Precar: An Adaptive ML-Based Tool for Predicting Cardiovascular Disease

Predicting the probability of having a cardiovascular event during the next 10 years is a medical practice that allows physicians to provide ad-hoc treatment on time. Based on this indicator, medical doctors routinely request additional medical and clinical examinations to ascertain the presence of heart disease in patients. Previous work in Chile has shown that cardiovascular risk models built in other regions may not be directly applicable without first adapting them to the realities of the local population. In this paper, we introduce Precar, a comprehensive tool designed to predict cardiovascular risk using machine-learning models. Notably, Precar implements four powerful methods - decision trees, random forests, support vector machines, and k-nearest neighbors - with an accuracy over 90%. Additionally, the renowned Framingham risk indicator is integrated to enhance risk assessment. The early model versions were developed using the well-known Cleveland Heart Disease Dataset, ensuring a robust foundation for risk prediction. A key aspect of Precar is its support for a continuous learning approach, enabling collaboration between data scientists and medical doctors. Medical practitioners can actively decide to retrain the models with daily patient data when the model's performance during medical practice falls below a specified threshold. This adaptive learning process ensures that the tool remains current and relevant in real-world clinical settings. Results are encouraging. The transparency and collaboration provided by Precar empower clinicians, enhancing their confidence in machine learning predictions and driving more effective decision-making in patient care.