Evidence-Based Case Recommendation System for Cardiac Health Diagnosis Using Non-Negative Factorization Data-Driven Similarity Approach

Effective management and continuous monitoring of patients with cardiovascular risk are essential to reducing adverse events. Diagnostic tools, particularly evidence-based ones, have significantly improved healthcare and enabled advances in personalized medicine. Recommender systems (RS) commonly used to offer personalized recommendations based on historical clinical data and patterns in large health databases. This work presents a RS based on case evidence for monitoring and diagnosing cardiovascular health. Using a similarity approach through the clustering algorithm and multivariate analysis of nonnegative matrix factorization (NMF), using the Framingham heart disease dataset to identify patient groups and their clinical profiles. This enables the evaluation of new patient cases, identification of similar cases, and inference of possible conditions or risks from a semi-supervised approach. The experimental results during train-validation show the best performances of NMF with lower average RMSE and lower variance of 0.2526 ± 0.0022 for 13 groups and test performance of RMSE of 0.2550 and Accuracy of 0.7386, suggesting the potential use to retrieve similar cases and their associated prevalence diagnosis as a complementary diagnostic support tool.