Tailored Risk Stratification in Severe Mitral Regurgitation and Heart Failure Using Supervised Learning Techniques.

Area of Science:

• Cardiology
• Medical Informatics
• Machine Learning

Background:

• Secondary mitral regurgitation (sMR) significantly impacts heart failure (HF) patient outcomes, including quality of life, rehospitalizations, and mortality.
• Identifying high-risk patient cohorts is crucial for understanding disease progression and implementing effective risk stratification strategies.

Purpose of the Study:

• To develop a structured, decision tree-like approach for risk stratification in patients with severe sMR and HF.
• To identify distinct patient subgroups with varying mortality risks within the HF spectrum.

Main Methods:

• An observational study involving 1,317 patients with severe sMR across the full spectrum of HF.
• Clinical, echocardiographic, and laboratory data were collected.
• Survival tree analysis, a supervised machine learning technique, was employed to identify mortality risk subgroups, stratified by HF subtype.

Main Results:

• Eight distinct patient subgroups with significantly different long-term survival rates were identified using survival tree analysis.
• Subgroup 7 (younger, higher hemoglobin and albumin) exhibited the best survival.
• Subgroup 5 (older, low albumin, high NT-proBNP) showed a 20-fold increased mortality risk (HR: 20.38).
• Unique risk subgroups were identified for HF with preserved, mildly reduced, and reduced ejection fraction.

Conclusions:

• Supervised machine learning effectively reveals significant heterogeneity in mortality risk among patients with sMR and HF.
• A decision tree-like model provides a valuable tool for tailored risk stratification by differentiating outcomes among identified subgroups.