Janus kinase inhibitor exposure and acute coronary event patterns in rheumatoid arthritis: a treatment-aware retrospective study.

BACKGROUND

Patients with rheumatoid arthritis (RA) have excess cardiovascular risk, and cardiovascular safety concerns surrounding Janus kinase inhibitors (JAKi) make acute coronary event assessment clinically important. Real-world analyses are difficult because treatment selection, sparse events, and complete separation can distort conventional regression. We evaluated treatment-specific acute coronary event patterns in a single-center RA cohort using sparse-data and balance-oriented methods.

METHODS

We performed a retrospective study of 316 RA patients. The primary exposure was JAKi treatment; non-JAKi treatment included TNFi or csDMARD regimens, although the pooled comparator was predominantly csDMARD-only therapy. The primary outcome was a recorded acute coronary event (acute myocardial infarction [AMI] or unstable angina), with a definite-AMI sensitivity analysis. We used exact event-rate estimation, Firth bias-reduced logistic regression, overlap-weighted propensity-score diagnostics with effective sample size estimation, Bayesian beta-binomial zero-event shrinkage, and tipping-point analysis. Missingness was limited to 2 HDL-C values and 1 BMI value (0.6% and 0.3%).

RESULTS

Among 316 patients, 115 received JAKi and 201 received non-JAKi therapy; 190 of 201 non-JAKi patients received csDMARD-only therapy. Acute coronary events occurred in 0/115 JAKi-treated patients, 2/11 TNFi-treated patients, and 87/190 csDMARD-treated patients (exact 95% CI for JAKi: 0.0%-3.2%). In the clinically adjusted Firth model, JAKi exposure was associated with lower odds of an acute coronary event (OR 0.022, 95% CI 0.001-0.344; P = 0.007). The overlap-weighted effective sample size was 215.5 overall (100.7 JAKi; 115.8 non-JAKi), with a largest post-weighting standardized mean difference of 0.053 and a weighted risk difference of - 16.4 percentage points. Marginal standardization of the clinical Firth model yielded model-predicted event probabilities of 9.5% under JAKi exposure and 32.1% under non-JAKi exposure. Bayesian analysis estimated a posterior median JAKi event rate of 0.20% (95% credible interval 0.0004%-2.16%). Thirty-eight hypothetical JAKi events were required before the exact comparison became non-significant.

CONCLUSIONS

In this cohort, JAKi exposure identified a selected RA subgroup with a zero-event acute coronary pattern across multiple sparse-data analyses. These findings should not be interpreted as causal cardioprotection; rather, they support transparent reporting of treatment selection, baseline cardiovascular phenotype, and absolute-risk summaries in real-world JAKi safety studies. Key Points • This retrospective study used sparse-data methods and overlap-weighted diagnostics to address zero acute coronary events among JAKi-treated patients in a real-world RA cohort. • Missingness in model covariates was minimal: 2 HDL-C values and 1 BMI value were missing (0.6% and 0.3%), with no variable exceeding 5% missingness. • The primary binary comparison was dominated by csDMARD-treated patients in the non-JAKi group; TNFi estimates were exploratory because only 11 patients received TNFi therapy. • Absolute-risk summaries were added: the overlap-weighted risk difference was - 16.4 percentage points, and clinical Firth model-predicted probabilities were 9.5% under JAKi exposure and 32.1% under non-JAKi exposure.