Three-year cardiovascular risk prediction among people who use cocaine or methamphetamine

Cardiovascular disease (CVD) risk prediction tools have not been developed or validated for adults who use nonmedical stimulants (cocaine or methamphetamine), despite substantially elevated event rates and pathophysiologically distinct risk profiles in this population. Accurate estimation of absolute risk is needed to support individualized clinical decision-making.

We developed a 3-year CVD risk prediction model using a cohort of 6940 adults aged 18-79 with documented stimulant use, identified from electronic health record (EHR) data of a large academic hospital system spanning 2016-2024. Type of stimulants used was a candidate predictor, alongside demographic characteristics, cigarette smoking status, systolic blood pressure, and baseline cardiovascular medications. Variable selection used least absolute shrinkage and selection operator (LASSO) with 10-fold cross-validation, followed by unpenalized logistic regression on selected variables. Model performance was assessed using discrimination and calibration metrics; bootstrap internal validation estimated and corrected for optimism in apparent performance.

LASSO selected 8 predictors: age, sex, Black race, Hispanic ethnicity, current cigarette smoking, cocaine-only use, any cardiovascular medication use, and systolic blood pressure. Calibration was excellent. Observed and predicted 3-year event rates were 29.6% and 30.1%, respectively (observed/expected [O/E] ratio 0.985, calibration slope 1.000, integrated calibration index [ICI] 0.008). Performance was maintained after optimism correction: O/E ratio 0.988, calibration slope 0.992, ICI 0.010. The apparent C-statistic was 0.730 (95% CI 0.717-0.743); the optimism-corrected C-statistic was 0.728. For illustration, a 45-year-old Black adult with cocaine use, current cigarette smoking, and systolic blood pressure of 130 mmHg has an estimated 3-year CVD risk of approximately 27.7%; the same profile with methamphetamine use yields approximately 22.1%.

A CVD risk prediction model tailored to individuals with stimulant use accurately estimated absolute 3-year CVD risk, with excellent calibration and clinically meaningful risk stratification. These findings establish proof of concept for a stimulant-specific CVD risk prediction tool. External validation and implementation studies are needed before clinical deployment.