Machine learning-based cardiovascular risk prediction in systemic lupus erythematosus: development and internal validation of a prognostic model

BACKGROUND: Cardiovascular (CV) diseases are the leading cause of mortality in patients with systemic lupus erythematosus (SLE). While traditional risk factors inadequately assess CV risk, SLE-specific determinants remain elusive. We have conducted a study of CV outcomes in a cohort of SLE patients, aiming to facilitate individualized risk assessment. METHODS: The LESLY cohort comprised patients diagnosed with SLE at the University Hospital of Lyon between January 2002 and August 2020. CV events (CVE) were defined as myocardial ischemia or stroke. Complete-case multivariable analyses identified predictors of CV risk, among baseline SLE characteristics and traditional CV factors. The identified predictors were subsequently employed to develop machine learning models estimating CV risk in patients with SLE. The dataset was partitioned into training (80%, n = 699) and validation (20%, n = 175) sets. RESULTS: CVE occurred in 55 LESLY patients (6.3%), including 27 acute coronary syndromes and 25 ischemic strokes, over a mean follow-up of 8.8 ± 5.2 years. Antiphospholipid antibodies (HR = 3.51 [1.91-6.43], p < 0.001) and inaugural skin involvement (HR = 2.69 [1.25-5.77], p = 0.011) were the most potent predictors of CVE occurrence. Finally, an Elastic Net Penalized Cox model accurately predicted individualized CV risks in an internal validation cohort (C-index 0.791 [95% CI: 0.674-0.906]; Brier score 6.4% [95% CI: 3.7-9.8%]). CONCLUSION: This study corroborates the primacy of CVD in SLE, underscoring the predictive roles of inaugural skin involvement and antiphospholipid antibodies. These findings enabled the development of an exploratory risk stratification model that may inform clinical decision-making for cardiovascular risk management in SLE, pending external validation.