Artificial intelligence-based clustering to identify functional risk phenotypes in heart failure.

Area of Science:

• Cardiology
• Artificial Intelligence
• Biostatistics

Background:

• Heart failure (HF) patients often experience undetected cardiorespiratory fitness (CRF) decline, increasing poor outcome risks.
• Current clinical practices lack effective tools for early CRF risk stratification in HF patients.

Purpose of the Study:

• To identify novel cardiorespiratory fitness (CRF) risk phenotypes in heart failure (HF) patients using AI.
• To develop and validate an interpretable and generalizable risk stratification model for early functional assessment in HF.

Main Methods:

• An AI-driven unsupervised clustering analysis was performed on 505 HF patients using 15 multimodal clinical variables.
• Associations between identified phenotypes and CRF impairment (VO2 max ≤20 mL/kg/min) were evaluated using logistic regression and machine learning models.
• SHapley Additive exPlanations (SHAP) analysis ensured model interpretability, with external validation in 201 HF patients.

Main Results:

• Three distinct HF phenotypes were identified: balanced, inflammatory-sarcopenic, and metabolically dysregulated.
• Both non-balanced phenotypes demonstrated significantly higher odds of impaired cardiorespiratory fitness (VO2 max).
• Machine learning models, including random forest and XGBoost, showed strong discriminative performance (AUC ≈ 0.75) in derivation and validation cohorts.

Conclusions:

• AI-driven integration of multimodal data successfully identified novel CRF risk phenotypes in HF patients.
• A highly interpretable and generalizable risk stratification model was established for early functional assessment.
• These findings provide a framework for precision rehabilitation strategies in heart failure management.