Effect of Comorbidities Features in Machine Learning Models for Survival Analysis to Predict Prodromal Alzheimer's Disease

Alzheimer's Disease (AD) is the most common form of dementia, specifically a progressive degenerative disorder affecting 47 million people worldwide and is only expected to grow in the elderly population. The detection of AD in its early stages is crucial to allow early intervention aiding in the prevention or slowing down of the disease. The effect of using comorbidity features in machine learning models to predict the time until a patient develops a prodrome was observed. In this study, we used Alzheimer's Disease Neuroimaging Initiative (ADNI) high-dimensional clinical data to compare the performance of six machine learning algorithms for survival analysis, combined with six feature selection methods trained on two settings: with and without comorbidities features. Our ridge model combined with permutation feature selection achieves maximum performance of 0.90 when using comorbidity features with the concordance index as a performance indicator. This demonstrated that incorporating comorbidities into the feature set enhances the performance of survival analysis for Alzheimer's disease. There is potential to identify risk factors (coronary artery disease) from comorbidities which could guide preventative care based on medical history.