffiths, Youssef, Bird, Smith, Segelov. Conflict of Interest Disclosures: Dr Griffiths reported serving on the advisory board for AbbVie, Eli Lilly, Organon, and Amgen. Dr Youssef reported receiving personal fees for consulting or serving on the speakers bureau from Pfizer, AbbVie, Eli Lilly, Celltrion, Pfizer, and Sandoz during the conduct of the study. Dr Bird reported serving on the advisory board for AbbVie, Pfizer, Novartis, and Bristol-Myers Squibb and consulting or serving on the speakers bureau for Janssen, Pfizer, Novartis, Gilead, Eli Lilly, GlaxoSmithKline, Synarc, and Boston Imaging Core Lab outside the submitted work. Ms Segelov reported being the director of Software4Specialists Pty Ltd. No other disclosures were reported. Group Information: The OPAL Rheumatology Network members are listed in Supplement 2. Meeting Presentation: This paper was presented at the EULAR 2022; June 1, 2022; Copenhagen, Denmark. Data Sharing Statement: See Supplement 3. Additional Contributions: We thank the members of OPAL Rheumatology Ltd and their patients for participating in this study and Software4Specialists Pty Ltd for providing the Audit4 platform and deidentifying and aggregating the clinical data. Additional Information: Software4Specialists has a financial arrangement with OPAL for software and data management service provided outside the submitted work. 29 6 2023 6 2023 29 6 2023 6 6 e232085113 1 2023 10 5 2023 Copyright 2023 Deakin CT et al. JAMA Network Open. https://creativecommons.org/licenses/by/4.0/ This is an open access article distributed under the terms of the CC-BY License. jamanetwopen-e2320851.pdf This comparative effectiveness reseach study assesses the effecetiveness of adalimumab vs tofacitinib for treatment of rheumatoid arthiritis in the clinical setting. Key Points Question What is the effectiveness of adalimumab (ADA) compared with tofacitinib (TOF) for treatment of rheumatoid arthritis in routine clinical practice? Findings In this comparative effectiveness study of 842 patients with rheumatoid arthritis in Australia, TOF was favored slightly at 3 months vs ADA, but there was no difference in scores between patients receiving TOF and those receiving ADA at 9 months. Meaning This study showed similar treatment effects for TOF and ADA, which is consistent with data from a randomized trial and current European Alliance of Associations for Rheumatology treatment guidelines. Importance There is a need for observational studies to supplement evidence from clinical trials, and the target trial emulation (TTE) framework can help avoid biases that can be introduced when treatments are compared crudely using observational data by applying design principles for randomized clinical trials. Adalimumab (ADA) and tofacitinib (TOF) were shown to be equivalent in patients with rheumatoid arthritis (RA) in a randomized clinical trial, but to our knowledge, these drugs have not been compared head-to-head using routinely collected clinical data and the TTE framework. Objective To emulate a randomized clinical trial comparing ADA vs TOF in patients with RA who were new users of a biologic or targeted synthetic disease-modifying antirheumatic drug (b/tsDMARD). Design, Setting, and Participants This comparative effectiveness study emulating a randomized clinical trial of ADA vs TOF included Australian adults aged 18 years or older with RA in the Optimising Patient Outcomes in Australian Rheumatology (OPAL) data set. Patients were included if they initiated ADA or TOF between October 1, 2015, and April 1, 2021; were new b/tsDMARD users; and had at least 1 component of the disease activity score in 28 joints using C-reactive protein (DAS28-CRP) recorded at baseline or during follow-up. Intervention Treatment with either ADA (40 mg every 14 days) or TOF (10 mg daily). Main Outcomes and Measures The main outcome was the estimated average treatment effect, defined as the difference in mean DAS28-CRP among patients receiving TOF compared with those receiving ADA at 3 and 9 months after initiating treatment. Missing DAS28-CRP data were multiply imputed. Stable balancing weights were used to account for nonrandomized treatment assignment. Results A total of 842 patients were identified, including 569 treated with ADA (387 [68.0%] female; median age, 56 years [IQR, 47-66 years]) and 273 treated with TOF (201 [73.6%] female; median age, 59 years [IQR, 51-68 years]). After applying stable balancing weights, mean DAS28-CRP in the ADA group was 5.3 (95% CI, 5.2-5.4) at baseline, 2.6 (95% CI, 2.5-2.7) at 3 months, and 2.3 (95% CI, 2.2-2.4) at 9 months; in the TOF group, it was 5.3 (95% CI, 5.2-5.4) at baseline, 2.4 (95% CI, 2.2-2.5) at 3 months, and 2.3 (95% CI, 2.1-2.4) at 9 months. The estimated average treatment effect was −0.2 (95% CI, −0.4 to −0.03; P = .02) at 3 months and −0.03 (95% CI, −0.2 to 0.1; P = .60) at 9 months. Conclusions and Relevance In this study, there was a modest but statistically significant reduction in DAS28-CRP at 3 months for patients receiving TOF compared with those receiving ADA and no difference between treatment groups at 9 months. Three months of treatment with either drug led to clinically relevant average reductions in mean DAS28-CRP, consistent with remission. ==== Body pmcIntroduction In the past 20 years, the availability of tumor necrosis