Functional decline and frailty in older heart failure patients can rapidly decrease their lifespan and it’s hard to quantify by how much – but a new study in the Journal of Cardiac Failure suggests machine learning could change that.
- One of the biggest issues with HF’s disease trajectory is the decline in activities of daily living (ADL) following a patient’s first hospitalization.
- When daily activity decreases, patients quickly lose independence, which actually increases rehospitalization and mortality risk.
- That means it’s critical to predict a patient’s timeline toward this first hospitalization and take steps to prevent it for as long as possible.
That’s where this JCF study comes in, with Japanese researchers developing and training an ML model on 6k older hospitalized HF patients from the J-Proof HF Registry to predict where patients would fall into one of three one-year functional categories: maintenance, decline, or death.
- Specifically, researchers tested two different versions of the model (one with 77 variables and one with only 10) against the Kihon Checklist frailty screening tool.
- Both the 77 and 10 variable versions achieved an AUC of 0.75 versus 0.66–0.69 for the KCL tools.
Diving deeper into the model’s prediction data then revealed that physical function mattered more than cardiac function for a patient’s frailty trajectory.
- For example, variables like gait speed, preadmission frailty, and discharge ADL scores outweighed ejection fraction as predictors.
- In fact, it’s possible that a patient’s extracardiac reserve (the blood supply to organs outside the heart) largely dictated whether a patient remained independent.
Importantly, researchers also identified a subgroup of patients (n=1.7k) that had an 89% survival probability according to traditional screening tools but faced a 57% rate of functional decline.
- A mortality-focused model would flag them as low-risk but these are the people who survive heart failure just to lose their independence.
The Takeaway
With heart failure, time is of the essence, and the sooner a patient’s decline is identified, the more we can improve disease trajectory. This study tells us that machine learning can aid in predicting that decline while questioning the way we currently measure it.
