Enhancing heart failure treatment decisions: interpretable machine learning models for advanced therapy eligibility prediction using EHR data

• DOI: 10.1186/s12911-024-02453-y
• Publication Date: 2024-02-14T14:02:39Z
• Authors: Yufeng Zhang; Jessica R. Golbus; Emily Wittrup; Keith D. Aaronson; Kayvan Najarian
• Keywords: 03 medical and health sciences; 0302 clinical medicine; Research; Machine learning; Computer applications to medicine. Medical informatics; R858-859.7; Electronic health records; Heart failure; Interpretability

Abstract Timely and accurate referral of end-stage heart failure patients for advanced therapies, including transplants mechanical circulatory support, plays an important role in improving patient outcomes saving costs. However, the decision-making process is complex, nuanced, time-consuming, requiring cardiologists with specialized expertise training transplantation. In this study, we propose two logistic tensor regression-based models to predict warranting evaluation therapies using irregularly spaced sequential electronic health records at population individual levels. The clinical features were collected previous visit predictions made very beginning subsequent visit. Patient-wise ten-fold cross-validation experiments performed. Standard LTR achieved average F1 score 0.708, AUC 0.903, AUPRC 0.836. Personalized obtained 0.670, 0.869 0.839. not only outperformed all other machine learning which they compared but also improved performance robustness via weight transfer. scores support vector machine, random forest, Naive Bayes are by 8.87%, 7.24%, 11.38%, respectively. can evaluate importance associated therapy referral. five most medical codes, chronic kidney disease, hypotension, pulmonary mitral regurgitation, atherosclerotic reviewed validated literature cardiologists. Our proposed effectively utilize EHRs potential necessity while explaining comorbidities events. information learned from trained model could offer further insight into risk factors contributing progression both