factor inhibitors (TNFis) and other biologic disease-modifying antirheumatic drugs (bDMARDs) has transformed treatment for patients with rheumatoid arthritis (RA). More recently, targeted synthetic DMARDs (tsDMARDs), including janus kinase inhibitors (JAKis), have become available and are considered equivalent to bDMARDs for patients with moderate to severe disease refractory to conventional synthetic DMARD (csDMARD) therapy.1 Although trials directly comparing specific b/tsDMARDs head-to-head are limited, a double-blind phase 3b/4 randomized clinical trial2 showed that tofacitinib (TOF) combined with methotrexate was noninferior to adalimumab (ADA) combined with methotrexate. Other trials have demonstrated the efficacy of JAKi therapy in patients who have not responded to methotrexate or TNFi therapy.3,4 Drug retention was longer for TOF compared with TNFis in an observational study of 4023 treatment courses that combined multiple lines of therapy.5 Similar outcomes have been described for TOF and JAKi therapy compared with TNFi therapy.6,7 To our knowledge, there are no observational studies evaluating the effectiveness of a JAKi drug compared with a TNFi drug in a b/tsDMARD-naive patient population. Registries and real-world data sets (RWDs), which include routinely collected data such as electronic medical records (EMRs) and medical claims data, are a valuable source of information for understanding the effectiveness of treatments. There is increasing recognition of the complementary role for real-world evidence based on analyses of RWDs in health care and regulatory decision-making.8 However, there are significant challenges to the reliability of comparative effectiveness studies using RWDs.9 In this study, RWD refers to clinical records in patients’ EMRs that are routinely collected at the point of care. Outcomes are often recorded incompletely in registries and EMRs, patients are not randomized to treatment groups, and differences between groups need to be accounted for. There can also be differential durations of follow-up and attrition. Target trial emulation (TTE) is a framework for comparative effectiveness analyses whereby principles from the design of randomized clinical trials (RCTs) are applied to observational research by making explicit the design of the trial that is intended to be emulated.10 In this study, we sought to emulate an RCT of ADA vs TOF in b/tsDMARD-naive patients with RA in the Optimising Patient Outcomes in Australian Rheumatology (OPAL) data set using an intention-to-treat analysis.11 To generate evidence from a large observational RWD, we developed a methodological approach to address the challenges of missing baseline and outcome data and nonrandomized treatment assignment. Our approach aimed to avoid the selection bias that could result from excluding patients with missing outcomes. Methods Detailed technical methods of this comparative effectiveness study are described in eMethods 1 through 7 and the eAppendix in Supplement 1. Ethics approval was obtained for research using deidentified data in the OPAL data set from the University of New South Wales Human Research Ethics Committee (HC17799) and for the specific protocol (HC210647). Patients consented to their deidentified EMR data being used for research via an opt-out consent model. This report followed the International Society for Pharmacoeconomics and Outcomes Research (ISPOR) reporting guideline. Design The prespecified protocol for the target trial to be emulated is described in eMethods 1 in Supplement 1. In brief, the analysis was designed to emulate an RCT of ADA vs TOF in patients with RA who were new users of a b/tsDMARD using an intention-to-treat analysis. Mean disease activity score in 28 joints using C-reactive protein (DAS28-CRP) was assessed at 3 and 9 months after treatment with ADA or TOF was initiated. Participants Eligible patients in the emulated target trial were adults aged 18 years or older who were diagnosed with RA; whose first visit occurred between April 1, 2015, and January 1, 2021; who had no prior recorded b/tsDMARD; who had at least 6 months from their first recorded visit until baseline; and who had at least 6 months of treatment with a csDMARD immediately prior to baseline. These criteria defined (or enriched for) a cohort of new users who were b/tsDMARD naive based on their EMR and the government criteria for b/tsDMARD reimbursement. Patients were excluded if they did not have at least 1 component of the DAS28-CRP recorded at baseline, 3 months, or 9 months. Interventions Patients initiated treatment with either ADA (40 mg every 14 days) or TOF (10 mg daily) and, in the target trial, would be expected to continue treatment during follow-up unless an adverse event or contraindication occurred. The limited duration of availability of the biosimilar for ADA meant that all ADA interventions were the originator and not the biosimilar. End Point The primary outcome was disease activity at 3 and 9 months after initiating treatment. The average treatment effect (ATE) was defined as the difference in mean DAS28-CRP among patients receiving TOF compared with those receiving ADA at 3 and 9 months. These time points were selected because joint counts and pathologic markers are assessed for government reimbursement at 3 and 9 months. American College of