EATG » Performance of CVD Risk Prediction Algorithms in People Living With HIV

Performance of CVD Risk Prediction Algorithms in People Living With HIV

Algorithms may help predict cardiovascular disease (CVD) risk in people living with HIV (PLWHIV), according to a study published in the Journal of Acquired Immune Deficiency Syndromes.

There is a higher burden of CVD observed in PLWHIV compared with people who are not infected with HIV. This is most likely a result of an interplay between HIV-related factors and CVD risk factors, which include persistent inflammation and immune activation, certain antiretrovirals, and damage to the immune system. Guidelines that outline how to manage CVD risk recommend initiation of primary prevention that is based on a person’s individual risk. Currently, there are 3 commonly used algorithms used to predict CVD risk; however, they were based on the general population and do not consider HIV‑related CVD risk factors.

Therefore, an algorithm was developed to more accurately predict the risk for CVD in PLWHIV, called the Data Collection on Adverse Effects of Anti-HIV Drugs (D:A:D) algorithm. However, previous studies have shown conflicting results on whether this algorithm is superior in predicting CVD risk in PLWHIV. This study evaluated the performance of 4 commonly used algorithms to estimate CVD risk.

In total, data from 16,070 people living with HIV were used. Participants were aged ≥18 years, were in care between 2000 and 2016, initiated first combination antiretroviral therapy >1 year ago, had no preexisting CVD, and had available data on CD4 count, cholesterol, and blood pressure measurements. The predictive performance of the following 4 algorithms was performed, using a Kaplan-Meier approach: D:A:D, Systemic Coronary Risk Evaluation adjusted for national data (SCORE-NL); Framingham CVD Risk Score (FRS);, and the American College of Cardiology and American Heart Association Pooled Cohort Equations (PCE). Model discrimination was assessed using Harrell’s C-statistic. Observed vs expected rations, calibration plots, and Greenwood-Nam-D’Agostino goodness-of-fit-tests were used to assess calibration.

Results suggested that all 4 algorithms showed acceptable discrimination with a Harrell’s C-statistic of 0.73 to 0.79. On a population level, the D:A:D, SCORE-NL, and PCE algorithms slightly underestimated CVD risk, whereas FRS slightly overestimated CVD risk (observed vs expected-ratios, 1.35, 1.38, 1.14, and 0.92, respectively). The algorithms that best fit the data were D:A:D, FRE, and PCE, but were still statistically significant in their lack of fit, whereas SCORE-NL performed the poorest. The researchers highlighted that although these algorithms are useful in clinical practice, clinicians should be aware of their limitations when using them.

Overall, the study authors concluded that, “Future studies should investigate the effect of immune activation and inflammatory markers, newer antiretrovirals, and [combined antiretroviral therapy] initiation soon after HIV diagnosis on CVD risk and prediction algorithms in [PLWHIV].”

By Zahra Masoud


van Zoest RA, Law M, Sabin CA, et al. Predictive performance of cardiovascular disease risk prediction algorithms in people living with HIV [published April 23, 2019]JAIDS. doi:10.1097/QAI.0000000000002069