Integrating probabilistic trees and causal networks for clinical and epidemiological data

Area of Science:

• * Computational biology and bioinformatics
• * Machine learning and artificial intelligence in healthcare
• * Causal inference and probabilistic modeling

Background:

• * Traditional machine learning (ML) models excel at prediction but struggle with 'what if' scenarios regarding interventions.
• * Accurate healthcare decision-making requires understanding how factors influence patient outcomes, not just predicting them.
• * Existing predictive models lack the ability to quantify the impact of specific interventions or simulate hypothetical scenarios.

Purpose of the Study:

• * To introduce the Probabilistic Causal Fusion (PCF) framework, integrating Causal Bayesian Networks (CBNs) and Probability Trees (PTrees).
• * To enable prediction, quantification of factor impacts, and simulation of hypothetical interventions.
• * To provide a unified toolkit for clinical decision support, enhancing interpretability and causal reasoning.

Main Methods:

• * Integration of Causal Bayesian Networks (CBNs) for causal relationships and Probability Trees (PTrees) for structured prediction.
• * Evaluation on diverse real-world datasets: MIMIC-IV, Framingham Heart Study, and BRFSS (Diabetes).
• * Incorporation of sensitivity analysis and SHapley Additive exPlanations (SHAP) for dual-layered interpretability.

Main Results:

• * PCF demonstrated consistent predictive performance comparable to conventional ML models across three datasets.
• * The framework successfully quantified factor impacts and simulated hypothetical interventions.
• * Enhanced interpretability was achieved through sensitivity analysis (macro-level causal pathways) and SHAP (micro-level feature importance).

Conclusions:

• * PCF offers a unified approach for prediction, intervention modeling, and counterfactual analysis in clinical decision support.
• * The framework bridges the gap between clinical intuition and data-driven insights by combining causal reasoning with predictive modeling.
• * PCF supports more informed, evidence-based decision-making by clarifying relationships between modifiable factors and simulating intervention outcomes.