Rheumatology response of at least 50% was used in the trial of ADA and TOF.2 However, its calculation requires multiple variables that are missing in the OPAL data set. Therefore, DAS28-CRP, used in other landmark trials for RA,12,13 was selected as a composite outcome that represents clinician- and patient-assessed disease and an objective pathologic marker. Setting and Data Source This analysis used the multicenter OPAL data set, which includes the EMRs for 216 138 patients with rheumatic conditions treated by 112 rheumatologists across Australia at 43 different clinics since 2004. In Australia, government reimbursement is available for b/tsDMARDs for patients with moderate to severe disease who have not responded to at least 6 months of treatment with csDMARDs, and physicians can prescribe the b/tsDMARD that fits the patient’s clinical need. As such, there are no binding guidelines as to the order in which a b/tsDMARD class can be prescribed. Response to treatment is assessed at 3 months and at 6-month intervals thereafter. Tofacitinib is approved by the Australian Therapeutic Goods Administration for use for RA at a dosage of 5 mg twice daily. The deidentified data in the OPAL data set include demographics, disease history, disease activity measures, comorbidities, pathology, medications, patient-reported outcomes, and characteristics of the treating rheumatologist. Data were collected from April 1, 2015, to June 30, 2021. A more detailed description of the data set and a summary of the characteristics of the variables used in this analysis are given in eMethods 2 in Supplement 1. Safety Treatment cessations due to an adverse reaction were described for all patients who satisfied the inclusion criteria. Treating physicians have discretion to record an adverse reaction to a medication in the EMR, and there may have been unrecorded adverse reactions. All recorded adverse reactions that were considered more serious than nonserious were described. An adverse reaction recorded using the International Statistical Classification of Diseases and Related Health Problems, Tenth Revision (ICD-10) codes C* or D00-D48 was considered to be cancer.14 A major cardiovascular event (MACE) was defined according to previously published ICD-10 codes for MACE in the context of administrative data sets and rheumatology: I21-I24 (myocardial infarction); I63-I66 (stroke); I11, I50, and I97.1 (heart failure); and Z95 (coronary artery bypass grafting).15,16 Patient Involvement Patients were not involved in the design of this study or consulted during selection of outcomes or the interpretation of findings. Patients will be involved in the dissemination of this research. Statistical Analysis A multistep method was developed to address the challenges of missing disease activity data and nonrandomized treatment assignment. Analysis that only uses complete cases could lead to selection bias, and thus, multiple imputation was used instead of complete-case analysis. An overview of the methods is presented in Figure 1, and the details are fully described in eMethods 1 through 7 and the eAppendix in Supplement 1. Figure 1. Multistep Analytical Procedure Developed to Estimate the Average Treatment Effect (ATE) of Tofacitinib (TOF) Compared With Adalimumab (ADA) Random forest multiple imputation (RF-MI) was used to impute plausible values for missing data in the original electronic medical record (EMR) data set. Stable balancing weights (SBWs) were used to balance the baseline characteristics of the ADA and TOF treatment groups. Gray squares in the grid that represents EMR indicate complete data items and white squares, missing data items. In the complete data sets generated by RF-MI, colored squares represent the imputed values. The different colors used for the same data item in different data sets indicate that imputed values were slightly different in each data set. The covariate balance plots indicate that after SBWs were applied, the mean standardized difference in baseline characteristics between the treatment groups was close to 0 compared with before weighting. The ATE was then calculated separately in each imputed data set before pooling to generate a final estimate. In brief, multiple imputation by chained equations using the random forest algorithm under the missing-at-random assumption was used to impute missing data for the components of the DAS28-CRP at baseline and follow-up, generating 10 imputed data sets.17 Patients with no observed DAS28-CRP components at follow-up were then excluded after multiple imputation because these patients would not be informative about the treatment effect. Stable balancing weights (SBWs) were then used to account for differences in baseline characteristics between the included patients receiving ADA and the total eligible patients and between the included patients receiving TOF and the total eligible patients.18 Balance for the baseline characteristics was evaluated, including calculating standardized mean differences between the treatment groups, which express these differences in terms of the observed SDs in the sample of eligible patients. The difference in weighted DAS28-CRP was then calculated at 3 and 9 months within each imputed data set, and these estimates were pooled using Rubin rules to yield single point estimates at 3 and 9 months.19 The whole