AI Uncovers Hidden Crisis in Heart Failure Care: Thousands Survive Hospitalization Only to Lose Their Independence

For older adults hospitalized with heart failure, survival is only half the battle. A machine learning model from Juntendo University has revealed a previously overlooked population: nearly 2,000 patients who survive the year following discharge but lose their independence—unable to perform daily activities despite beating the odds on mortality.

The research team, led by Assistant Professor Kanji Yamada and colleagues from Juntendo's Department of Physical Therapy, developed an XGBoost algorithm that analyzes 77 clinical factors to predict both functional decline and death one year after hospital discharge. Published in the Journal of Cardiac Failure in August 2026, the model was validated on 6,382 patients aged 65 and older from Japan's J-Proof HF Registry—all of whom were independent before their heart failure hospitalization and received physical therapy during their stay.

A New Risk Category Emerges

The model's most striking discovery was the identification of four distinct risk phenotypes, including a "low-mortality/high-decline" group of 1,732 patients. While 89% of these individuals survived the year, 57% experienced functional decline severe enough to compromise their independence.

This represents a critical gap in traditional heart failure prognosis. Conventional assessments focus primarily on mortality risk, leaving clinicians without tools to identify patients who will survive but struggle to maintain the functional capacity needed for independent living.

The machine learning approach outperformed traditional frailty assessments, including the modified Kihon Checklist (KCL) score and total KCL score, by prioritizing physical function measures over cardiac metrics. Surprisingly, maximum gait speed, preadmission frailty scores, discharge Barthel Index scores, Short Physical Performance Battery results, and handgrip strength proved more predictive than cardiac ejection fraction—the standard measure of heart pump function.

From Algorithm to Clinical Action

The research team refined their initial 77-predictor model into a streamlined "Top-10" version and developed a prototype web application that clinicians can use at discharge. The tool not only estimates a patient's risk profile but includes simulation capabilities that show how improvements in specific areas—such as gait speed—might alter outcomes.

This feature transforms the model from a passive prediction tool into an active planning resource. Physical therapists and cardiologists can use it to set rehabilitation targets and demonstrate to patients the potential functional benefits of intensive post-discharge therapy.

Of the 6,382 patients studied, 3,378 maintained their independence through the one-year follow-up period, while 2,076 experienced functional decline and 928 died. The model's nested leave-one-site-out cross-validation approach—testing the algorithm's performance when applied to data from medical centers it hadn't been trained on—demonstrated robust generalizability across different hospital settings.

Implications for Care Delivery

The findings carry significant implications for how healthcare systems allocate rehabilitation resources and structure discharge planning. Rather than concentrating support solely on patients at highest mortality risk, the model enables targeted intervention for those facing independence loss—an outcome that profoundly affects quality of life, caregiver burden, and long-term care costs.

For patients and families, the tool offers a framework for realistic post-discharge planning conversations that extend beyond survival to address the practical question: Will I be able to live independently in my own home?

As healthcare increasingly embraces AI-driven decision support, this research demonstrates how machine learning can reveal clinical realities hidden within traditional risk stratification approaches—and potentially preserve independence for thousands of older adults navigating life after heart failure hospitalization.

Sources

1. Machine learning can identify older heart failure patients at risk of losing independence — News-Medical.Net