Heart failure (HF) is a significant problem, contributing to one in every eight deaths in the United States, but the risk prediction models currently available for HF remain limited at best.
With this ongoing issue in mind, the authors of a new study published in JAMA Network Open aimed to find out if machine learning-based prediction models could provide better results than traditional logistic regression.
Lead author Rishi J. Desai, MS, PhD, Harvard Medical School in Boston, and colleagues tracked data from more than 9,000 Medicare-enrolled patients with HF from Jan. 1, 2007, to Dec. 31, 2014. The mean patient age was 78 years, more than 54% of patients were women and more than 91% were white.
Desai et al. compared the performance of claims-only prediction models with models that used machine learning techniques and incorporated additional predictors from electronic medical records (EMRs). The machine learning models tested in this study included least absolute shrinkage and selection operator (LASSO), classification and regression tree, random forest and gradient-boosted model. In all instances, the primary outcomes the authors wanted to predict were all-cause mortality, HF hospitalization, high cost (top cost decile) and home days loss (greater than 25%).
Overall, the team found that machine learning “offered only limited improvement” over logistic regression. The GBM model “consistently provided the highest discrimination and lowest Brier scores across all four outcomes,” followed by the random forest and LASSO models.
“Although augmenting claims data with detailed EMR-derived predictors resulted in notable improvement in model performance for certain outcomes, including mortality and home days loss, such improvement was not seen for prediction of high future costs,” the authors wrote.
The team added that the models they developed for this study could potentially be used by outside parties to track certain data.
“For instance, an insurer who is interested in deploying interventions, such as home nurse visits, to ensure optimal HF management and downstream cost savings may benefit from identifying a population with a high one-year risk of HF hospitalization based on their administrative data using models from this study to possibly ensure the most efficient use of finite resources,” Desai and colleagues wrote.
Michael has more than 16 years of experience as a professional writer and editor. He has written at length about cardiology, radiology, artificial intelligence and other key healthcare topics.