Estimating the degree of cardiovascular disease risk under-prediction in people with HIV by existing risk models: an updated meta-analysis.

BACKGROUND

The application of cardiovascular disease (CVD) risk models in people with HIV (PWH) may be limited due to concerns that the models tend to underestimate risk in this population. We conducted an updated meta-analysis to assess the degree of CVD risk under prediction in PWH by existing CVD risk models.

METHODS

We searched electronic databases and the reference lists of relevant articles. Data were abstracted using standardized forms. Calibration data included observed: expected ratios, which were graphically summarized. Weighted-regression models were fitted to estimate the relationship between observed and expected event rates across studies. Discrimination data included C -statistics which were pooled using random-effects model meta-analysis.

RESULTS

Sixteen publications reporting data on 97 945 PWH (mean age, 41 years; 75% male) were included. The risk scores tended to underpredict CVD risk in PWH, especially for individuals with lower expected risk. For PWH with expected 10-year CVD risk of 3%, the risk scores can underestimate risk by 71% on average. By contrast, for PWH with 6% and 9% expected 10-year CVD risk, the underestimation is by ~23% and ~7%, respectively. The risk scores had moderate performance in discrimination (pooled C -statistic values ranged from 0.71 to 0.82). There was substantial heterogeneity across studies and risk models. The 10-year models of the American Joint Societies Pooled Cohort Equation (PCE-10) and the Framingham Risk Score (FRS-10) had better calibration.

CONCLUSIONS

Extant risk scores underpredict CVD risk in PWH, especially for individuals with lower expected risk, suggesting a need for implementation of HIV-specific calibration factors in this population. PCE-10 and FRS-10 showed better calibration.