Understanding risk factors in cardiac rehabilitation patients with random forests and decision trees

Cardiac rehabilitation is a well-recognised non-pharmacological intervention recommended for the prevention of cardiovascular disease. Numerous studies have produced large amounts of data to examine the above aspects in patient groups. In this paper, datasets collected for over a 10 year period by one Australian hospital are analysed using decision trees to derive prediction rules for the outcome of phase II cardiac rehabilitation. Analysis includes prediction of the outcome of the cardiac rehabilitation program in terms of three groups of cardiovascular risk factors: physiological, psychosocial and performance risk factors. Random forests are used for feature selection to make the models compact and interpretable. Balanced sampling is used to deal with heavily imbalanced class distribution. Experimental results show that the outcome of phase II cardiac rehabilitation in terms of physiological, psychosocial and performance risk factor can be predicted based on initial readings of cholesterol level and hypertension, level achieved in six minute walk test, and Hospital Anxiety and Depression Score (HADS) anxiety score and HADS depression score respectively. This will allow for identifying high risk patient groups and developing personalised cardiac rehabilitation programs for those patients to increase their chances of success and minimize their risk of failure.