From statistical inference to machine learning: A paradigm shift in contemporary cardiovascular pharmacotherapy

Area of Science:

• Cardiology
• Medical Informatics
• Artificial Intelligence

Background:

• Heart failure with reduced ejection fraction (HFrEF) presents complex clinical challenges.
• Artificial intelligence (AI) offers novel methods for analyzing intricate clinical data.
• Machine learning (ML) can uncover deeper insights into HFrEF patient datasets.

Purpose of the Study:

• To explore the predictive potential of ML algorithms in HFrEF patient data.
• To identify key factors influencing long-term outcomes in HFrEF.
• To compare ML-based predictions with traditional statistical analyses.

Main Methods:

• Analysis of a cohort of 386 HFrEF patients with a minimum 6-month follow-up.
• Application of eXtreme Gradient Boosting (XGBoost) algorithms.
• Utilized Shapley additive explanations (SHAP) for model interpretation.

Main Results:

• Traditional methods showed associations but lacked independent predictive power for key pharmacological factors.
• XGBoost identified 'newly initiated treatment with sacubitril/valsartan' and 'beta-blocker (BB) dose escalation' as strong predictors.
• XGBoost effectively handled non-linear distributions, multicollinearity, and confounding factors.

Conclusions:

• ML, particularly XGBoost with SHAP, provides significant predictive power for HFrEF outcomes.
• This approach offers valuable insights beyond traditional statistical models.
• Consideration of limitations, including potential for clinically irrelevant predictors, is crucial.