(VP073) MACHINE LEARNING-BASED APPROACH FOR PREDICTING POST-TREATMENT SURVIVAL FOR PATIENTS WITH CORONARY ARTERY DISEASE

Background: Coronary artery disease (CAD) is a leading cause of mortality worldwide. Coronary artery bypass graft surgery (CABG) and percutaneous coronary intervention (PCI) are two commonly used treatments for CAD, each with its own risks and benefits. Predicting patient survival is critical for developing personalized treatment plans and improving medical decision-making. In this study, we aimed to predict the survival of each patient with CAD based on a range of clinical features, including patient covariates, cardiovascular problems, previous treatments, and relevant medical history, which are collected prior to the treatment, as well as the treatment type (PCI or CABG) received by each patient.

METHODS AND RESULTS: We conducted a survival analysis on a cohort of patients with CAD who received either PCI or CABG treatment. For this project, we employed neural multi-task logistic regression (MTLR) to predict patient survival after receiving PCI or CABG, using clinical features such as patient demographics, cardiovascular history, and treatment type as input variables. This study utilizes real-world datasets collected by the Canadian VIGOUR Centre from 2002 to 2019 that we used to train our model.

In this study, we investigated the survival of patients with stable angina as their indication type. Of the 27199 patients, 70% received PCI, and the rest received CABG. The age of the patients ranged from 16 to 96 years old, with an average age of 65.1 years old; also, 18.9% were female. The observed mortality rates for patients who underwent PCI and CABG were 18% and 26%, respectively. Our survival model showed promising performance in predicting patient survival, with a concordance index of 0.896. These findings highlight that our algorithms can predict individual survival and assist in clinical decision-making for patients with CAD.

Conclusion: Our study findings suggest that the survival model employed in this research is effective in predicting the survival of patients receiving the target treatment by leveraging their pertinent medical history and cardiovascular conditions. Our approach integrates patient-specific information to determine their potential for survival. As a potential direction for future work and applications, the information gathered from this approach can assist clinicians in making informed decisions regarding the most suitable treatment options. Overall, our study highlights the importance of personalized medicine and the need for accurate predictive models to assist healthcare providers in delivering optimal care.