procedure was then bootstrapped using 1000 bootstrap samples to generate a 95% CI for the estimates using the percentile method.20 All analyses were performed using R, version 4.0.2 (R Project for Statistical Computing). Differences in mean DAS28-CRP at follow-up were assessed using a 2-sided test and a significance level of P < .05. eMethods 1 through 7, eResults 1 through 4, and the eAppendix in Supplement 1 give further details of the analysis, including the R code and checks of the performance of the multiple imputation algorithm and the balance between the treatment groups. Results Patients Of the 52 338 patients with RA in the OPAL data set, 842 eligible b/tsDMARD-naive patients who were new starters of ADA (n = 569; 387 [68.0%] female; 175 [30.8%] male; median age, 56 years [IQR, 47-66 years]) or TOF (n = 273; 201 [73.6%] female; 72 [26.4%] male; median age, 59 years [IQR, 51-68 years]) were identified (Figure 2). Patients without any components of DAS28-CRP recorded at baseline, 3 months, or 9 months who were otherwise eligible new starters were excluded (n = 339). The baseline characteristics of the excluded patients are described in eMethods 2 in Supplement 1. Figure 2. Flow Diagram of Inclusion and Exclusion of Patients with Rheumatoid Arthritis (RA) in the Optimising Patient Outcomes in Australian Rheumatology Data Set Stable balancing weights were used to make each treatment group balanced with each other and the 842 total eligible patients. ADA indicates adalimumab; TOF, tofacitinib. aMultiple imputation was applied to these patients. Baseline Data Before SBWs were applied, there were small differences between the treatment groups (Table). For example, patients in the TOF group were slightly older and more likely to be female, to be located in states other than Victoria, to be located in outer regional or remote areas, and to have prior kidney disease. The TOF group also included a lower percentage of patients treated at a clinic with a nurse and a higher percentage of patients treated by rheumatologists with more years of experience, a higher overall tendency to prescribe b/tsDMARDs in their practice, and a lower overall tendency to complete the patient global score. There were small differences between the treatment groups in the csDMARD drugs that had been prescribed to patients prior to baseline. Table. Baseline Characteristics of Patients With RA Treated With ADA or TOF Characteristic Patients with RAa Received ADA (n = 569) Received TOF (n = 273) Age, median (IQR) [range] 56 (47-66) [18-88] 59 (51-68) [21-86] Gender, No./total No. (%)b Female 387/562 (68.0) 201/273 (73.6) Male 175/562 (30.8) 72/273 (26.4) State, No./total No. (%)c Victoria 288/568 (50.6) 73/272 (26.7) New South Wales 149/568 (26.2) 108/272 (39.6) Australian Capital Territory 42/568 (7.4) 25/272 (9.2) Tasmania 20/568 (3.5) 22/272 (8.1) Queensland and Western Australia 69/568 (12.1) 44/272 (16.1) Regional location, No./total No. (%)d Major cities 355/568 (62.4) 165/273 (60.4) Inner regional 166/568 (29.2) 68/273 (24.9) Outer regional and remote 47/568 (8.3) 40/273 (14.7) Disease duration recorded in EMR, median (IQR) [range], y 1.0 (0.7-1.7) [0.5-5.3] 1.2 (0.8-2) [0.5-5.6] Year of treatment start 2015 and 2016 64 (11.2) 29 (10.6) 2017 110 (19.3) 60 (22.0) 2018 104 (18.3) 73 (26.7) 2019 118 (20.7) 65 (23.8) 2020 141 (24.8) 39 (14.3) 2021 32 (5.6) 7 (2.6) DAS28-CRP components, median (IQR) [range] SJC28e 13 (6-22) [0-28] 12 (5-22) [0-28] TJC28e 14 (7-22) [0-28] 12 (6-22) [0-28] Patient global scoref 60 (46.5-75) [0-100] 51 (30.5-74.2) [0-100] CRP levelg 6 (3-14) [1-221] 6 (3-14.7) [1.5-169] Physician global score, median (IQR) [range]h 50 (40-70) [0-100] 50 (30-70) [3-100] ESR, median (IQR) [range]i 10 (5-22.2) [1-106] 11 (5-28) [1-107] DAS28-CRP, median (IQR) [range]i 5.7 (4.7-6.4) [1.2-7.8] 5.4 (4.2-6.4) [1.4-8.1] Nurse at clinic Yes 284 (49.9) 74 (27.1) No 285 (50.1) 199 (72.9) Practitioner gender Female 230 (40.4) 117 (42.9) Male 339 (59.6) 156 (57.1) Practitioner experience, y 0-15 187 (32.9) 74 (27.1) 16-30 226 (39.7) 142 (52) >30 156 (27.4) 57 (20.9) Practitioner overall tendency to complete patient global score, % 0-75 219 (38.5) 126 (46.2) 76-100 350 (61.5) 147 (53.8) Practitioner overall tendency to prescribe b/tsDMARDs, % total prescriptions 0-25 494 (86.8) 216 (79.1) 26-100 75 (13.2) 57 (20.9) Prior kidney disease Yes 229 (40.2) 130 (47.6) No 340 (59.8) 143 (52.4) Prior treatment Methotrexate Yes 355 (62.4) 184 (67.4) No 214 (37.6) 89 (32.6) Hydroxychloroquine Yes 204 (35.9) 130 (47.6) No 365 (64.1) 143 (52.4) Leflunomide Yes 161 (28.3) 95 (34.8) No 408 (71.7) 178 (65.2) Sulfasalazine Yes 158 (27.8) 74 (27.1) No 411 (72.2) 199 (72.9) Oral corticosteroids Yes 324 (56.9) 159 (58.2) No 245 (43.1) 114 (41.8) Concomitant treatment Methotrexate Yes 151 (26.5) 79 (28.9) No 418 (73.5) 194 (71.1) Hydroxychloroquine Yes 133 (23.4) 69 (25.3) No 436 (76.6) 204 (74.7) Leflunomide Yes 101 (17.8) 49 (17.9) No 468 (82.2) 224 (82.1) Sulfasalazine Yes 89 (15.6) 39 (14.3) No 480 (84.4) 234 (85.7) Oral corticosteroids Yes 282 (49.6) 131 (48.0) No 287 (50.4) 142 (52.0) Abbreviations: ADA, adalimumab; b/tsDMARDs, biologic or targeted synthetic disease-modifying antirheumatic drugs; CRP, C-reactive protein; DAS28-CRP, disease activity score in 28 joints using C-reactive protein; EMR, electronic