Development of an accessible 10-year Digital CArdioVAscular (DiCAVA) risk assessment: a UK Biobank study

Background: Cardiovascular diseases (CVDs) are among the leading causes of\ndeath worldwide. Predictive scores providing personalised risk of developing\nCVD are increasingly used in clinical practice. Most scores, however, utilise a\nhomogenous set of features and require the presence of a physician.\n Objective: The aim was to develop a new risk model (DiCAVA) using statistical\nand machine learning techniques that could be applied in a remote setting. A\nsecondary goal was to identify new patient-centric variables that could be\nincorporated into CVD risk assessments.\n Methods: Across 466,052 participants, Cox proportional hazards (CPH) and\nDeepSurv models were trained using 608 variables derived from the UK Biobank to\ninvestigate the 10-year risk of developing a CVD. Data-driven feature selection\nreduced the number of features to 47, after which reduced models were trained.\nBoth models were compared to the Framingham score.\n Results: The reduced CPH model achieved a c-index of 0.7443, whereas DeepSurv\nachieved a c-index of 0.7446. Both CPH and DeepSurv were superior in\ndetermining the CVD risk compared to Framingham score. Minimal difference was\nobserved when cholesterol and blood pressure were excluded from the models\n(CPH: 0.741, DeepSurv: 0.739). The models show very good calibration and\ndiscrimination on the test data.\n Conclusion: We developed a cardiovascular risk model that has very good\npredictive capacity and encompasses new variables. The score could be\nincorporated into clinical practice and utilised in a remote setting, without\nthe need of including cholesterol. Future studies will focus on external\nvalidation across heterogeneous samples.\n