Abstract

Abstract Aims Several risk prediction models are available for determining the 10-year risk of cardiovascular disease (CVD), including the Norwegian NORRISK 2 model. However, the existing models explain only a modest proportion of the incidence. Therefore, this study aimed to develop improved models for predicting the 10-year risk of myocardial infarction (MI) for both sexes. Methods Data from 31,946 participants without prior CVD were analyzed. The data set was divided into a training set (for estimation) and a test set (for model evaluation). Prediction models were developed on the training set for each sex using XGBoost and logistic regression, using 96 (men) and 100 (women) variables. The models were evaluated on the test set using Receiver-Operating-Characteristic (ROC) and Precision-Recall (PR) curves. Age and sex-specific thresholds for intervention were explored through cross-validation on the training set. Model performance was compared to the NORRISK 2 model using the test set. Results The XGBoost model improved CVD risk prediction for men across all age groups (AUCROC for XGBoost and NORRISK 2, respectively, 0.72 and 0.65 (age 45-54), 0.63 and 0.62 (age 55-64), 0.69 and 0.62 (age 65-74)). For women, NORRISK2 performed better than XGBoost in ROC curve evaluation. However, XGBoost were superior to NORRISK2 when evaluated by PR-curves in women aged 55-64 (PR-AUC for XGBoost and NORRISK 2, respectively, 0.20 and 0.12 compared to 0.06 for a no-skills model). Potential new risk predictors, such as alkaline phosphatase (ALP) for men and thyroid stimulation hormone (TSH) for women, were identified. The results also indicate that the thresholds for intervention should be sex specific. Conclusion By employing machine learning and incorporating sex-specific risk factors, we propose improved risk prediction models for CVD, particularly in men. Introducing sex-specific thresholds for intervention could enhance CVD prevention for both women and men.