Background:  Prevention strategies for coronary heart disease (CHD) require reliable estimates of CHD risk. No such equations exist for HIV⁺ persons, where components of ART may contribute to this risk. We developed a CHD risk equation tailored to HIV⁺ patients.

Methods:  Prospective multi-national cohort study of HIV⁺ subjects. Step 1 developed a model based on 9023 subjects who had full covariate data and were free of CHD at entry into the study. The risk equation to predict CHD was developed based on parametric survival models. Estimates from the risk equation and corresponding hazard ratios (HR) from a Cox model are reported. The performance of the equation was assessed on this development dataset by testing the prognostic system’s discrimination, calibration, and accuracy. The predictive performance was also compared to that of a conventional prediction model (Framingham). Step 2 will validate the model on a separate dataset (D:A:D cohort II).

Results:  Over 33,594 person-years, 157 cases of CHD occurred. The best fitting parametric model was log-logistic, and included the conventional risk factors of (b-coefficient from log-logistic model with constant = 11.498 and g = 0.933; HR from Cox model): age (per 5 years older –0.334; 1.42), sex (male –0.796; 2.35), family history of CHD (–0.478; 1.66), systolic blood pressure (per 10 mmHg higher  –0.050; 1.05) and smoking status (current –1.042, ex –0.456; 2.97, 1.78), the ratio of TC/HDL (per unit higher –0.144; 1.16), diabetes (fitted separately by sex due to interaction (in men –0.683, women –1.349; 1.94 and 4.04)), and in addition duration of protease inhibitor (PI) exposure (per additional year –0.114; 1.13). Age, PI exposure, and smoking status were fitted as time-updated, while all other covariates took the fixed value at baseline for the analyses. The area under the receiver-operator characteristic curve discrimination statistic was 0.78 (95%CI 0.75 to 0.82). Overall, the D:A:D equation predicted 153 CHD events, compared with 187 events predicted by the Framingham equation. Predictions were accurate in sub-groups of patients according to sex and smoking status (see the table).

Conclusions:  The Framingham risk equation over-predicted CHD events in this cohort. In contrast, the D:A:D equation, developed on HIV⁺ subjects and incorporating PI exposure as well as conventional CHD risk parameters, accurately predicted CHD outcomes in the development dataset.