medical record; ESR, erythrocyte sedimentation rate; RA, rheumatoid arthritis; SJC28, swollen joint count based on 28-joint assessment; TJC28, tender joint count based on 28-joint assessment; TOF, tofacitinib. a Data are presented as number (percentage) of patients unless otherwise indicated. b Data were missing for 7 patients in the ADA group (1.2%) and 0 patients in the TOF group. c Data were missing for 1 patient in the ADA group (0.2%) and 1 patient in the TOF group (0.4%). d Data were missing for 1 patient in the ADA group (0.2%) and 0 patients in the TOF group. e Data were missing for 68 patients in the ADA group (12.0%) and 40 in the TOF group (14.7%). f Data were missing for 213 patients in the ADA group (37.4%) and 115 in the TOF group (42.1%). g Data were missing for 75 patients in the ADA group (13.2%) and 45 in the TOF group (16.5%). h Data were missing for 210 patients in the ADA group (36.9%) and 118 in the TOF group (43.2%). i Data were missing for 217 patients in the ADA group (38.1%) and 120 in the TOF group (44.0%). After SBWs were applied, the standardized mean differences in baseline characteristics were between −0.03 and 0.03, within the conventional threshold of 0.1 for propensity score matching (Figure 3). This indicated that the treatment groups were balanced on these measured characteristics using the SBW method. Figure 3. Standardized Mean Difference in Baseline Characteristics of Adalimumab (ADA) and Tofacitinib (TOF) Treatment Groups Before and After Weighting Error bars for disease activity score in 28 joints using C-reactive protein (DAS28-CRP) represent the minimum and maximum standardized mean difference values across 10 imputed data sets. The standardized mean difference is the difference between treatment groups in the mean for each covariate divided by its SD for the entire sample. The vertical dashed black lines indicate standardized mean differences of −0.03 and 0.03. ACT indicates Australian Capital Territory; NSW, New South Wales; QLD, Queensland and Western Australia; and TAS, Tasmania. Estimated Comparative Effectiveness After weighting, mean DAS28-CRP decreased from 5.3 (95% CI, 5.2-5.4) at baseline to 2.6 (95% CI, 2.5-2.7) at 3 months and 2.3 (95% CI, 2.2-2.4) at 9 months in the ADA group (eResults 5 in Supplement 1). Mean DAS28-CRP decreased from 5.3 (95% CI, 5.2-5.4) at baseline to 2.4 (95% CI, 2.2-2.5) at 3 months and 2.3 (95% CI, 2.1-2.4) at 9 months in the TOF group. These follow-up values are consistent with the threshold for remission (≤2.6).21 The ATE for TOF compared with ADA was −0.2 (95% CI, −0.4 to −0.03; P = .02) at 3 months and −0.03 (95% CI, −0.2 to 0.1; P = .60) at 9 months (Figure 4). This indicated that patients who received TOF had, on average, slightly lower DAS28-CRP at 3 months compared with patients who received ADA, but there was no difference in DAS28-CRP between the treatment groups at 9 months. Figure 4. Estimated Average Treatment Effect (ATE) for Tofacitinib (TOF) Compared With Adalimumab (ADA) at 3 and 9 Months Squares represent ATEs, with horizontal lines representing 95% CIs based on bootstrap distributions of the estimates. Safety Due to recent concerns about MACE and cancer associated with JAKi drugs,22 any cessations of ADA or TOF due to an adverse event are described in eResults 6 in Supplement 1 for all eligible patients (n = 842) and patients who had been excluded due to missing data on the components of the DAS28-CRP (n = 326). For these 1168 patients, there were 28 recorded treatment cessations due to adverse reactions in the ADA group (3.8%) and 19 in the TOF group (4.4%). Median follow-up in the ADA group was 2.1 years (IQR, 0.9-3.4 years) and in the TOF group was 2.5 years (IQR, 1.4-3.4 years). There were no adverse reactions recorded using ICD-10 codes for cancer and 1 (0.2%) for MACE (a medically significant case of embolic stroke in the TOF group). Additionally, there was 1 case of pulmonary embolism in each of the ADA (0.1%) and TOF (0.2%) treatment groups, the former of which was life-threatening but considered unrelated to treatment. There was 1 case of deep vein thrombosis in the TOF group. Discussion In this comparative effectiveness study using the TTE framework, we found a modest but statistically significant reduction in disease activity associated with TOF compared with ADA at 3 months and no difference between drugs at 9 months in patients with RA who were b/tsDMARD naive. These results may be generalizable to patients with RA who have not responded to csDMARD therapy and are eligible to initiate their first b/tsDMARD, subject to the limitations of the study. Although the outcomes, time points, and superiority design differ, the small effect size in our analysis is consistent with the previous findings of noninferiority in the American College of Rheumatology response of at least 50% at 6 months for these drugs in a clinical trial.2 Our findings are also consistent with an observational study that combined multiple lines of therapy and showed slightly longer drug retention for TOF therapy compared with TNFi therapy.5 Our study benefitted from the large number of patients with RA in the OPAL data set, which enabled the analysis to focus on the effectiveness of TOF vs ADA received as the first b/tsDMARD. Our findings support the 2019 European Alliance of Associations for