Cardiovascular Risk Prediction Functions Underestimate Risk in HIV Infection

Abstract

Background

Cardiovascular disease (CVD) risk is elevated in HIV-infected individuals, with contributions from both traditional and non-traditional risk factors. The accuracy of established CVD risk prediction functions in HIV is uncertain. We sought to assess the performance of three established CVD risk prediction functions in a longitudinal cohort of HIV-infected men.

Methods

Framingham Heart Study (Framingham) functions for hard coronary heart disease (Framingham CHD) and atherosclerotic CVD (Framingham ASCVD) and the American College of Cardiology/American Heart Association (ACC/AHA ASCVD) function were applied to the Partners HIV cohort. Risk scores were calculated between 1/1/2006 and 12/31/2008. Outcomes included CHD (myocardial infarction [MI] or coronary death) for the Framingham CHD function and ASCVD (MI, stroke or coronary death) for the Framingham ASCVD and ACC/AHA ASCVD functions. We investigated the accuracy of CVD risk prediction for each function when applied to the HIV cohort using comparison of Cox regression coefficients, discrimination, and calibration.

Results

The HIV cohort was comprised of 1280 men followed for a median of 4.4 years. There were 80 (6.3%) ASCVD events; 5-year incidence rate was 16.7 per 1000 person years. Discrimination was moderate to poor as indicated by low c statistic (0.68 for Framingham CHD, 0.65 for ACC/AHA, and 0.67 for Framingham ASCVD). Observed CVD risk exceeded predicted risk for each of the functions in most deciles of predicted risk. Calibration, or goodness-of-fit of the models, was consistently poor, with significant chi-square p values for all functions. Recalibration did not significantly improve model fit.

Conclusions

Cardiovascular risk prediction functions developed for use in the general population are inaccurate in HIV infection and systematically underestimate risk in a cohort of HIV-infected men. Development of tailored CVD risk prediction functions incorporating traditional CVD risk factors and HIV-specific factors is likely to result in more accurate risk estimation to guide preventative CVD care.