Predicting in-hospital all-cause mortality in heart failure using machine learning.

Area of Science:

• Cardiology
• Biomedical Engineering
• Data Science

Background:

• Heart failure presents unique challenges in low- and middle-income countries (LMICs), including earlier onset and higher mortality rates.
• There is a critical need for innovative risk stratification tools tailored to LMIC populations.
• This study explored the application of machine learning for predicting mortality in hospitalized heart failure patients.

Purpose of the Study:

• To evaluate the effectiveness of supervised machine learning algorithms in predicting all-cause mortality among heart failure patients.
• To identify key predictors of mortality in this patient cohort.
• To lay the groundwork for a region-specific heart failure risk prediction tool.

Main Methods:

• Trained six supervised machine learning algorithms on data from 500 heart failure patients.
• Utilized patient demographics, clinical data, and electrocardiogram (ECG) findings.
• Focused on predicting in-hospital all-cause mortality for patients with left ventricular ejection fraction (LVEF) < 50%.

Main Results:

• Support vector machines (SVM) demonstrated the highest predictive accuracy (86%) and Area Under the Curve (AUC) of 0.77.
• Random forest achieved the highest AUC (0.82) with 88% accuracy.
• Key predictors included medications (furosemide, beta-blockers, spironolactone), clinical signs (diastolic murmur, parasternal heave), and comorbidities (coronary artery disease, ischaemic cardiomyopathy).

Conclusions:

• Supervised machine learning algorithms show promise in predicting heart failure mortality, even with a modest sample size.
• The SVM model warrants external validation across multiple South African cardiology centers.
• Development of a "uniquely African" risk prediction tool could revolutionize heart failure management via precision medicine.