Risk Factor Analysis for Cancer and Coronary Heart Disease: A Machine Learning Approach Using National Health and Nutrition Examination Survey Data

Objectives: The relative significance of predictive factors for cancer and coronary heart disease (CHD) is still unclear. This study aims to identify and evaluate the risk factors contributing to the development of both conditions using the CatBoost machine learning algorithm. Methods: Data from twelve datasets of the 2009 to 2010 National Health and Nutrition Examination Survey (NHANES), incorporating both survey responses and laboratory results, were used. Separate CatBoost models were developed to predict cancer and CHD occurrences, by using Shapley Additive Explanations (SHAP), with the help of Recursive Feature Elimination with Cross-Validation (RFECV), and by adjusting class weights, and model performance was assessed using Receiver Operating Characteristic (ROC) curves. Results: The datasets were combined to form a cohort of 5,012 participants, each with 24 selected features. The cancer prediction model achieved a ROC Area Under the Curve (AUC) of 0.76, with 13 selected features, yielding an accuracy of 0.70, sensitivity of 0.67, and specificity of 0.70. In contrast, the CHD prediction model achieved a higher ROC AUC of 0.87, with an accuracy of 0.83, sensitivity of 0.78, and specificity of 0.83. Accordingly, top predictive features for each disease have been ranked and selected by the CatBoost algorithm. Conclusions: This study identifies key demographic and laboratory features significantly associated with cancer and CHD risk in the NHANES dataset. The findings suggest that these factors could be valuable for estimating individual risk and could inform machine learning models aimed at early detection and screening.