Rheumatology (EULAR) recommendations, which treat JAKis and bDMARDs as equivalent when used as first-line therapy.1 Observational studies are an important supplement to RCTs for assessing whether trial findings can be reproduced in everyday practice in a less restricted patient population and for guiding treatment decisions that occur outside the idealized setting of a trial.23 Given the well-known limitations of observational research, the need for quality in comparative effectiveness studies has been recognized.9 Although there have been few applications of the TTE framework in rheumatology,24,25,26 there have been numerous studies using propensity score methods, including propensity score matching and inverse probability of treatment weighting,6,27,28,29,30,31,32,33,34 and some of these are limited by time-related biases, such as comparing drugs or classes received as different lines of therapy.35 Covariate balancing is an alternative approach to propensity score methods and can lead to better balance while avoiding some of the known pitfalls of propensity score methods, such as model misspecification and dispersed weights.36,37 Additionally, this study dealt with the challenge of missing outcomes data, which is a source of bias not often addressed in observational studies. Safety signals, although typically rare, are important in connection with ADA and TOF, especially in light of the relatively higher risks of MACE and cancer associated with JAKi therapy compared with TNFi therapy.22 Although the rates of MACE and other cardiovascular events in our study are consistent with those previously reported, this study was not set up to address safety, and we only described recorded adverse reactions in patients who received ADA or TOF as first-line b/tsDMARD therapy. Follow-up was limited, and physicians’ discretion to record adverse reactions limit the conclusions that can be drawn from these data. As more data on these risks become available, the safety of these drugs will be better understood.38 Limitations As an analysis of observational data, the challenges of missing outcomes data and nonrandomized treatment assignment were possible limitations to this study that may impact interpretation of the generalizability of the results. Our analysis relied on some assumptions, and violation of these would limit the reliability of the results. We assumed the joint counts, CRP levels, and patient global scores were missing at random (ie, whether an outcome was missing was not related to its value after conditioning on the observed data). Under this assumption, the missing data on joint counts, CRP levels, and patient global scores could be accounted for in the imputation model by the characteristics of the patients, the treating rheumatologists, and the clinics, which were fully observed, and accounting for these variables could produce unbiased results in the analysis. Availability of nursing staff time and other characteristics of the clinics and individual rheumatologists are plausible variables to explain whether a rheumatologist in a busy clinic would be able to prioritize recording complete data on all outcomes during a consultation with a patient. Nevertheless, we acknowledge that there may have been bias that could not be addressed if there were further unmeasured sources of missing data that were not accounted for in the imputation model. The assumptions for inferring a causal treatment effect include that (1) the probability of initiating treatment with ADA or TOF may have depended on the measured baseline characteristics but did not depend on future disease activity at follow-up time points, (2) the probability of treatment assignment did not equal 0, and (3) a patient's observed DAS28-CRP at follow-up was the same as that patient's potential DAS28-CRP at follow-up after they followed the treatment that they were observed to be assigned to.39,40 It is possible that assignment to ADA or TOF was confounded by unmeasured characteristics that were not accounted for in this analysis. E-values for the ATE suggest that these results may be sensitive to unmeasured confounding (eResults 7 in Supplement 1). Although these assumptions cannot be verified, there was consistency between our analysis and other comparisons of ADA or TNFi therapy with TOF.2,5 This analysis emulated an intention-to-treat effect and not a per-protocol effect. Future work may address possible effects of patients stopping treatment between the 3- and 9-month time points. Conclusions In this comparative effectiveness study, DAS28-CRP was significantly lower at 3 months for patients treated with TOF compared with ADA. However, 3 months of treatment with either drug led to substantive reductions in mean DAS28-CRP, consistent with remission. There was no difference in DAS28-CRP between patients receiving TOF or ADA at 9 months. The results of this observational study are consistent with clinical trial data1,2 and support the current EULAR treatment guidelines. The analysis serves as an exemplar of the TTE framework applied to an incomplete RWD. Supplement 1. eMethods 1. Protocol for the Target Trial to be Emulated eMethods 2. Description of the OPAL Dataset eMethods 3. Directed Acyclic Diagram eMethods 4. Overview of the Analytical Procedure eMethods 5. Multiple Imputation eMethods 6. Stable Balancing