Heart failure (HF) continues to be a significant global health challenge, contributing to high rates of morbidity, mortality, and healthcare costs. Patients who seek care in the emergency department (ED) or are hospitalized for HF face a greater risk of future adverse events, but accurately assessing and categorizing their risk level remains a complex task. In order to address this, researchers in a recent publication in PLOS Digital Health utilized a machine learning (ML) approach to predict 30-day and 1-year risks of ED visits, hospital readmissions, or death in HF patients.
The study utilized administrative healthcare data from a large regional system in Alberta, Canada that included over 50,000 HF patients who visited the ED or were hospitalized between 2002 and 2016. The researchers employed deep feature synthesis to extract health data from multiple sources and compared the performance of several machine learning algorithms, including gradient boosting (CatBoost), deep neural networks, and conventional logistic regression modeling.
The study found that CatBoost consistently outperformed logistic regression across all performance metrics and endpoints. This model could potentially address longstanding challenges in risk assessment and the utilization of healthcare resources, however, additional research employing administrative health data at a larger population level is needed to determine if it can be effectively applied across large regional healthcare systems to reduce adverse events following hospital discharge for HF.
The lead authors of this publication are Nowell Fine, MD, SM (University of Calgary) and the CVC’s Sunil Vasu Kalmady, PhD. Co-authors include the CVC’s Weijie Sun (PhD student), Finlay McAlister, MD, MSc, Justin Ezekowitz, MBBCh, MSc, and Padma Kaul, PhD, along with Russ Greiner, PhD (University of Alberta), and Jonathan Howlett, MD and James White, MD (University of Calgary).
