The CorEvitas RA Japan registry (NCT02737449) is a prospective, multicenter, observational disease-based registry that was launched in February 2016. Longitudinal follow-up data were collected from patients and their treating rheumatologists during routine clinical encounters using questionnaires. As of June 30, 2022, the registry included data from 48 private and academic active clinical sites, with over 219 physicians and 2359 patients with RA across 27 prefectures in Japan.

This study (NCT05572567) included patients with rheumatologist-diagnosed RA, who initiated MTX, tofacitinib, TNFi (adalimumab [originator or biosimilar], certolizumab pegol, etanercept [originator or biosimilar], golimumab, infliximab [originator or biosimilar] or any other TNFi biosimilar approved during the study), or non-TNFi bDMARDs (abatacept, tocilizumab, or sarilumab) between March 1, 2016 and June 30, 2022, (October 5, 2022 data extract) and had an index visit (defined as the visit during which the first ever initiation of a drug took place). For patients with more than one initiation in a drug class, the index date was the initiation date of the first drug.

For the analyses of the safety cohort, eligible patients had at least 1 day of registry follow-up time after the index visit unless the individual had an event on the day of initiation. For the analyses of the effectiveness cohort, eligible patients had a 6-month follow-up visit (± 3 months) after initiation of the drug.

This structured, secondary study was conducted within the CorEvitas Registry in accordance with the International Council for Harmonisation Guidelines for Good Clinical Practice and Ethical Guidelines for Medical and Health Research Involving Human Subjects. All patients provided written informed consent for participation in the registry.

In both the safety and effectiveness cohorts, patient characteristics were measured at baseline and included sociodemographic characteristics (age, sex, race, education, employment), lifestyle characteristics (smoking status and alcohol use), and health measures (such as body mass index [BMI], weight, provider-reported depression, provider-reported history of cardiovascular comorbidity, and non-cardiovascular comorbidity). Disease characteristics and some patient-reported outcomes (PROs) were also recorded at baseline.

In the safety cohort, safety events of interest included MACE, total cardiovascular disease (CVD), VTE, total serious infections, total herpes zoster (HZ; non-serious and serious), and total malignancies excluding NMSC. The definitions of these events are listed in the Supplementary Material. Patients were followed according to a risk window, which, for all events except malignancies, began with the index visit and continued until the visit closest to 90 days after the end of therapy, or end of data collection, whichever came first. If the 90-day window overlapped with an initiation of a drug in a different group, then an event during this period was attributed to both therapies. An “any exposure” approach was utilized for analyses of risk for malignancies, in which the risk window began with the index visit and extended until the end of data collection, even in the case of subsequent switching to another therapy. If a malignancy was diagnosed after the start of a subsequent therapy, the event was attributed to both therapies.

In the effectiveness cohort, discontinuation rates were evaluated at the 6-month follow-up visit. Effectiveness outcomes, also evaluated at the 6-month follow-up visit, included mean change from baseline in Clinical Disease Activity Index (CDAI; 0–76), proportion of patients achieving a minimum clinically important difference (MCID) in CDAI, proportion of patients achieving modified American College of Rheumatology (mACR)20/50/70 responses, and the mean change from baseline in PROs: Japanese version of the Health Assessment Questionnaire (J-HAQ; 0–3), patient pain (Visual Analog Scale [VAS]; 0–100), Patient Global Assessment (VAS; 0–100), patient fatigue (VAS; 0–100), morning stiffness (yes/no; if yes, the patient was asked to state the duration of morning stiffness [0–24 h and 1–60 min]), and EuroQol 5-Dimension 5-Level (EQ-5D-5L; 0–1). MCID in CDAI was based on the CDAI value at initiation (the cut points for improvement were 1 if CDAI at baseline was < 10, 6 if CDAI at baseline was between 10–22 [inclusive], and 12 if CDAI at baseline was > 22) [16].

Follow-up time was summarized descriptively for each treatment group. For each outcome, only the first event for each patient was analyzed. Incidence rates (IRs; number of patients with events per 100 patient-years) and 95% confidence intervals (CIs) were calculated on the basis of time to the first event for patients with at least one event. For patients without events, the calculation of exposure time included the entire risk period. A Poisson distribution with 95% CIs was computed. A multivariate Poisson model was built for each safety outcome, adjusted for potential a priori-defined confounders, and was used to estimate marginal IRs for each safety outcome within each treatment group. Specifically, a multiplicative Poisson regression model was fitted as a log-linear regression (i.e., a log link and a Poisson error distribution), with an offset equal to the natural logarithm of person-time. Adjusted models included the following potential confounders: age, sex, baseline CDAI, number of prior bDMARDs, BMI, duration of RA, Physician Global Assessment, J-HAQ, and the baseline value of the outcome measurement. Each patient’s observed covariates and the model coefficients were used to compute the average probability of an event by treatment group. Patients with missing covariate data were omitted from both the unadjusted and adjusted analyses. No outcome or missing data for covariates were imputed.

A separate regression model was fit for each effectiveness outcome, with DMARD group as the exposure variable (primary independent variable of interest). Linear models were used for continuous measures and logistic models were used for binary measures. Unadjusted mean outcomes and 95% CIs were calculated for each treatment group. Adjusted regression models were fit, and marginal means, with associated 95% CIs, were estimated. A priori-defined potential confounders for adjusted models and mean marginal effect computations were the same as for the primary outcomes. Patients with missing covariate data were omitted from both the unadjusted and adjusted analyses. Discontinuations were summarized descriptively for each treatment group. For patients who discontinued the index therapy prior to the 6-month follow-up visit, a non-response was imputed for binary outcome measures, and the last observation carried forward was used as the 6-month value for continuous measures. For patients who discontinued at their 6-month visit, the 6-month outcome was evaluated at the visit.

For the purposes of this analysis, across all safety and efficacy outcomes evaluated, comparisons between treatment groups for marginal mean changes from baseline (continuous outcomes) and proportions (binary outcomes) are described as higher or lower if 95% CIs do not overlap, and numerically higher or lower if 95% CIs do overlap, as appropriate. Multiplicity adjustment was not conducted because of the exploratory nature of the analysis.