(296) Risk of Cardiovascular Events in Patients with Rheumatoid Arthritis Treated with Tofacitinib vs Select TNF Inhibitors: Assessing the Concordance of Trial Findings Using Real-World Data

Publication Number: 627

Background: The ORAL Surveillance (OS) post-authorization safety trial suggested an increased risk of major cardiovascular (CV) events in patients with rheumatoid arthritis (RA) taking tofacitinib (tofa) compared to the tumor necrosis factor inhibitors (TNFi) etanercept or adalimumab. To assess the concordance of OS findings in routine patient care, the association between tofa versus TNFi and CV events was assessed in a real-world setting using principled pharmacoepidemiology methods.

Objectives: To compare the risk of CV events between tofa vs. TNFi-treated patients with RA aged >50 with >1 CV risk factor, and methotrexate use in the year prior.

Methods: Using Optum Clinformatics® an OS cohort mimicking trial inclusion/exclusion criteria was created. Patients meeting study criteria from Nov 2012-Dec 2020 were followed from initiation of tofa or TNFi for a CV event until the end of index drug exposure, insurance disenrollment, or end of data. The outcome was a composite of incident myocardial infarction or stroke/transient ischemic attack. Cox proportional hazards regression estimated hazard ratios (HR) and 95% confidence intervals (CI) between 1:1 propensity score-matched treatment groups.

Results: The final cohort included 2,378 patients (mean age 63yrs, 81% female). The incidence rate per 100 person-years for CV events among tofa and TNFi initiators, respectively, was 1.62 (0.98-2.54) and 1.02 (0.53-1.78). The HR was elevated in the main analysis (1.61; 0.78-3.32). The point estimate attenuated in a sensitivity analysis using time-varying exposure/covariates (1.22; 0.63-1.61) albeit, despite increased precision in the sensitivity analysis, CIs included the null.

Conclusions: Replicating methodology from OS in real-world data yielded point estimates for tofa vs. TNFi treated patients above one for MACE in an OS like real-world cohort. In a time-dependent analysis, the point estimate was attenuated. Confidence intervals from both primary and sensitivity analyses included the null. Considerations related to power, the inability to reliably capture death in claims and unmeasured factors related to physician prescribing, such as disease activity, warrant further study.