Weights eMethods 7. Bootstrapping Procedure eResults 1. Performance of Multiple Imputation eResults 2. Properties of Stable Balancing Weights eResults 3. Balance Checks eResults 4. Weighted Components of the DAS28CRP eResults 5. Point Estimates eResults 6. Drug Cessations Due to Adverse Reactions eResults 7. E-Values to Quantify Sensitivity of Results to Possible Unmeasured Confounding eAppendix. Code eReferences Click here for additional data file. Supplement 2. OPAL Rheumatology Network Click here for additional data file. Supplement 3. Data Sharing Statement Click here for additional data file. ==== Refs References 1 Smolen JS, Landewé RBM, Bijlsma JWJ, . EULAR recommendations for the management of rheumatoid arthritis with synthetic and biological disease-modifying antirheumatic drugs: 2019 update. Ann Rheum Dis. 2020;79 (6 ):685-699. doi:10.1136/annrheumdis-2019-216655 31969328 2 Fleischmann R, Mysler E, Hall S, ; ORAL Strategy Investigators. Efficacy and safety of tofacitinib monotherapy, tofacitinib with methotrexate, and adalimumab with methotrexate in patients with rheumatoid arthritis (ORAL strategy): a phase 3b/4, double-blind, head-to-head, randomised controlled trial. Lancet. 2017;390 (10093 ):457-468. doi:10.1016/S0140-6736(17)31618-5 28629665 3 Keystone EC, Taylor PC, Drescher E, . Safety and efficacy of baricitinib at 24 weeks in patients with rheumatoid arthritis who have had an inadequate response to methotrexate. Ann Rheum Dis. 2015;74 (2 ):333-340. doi:10.1136/annrheumdis-2014-206478 25431052 4 Fleischmann R, Pangan AL, Song IH, . Upadacitinib versus placebo or adalimumab in patients with rheumatoid arthritis and an inadequate response to methotrexate: results of a phase iii, double-blind, randomized controlled trial. Arthritis Rheumatol. 2019;71 (11 ):1788-1800. doi:10.1002/art.41032 31287230 5 Finckh A, Tellenbach C, Herzog L, ; Physicians and Patients of the SCQM. Comparative effectiveness of antitumour necrosis factor agents, biologics with an alternative mode of action and tofacitinib in an observational cohort of patients with rheumatoid arthritis in Switzerland. RMD Open. 2020;6 (1 ):e001174. doi:10.1136/rmdopen-2020-001174 32385143 6 Reed GW, Gerber RA, Shan Y, . Real-world comparative effectiveness of tofacitinib and tumor necrosis factor inhibitors as monotherapy and combination therapy for treatment of rheumatoid arthritis. Rheumatol Ther. 2019;6 (4 ):573-586. doi:10.1007/s40744-019-00177-4 31707603 7 Lauper K, Iudici M, Mongin D, . Effectiveness of TNF-inhibitors, abatacept, IL6-inhibitors and JAK-inhibitors in 31 846 patients with rheumatoid arthritis in 19 registers from the ‘JAK-pot’ collaboration. Ann Rheum Dis. 2022;81 (10 ):1358-1366. doi:10.1136/annrheumdis-2022-222586 35705376 8 Arlett P, Kjaer J, Broich K, Cooke E. Real-world evidence in EU medicines regulation: enabling use and establishing value. Clin Pharmacol Ther. 2022;111 (1 ):21-23. doi:10.1002/cpt.2479 34797920 9 Courvoisier DS, Lauper K, Kedra J, . EULAR points to consider when analysing and reporting comparative effectiveness research using observational data in rheumatology. Ann Rheum Dis. 2022;81 (6 ):780-785. doi:10.1136/annrheumdis-2021-221307 35058229 10 Hernán MA, Robins JM. Using big data to emulate a target trial when a randomized trial is not available. Am J Epidemiol. 2016;183 (8 ):758-764. doi:10.1093/aje/kwv254 26994063 11 Littlejohn GO, Tymms KE, Smith T, Griffiths HT. Using big data from real-world Australian rheumatology encounters to enhance clinical care and research. Clin Exp Rheumatol. 2020;38 (5 ):874-880.31820719 12 Rubbert-Roth A, Enejosa J, Pangan AL, . Trial of upadacitinib or abatacept in rheumatoid arthritis. N Engl J Med. 2020;383 (16 ):1511-1521. doi:10.1056/NEJMoa2008250 33053283 13 Gabay C, Emery P, van Vollenhoven R, ; ADACTA Study Investigators. Tocilizumab monotherapy versus adalimumab monotherapy for treatment of rheumatoid arthritis (ADACTA): a randomised, double-blind, controlled phase 4 trial. Lancet. 2013;381 (9877 ):1541-1550. doi:10.1016/S0140-6736(13)60250-0 23515142 14 World Health Organization. ICD-10: International Statistical Classification of Diseases and Related Health Problems. World Health Organization; 2004. 15 Bosco E, Hsueh L, McConeghy KW, Gravenstein S, Saade E. Major adverse cardiovascular event definitions used in observational analysis of administrative databases: a systematic review. BMC Med Res Methodol. 2021;21 (1 ):241. doi:10.1186/s12874-021-01440-5 34742250 16 de Thurah A, Andersen IT, Tinggaard AB, . Risk of major adverse cardiovascular events among patients with rheumatoid arthritis after initial CT-based diagnosis and treatment. RMD Open. 2020;6 (1 ):e001113. doi:10.1136/rmdopen-2019-001113 31958282 17 van Buuren S, Groothuis-Oudshoorn K. mice: Multivariate imputation by chained equations in R. J Stat Softw. 2011;45 :1-67. doi:10.18637/jss.v045.i03 18 Zubizarreta JR. Stable weights that balance covariates for estimation with incomplete outcome data. J Am Stat Assoc. 