Mortality Prediction in Patients With or Without Heart Failure Using a Machine Learning Model.

Area of Science:

• Cardiology
• Medical Informatics
• Public Health

Background:

• Existing risk prediction models often target specific conditions, limiting their use in general patient populations.
• The MARKER-HF model was initially developed for heart failure (HF) patients.

Purpose of the Study:

• To evaluate the MARKER-HF model's capability in predicting 1-year mortality.
• To assess its performance in a large, community-based hospital registry including patients with and without HF.

Main Methods:

• Analysis of 41,749 consecutive patients undergoing echocardiography.
• Inclusion of patients with (n=4,640) and without HF (n=37,109).
• Subgroup analysis of non-HF patients based on cardiovascular disease, acute coronary syndrome, atrial fibrillation, COPD, CKD, diabetes, hypertension, and malignancy.

Main Results:

• MARKER-HF showed strong predictive performance for 1-year mortality in both HF (AUC=0.729) and non-HF patients (AUC=0.770).
• Consistent accuracy was observed across various subgroups, including those with cardiovascular disease and common comorbidities.
• Patients with malignancy exhibited higher mortality rates at similar MARKER-HF scores.

Conclusions:

• The MARKER-HF model effectively predicts mortality in heart failure patients.
• Its predictive capability extends to patients without heart failure, including those with various other diseases.
• MARKER-HF offers a versatile tool for mortality risk assessment across a broad spectrum of patients.