Machine Learning Tool Predicts Heart Failure Treatment Response

A research team from The Texas Heart Institute recently developed a machine learning (ML) tool capable of characterizing and predicting diuretic responsiveness in individuals with acute decompensated heart failure (ADHF).

ADHF is characterized by the presence of too much fluid in the body, and the condition is a major driver of hospitalization, morbidity, and mortality. Treatment for ADHF typically involves using diuretic drugs to relieve and manage heart congestion.

However, patients have different responses to these treatments, and some exhibit diuretic resistance. Treating these patients is a challenge, as there is a lack of consensus around how clinicians should approach diuretic resistance in ADHF patients who are hemodynamically stable, the researchers noted.

The study further indicated that in these cases, it is often recommended that clinicians should work to optimize the dosage of loop diuretics that patients are receiving before moving to consider combination therapy. But there is no consensus around how much the dosage should be adjusted before introducing other diuretic drugs.

“Inefficient diuretic response in hospitalized patients can hinder treatment progress and increase the risk of post-discharge rehospitalization and mortality. It’s crucial to identify individuals with low diuretic efficiency early on to tailor decongestion strategies and improve clinical outcomes,” explained Matthew Segar, MD, a third-year cardiovascular disease fellow at The Texas Heart Institute, in the news release.

To help predict treatment responsiveness and improve outcomes for patients with ADHF, the research team turned to ML.

Using de-identified, publicly available clinical trial data from the National Heart, Lung, and Blood Institute (NHLBI) Biologic Specimen and Data Repository Information Coordinating Center, the researchers built an ML-based diuretic efficiency phenomapping approach and diuresis score.

The model, known as BAN-ADHF, leveraged this information to identify and classify patients into subgroups based on diuretic efficiency. Patients within each group had similar characteristics, but were found to be clinically distinct from those outside of their subgroup in terms of diuretic therapy responsiveness.

Patients also demonstrated significant differences in clinical outcomes, such as length of stay and mortality. Patients in the group with the highest diuretic resistance had notable increases across both metrics compared to their peers.

The research team underscored that these findings show how the tool’s ability to predict a patient’s probability of being in the group with the least diuretic response may have significant prognostic utility.

“We know the BAN-ADHF score can accurately identify, characterize, and predict diuretic resistance among individuals with ADHF mathematically. Now we must take this medical knowledge and conduct a clinical study to evaluate whether implementing the BAN-ADHF score in our care protocols improves outcomes for patients hospitalized with acute decompensated heart failure,” said Segar.

ML and other AI tools have demonstrated high potential in a host of cardiovascular care-related use cases.

Last week, researchers from the University of Michigan shared that an ML model they built can accurately predict death, major bleeding events, and the need for blood transfusion in patients undergoing percutaneous coronary intervention (PCI).

PCI is minimally invasive and considered to be less risky than other approaches to angioplasty and stent placement, but it still carries significant risks that clinicians can struggle to stratify effectively.

The ML model is designed to provide support to these clinicians by predicting patients’ risk for adverse post-PCI outcomes. The model also incorporates patient feedback to improve the shared decision-making process.

The tool is freely available as a smartphone and computer application for researchers.