2015;110 :910-922. doi:10.1080/01621459.2015.1023805 19 Rubin DB. Multiple Imputation for Nonresponse in Surveys. Wiley-Interscience; 1987. doi:10.1002/9780470316696 20 Bartlett JW, Hughes RA. Bootstrap inference for multiple imputation under uncongeniality and misspecification. Stat Methods Med Res. 2020;29 (12 ):3533-3546. doi:10.1177/0962280220932189 32605503 21 Wells G, Becker JC, Teng J, . Validation of the 28-joint Disease Activity Score (DAS28) and European League Against Rheumatism response criteria based on C-reactive protein against disease progression in patients with rheumatoid arthritis, and comparison with the DAS28 based on erythrocyte sedimentation rate. Ann Rheum Dis. 2009;68 (6 ):954-960. doi:10.1136/ard.2007.084459 18490431 22 Ytterberg SR, Bhatt DL, Mikuls TR, ; ORAL Surveillance Investigators. Cardiovascular and cancer risk with tofacitinib in rheumatoid arthritis. N Engl J Med. 2022;386 (4 ):316-326. doi:10.1056/NEJMoa2109927 35081280 23 Sherman RE, Anderson SA, Dal Pan GJ, . Real-world evidence—what is it and what can it tell us? N Engl J Med. 2016;375 (23 ):2293-2297. doi:10.1056/NEJMsb1609216 27959688 24 Zhao SS, Lyu H, Solomon DH, Yoshida K. Improving rheumatoid arthritis comparative effectiveness research through causal inference principles: systematic review using a target trial emulation framework. Ann Rheum Dis. 2020;79 (7 ):883-890. doi:10.1136/annrheumdis-2020-217200 32381560 25 Nabi H, Georgiadis S, Loft AG, . Comparative effectiveness of two adalimumab biosimilars in 1318 real-world patients with inflammatory rheumatic disease mandated to switch from originator adalimumab: nationwide observational study emulating a randomised clinical trial. Ann Rheum Dis. 2021;80 (11 ):1400-1409. doi:10.1136/annrheumdis-2021-219951 33926921 26 McDermott G, Fu X, Cook C, . The effect of achieving serological remission on subsequent risk of relapse, end-stage renal disease and mortality in ANCA-associated vasculitis: a target trial emulation study. Ann Rheum Dis. 2022;81 (10 ):1438-1444. doi:10.1136/annrheumdis-2022-222439 35697489 27 Finckh A, Ciurea A, Brulhart L, ; Doctors of the Swiss Clinical Quality Management Programme for Rheumatoid Arthritis. Which subgroup of patients with rheumatoid arthritis benefits from switching to rituximab versus alternative anti-tumour necrosis factor (TNF) agents after previous failure of an anti-TNF agent? Ann Rheum Dis. 2010;69 (2 ):387-393. doi:10.1136/ard.2008.105064 19416802 28 Harrold LR, Reed GW, Kremer JM, . The comparative effectiveness of abatacept versus anti-tumour necrosis factor switching for rheumatoid arthritis patients previously treated with an anti-tumour necrosis factor. Ann Rheum Dis. 2015;74 (2 ):430-436. doi:10.1136/annrheumdis-2013-203936 24297378 29 Sparks JA, Wallace ZS, Seet AM, ; COVID-19 Global Rheumatology Alliance. Associations of baseline use of biologic or targeted synthetic DMARDs with COVID-19 severity in rheumatoid arthritis: results from the COVID-19 Global Rheumatology Alliance physician registry. Ann Rheum Dis. 2021;80 (9 ):1137-1146. doi:10.1136/annrheumdis-2021-220418 34049860 30 Pappas DA, St John G, Etzel CJ, . Comparative effectiveness of first-line tumour necrosis factor inhibitor versus non-tumour necrosis factor inhibitor biologics and targeted synthetic agents in patients with rheumatoid arthritis: results from a large US registry study. Ann Rheum Dis. 2021;80 (1 ):96-102. doi:10.1136/annrheumdis-2020-217209 32719038 31 Pawar A, Desai RJ, Solomon DH, . Risk of serious infections in tocilizumab versus other biologic drugs in patients with rheumatoid arthritis: a multidatabase cohort study. Ann Rheum Dis. 2019;78 (4 ):456-464. doi:10.1136/annrheumdis-2018-214367 30679153 32 Kubo S, Nakayamada S, Nakano K, . Comparison of the efficacies of abatacept and tocilizumab in patients with rheumatoid arthritis by propensity score matching. Ann Rheum Dis. 2016;75 (7 ):1321-1327. doi:10.1136/annrheumdis-2015-207784 26245754 33 Norvang V, Brinkmann GH, Yoshida K, ; ARCTIC Study Group; NOR-VEAC Study Group. Achievement of remission in two early rheumatoid arthritis cohorts implementing different treat-to-target strategies. Arthritis Rheumatol. 2020;72 :1072-1081. doi:10.1002/art.41232 32090491 34 Miyazaki Y, Nakano K, Nakayamada S, . Efficacy and safety of tofacitinib versus baricitinib in patients with rheumatoid arthritis in real clinical practice: analyses with propensity score-based inverse probability of treatment weighting. Ann Rheum Dis. 2021;80 (9 ):1130-1136. doi:10.1136/annrheumdis-2020-219699 33827788 35 Iudici M, Porcher R, Riveros C, Ravaud P. Time-dependent biases in observational studies of comparative effectiveness research in rheumatology: a methodological review. Ann Rheum Dis. 2019;78 (4 ):562-569. doi:10.1136/annrheumdis-2018-214544 30755417 36 Fong C, Hazlett C, Imai K. Covariate balancing propensity score for a continuous treatment: application to the efficacy of political advertisements. Ann Appl Stat. 2018;12 :156-177. doi:10.1214/17-AOAS1101 37 Chattopadhyay A, Hase CH, Zubizarreta JR. Balancing vs modeling approaches to weighting in practice. Stat Med. 2020;39 (24 ):3227-3254. doi:10.1002/sim.8659 32882755 38 Hoisnard L, Pina Vegas L, Dray-Spira R, Weill A, Zureik M, Sbidian E. Risk of major adverse cardiovascular and venous thromboembolism events in patients with rheumatoid arthritis exposed to JAK inhibitors versus adalimumab: a nationwide cohort study. Ann Rheum Dis. 2023;82(2):182-188. doi:10.1136/ard-2022-222824 36198438 39 Pearl J, Mackenzie D. The Book of Why: The New Science of Cause and Effect. Allen Lane; 2018. 40 Hernán MA, Robins JM. Causal Inference: What If. Chapman & Hall/CRC; 2020.