Comprehensive Summary
This study by Tokodi et al., examines the application of a machine learning (ML)-based risk stratification system to predict all-cause mortality of patients undergoing cardiac resynchronization therapy (CRT) during a 5-year follow up period. Researchers retrospectively analyzed 1510 CRT patients’ pre-implant clinical data and using 33 clinical variables, they trained and compared multiple machine learning algorithms. They found that the random forest model named the SEMMELWEIS-CRT score was the best performing model, and the researchers later tested it on an independent cohort of 158 patients. The average under the receiver operating characteristic curves for SEMMELWEIS-CRT score was 0.785 for prediction of 1-5 year post-mortality, which was significantly greater than the other evaluated scores. The model identified key predictors of mortality, including older age, reduced left ventricular ejection fraction, renal dysfunction, and lower serum sodium and hemoglobin levels.
Outcomes and Implications
Many clinicians face limitations in formulating conclusions regarding risk in an individual patient as they need to consider vast number of clinical variables associated with mortality. Therefore, accurate prediction of CRT outcomes is crucial because it may facilitate development of a more personalized approach for the risk assessment of patients rather than on a population level. ML algorithms such as the SEMMELWEIS-CRT score will be effective for medical prediction with greater potential to increase risk stratification. However, there are a few limitations which will hamster their implementation for future clinical practices as more precise and personalized methods are required.