Accurate machine learning-based CVD risk prediction in primary care may reduce the need for routine healthcare checks

Guideline recommended cardiovascular risk prediction models, such as QRISK3 [1] (commonly used in the UK), PCE [2] (US), and SCORE2 (Europe) are used in clinical practice to identify people at risk of future heart disease, enabling risk-based management and prevention of disease. Their routine application is however limited by the need to actively measure key variables in the clinic. Financial constraints for screening services and testing, coupled with limited population attendance, results in incomplete coverage. In the UK, over a 5-year period only around 60% of those eligible for a National Health Service health check were invited, and of these, only a half attended.

A set of partial models was developed to predict the 10-year risk of cardiovascular disease (CVD) and major CVD (including atrial fibrillation, heart failure, and peripheral arterial disease), using combinations of 14 predictors. This approach allows for application in settings where only a subset of variables is available. The set of partial models was evaluated across five studies carried out in the UK and the Netherlands, jointly comprising 105,550 participants.

A total of 4,096 unique models were trained to predict 10-year major CVD risk, showing nearly identical performance when evaluated against both CVD and major CVD outcomes. The c-statistic ranged between: quartiles (Q1: 0.71 and Q3: 0.73 across the five studies. This was comparable to the performance of the PCE (Q1: 0.70, Q3: 0.74, 10 predictors) and SCORE2 (Q1: 0.71, Q3: 0.75, 8 predictors). Due to the large number of required predictors (22/23 for men/women) the QRISK3 was evaluated in a single cohort: c-statistic 0.72 (95% CI 0.72; 0.73). Model performance remained adequate when focussing on the set of partial models using 2-4 predictors: c-statistic Q1: 0.70 and Q3: 0.71. The partial models demonstrated reasonable calibration across most studies, observing a limited risk underestimation in two cohorts. Partial models excluding traditional risk factors such blood pressure and lipids demonstrated similar performance to models incorporating these variables.

These results demonstrate that even with only a subset of predictor variables, the partial models approach can generate clinically relevant predictions of 10-year (major) cardiovascular disease risk, supporting earlier and more effective treatment decisions. The set of partial models are accessible through a python-based API (https://gitlab.com/cvd_in_t2dm/array-of-cvd-prediction-models), allowing for integration in personal or clinical care dashboards.