A multicentre evaluation of cardiovascular risk prediction tools in people living with HIV across England (ECARTH study).

introductionIt remains unclear which cardiovascular disease (CVD) risk scores are optimized for people living with treated HIV with high rates of viral suppression. We evaluated the performance of commonly used CVD risk scores in a population of primarily well-treated HIV patients in England.

methodsA multi-centre retrospective cohort analysis was performed, including people living with HIV (PLWH) above the age of 40 with no prior major adverse CVD events (MACE), regularly attending 7 HIV clinics in England during 2014. Outcomes were MACE over the following 5 years: myocardial infarction, invasive cardiac procedure e.g. primary angioplasty, cerebrovascular events, new heart failure, and CVD-related death. Clinical outcomes were aligned with the respective risk scores. The following CVD risk models were evaluated: the UK primary-care validated QRISK3, Framingham laboratory and non-laboratory (Office) scores, Data collection on Adverse events of anti-HIV Drugs (D:A:D), Pooled Cohort Equation (PCE), and the World Health Organization (WHO) laboratory and non-laboratory models. Models were scaled for 5-year CVD estimates. We assessed the discrimination and calibration of these 5-year models within our population. Multiple imputation was used to address missing data.

resultsOf 2582 people, 94 had at least one MACE documented during follow-up. All seven scores demonstrated moderate discrimination. QRISK3 demonstrated the best calibration in this cohort with an O:E ratio of 1.169, 95% CI 0.950, 1.439, mean calibration-in-the-large (CITL) value of 0.156 (95% CI -0.052, 0.364) and slope of 0.897 (0.682, 1.113). Framingham Office generally overpredicted risk, while WHO scores and D:A:D generally underpredicted risk. Imputing for missing data yielded results similar to the complete case analysis.

conclusionsCVD risk scores demonstrated moderate performance in a well-treated PLWH cohort. Our findings emphasize the need to calibrate each CVD risk score to the local population to guide prioritization of clinical resources for CVD prevention. EVIDENCE IN CONTEXT: Evidence before this study: We systematically searched Medline, Embase, and Cochrane Library database from January 1995 to August 2023, using the following terms 'HIV/AIDS and Cardiovascular Disease (CVD) outcomes', with an updated search to 2025. The risk of CVD has been reported to be up to two times that of people living without HIV, along with increased rates of heart failure and sudden death. Commonly used CVD risk scores have demonstrated variable performance for PLWH, and only the D:A:D study equation has been specifically validated in an HIV-positive population. Within an era of modern ART regimes and high rates of viral suppression, there have been numerous comparisons of CVD risk across scoring systems, but few studies with observed outcomes of CVD rates to validate the predictions. It remains unclear which CVD risk calculator is the most suitable for PLWH. The use of routinely collected clinical data may also impact the accuracy of results due to the presence of incomplete or missing data fields. In this study, we aimed to evaluate the performance of commonly used CVD risk scores in a population of primarily well-treated HIV patients in England using routinely collected clinical data. Added value of this study: We demonstrate that of the seven CVD risk scores evaluated (QRISK3, Framingham laboratory and non-laboratory (Office) scores, D:A:D, Pooled Cohort Equation (PCE), and the WHO laboratory and non-laboratory models), all showed moderate discrimination compared to observed major adverse cardiovascular events (MACE) rates in an English multicentre cohort of people living with HIV attending clinics. QRISK3, using a large UK primary care database to validate predicted CVD risk, appeared to be most closely calibrated to CVD outcomes in our cohort without the need for additional calibration. Our findings also suggest that the use of non-laboratory scores, which have been suggested for low and middle-income settings where laboratory results may not be available, should be interpreted with caution, and further testing may be required due to their miscalibration in this cohort. Implications of all the available evidence: Using real-world clinical data from people with HIV who are engaged in care in England, these findings emphasize the need to calibrate each CVD risk score to the local population it is used for in order to accurately guide prioritization of clinical resources for primary CVD prevention.