ADAPT Researchers Introduce New Framework to Enhance Healthcare Decision-Making

05 February 2025

ADAPT Researchers Introduce New Framework to Enhance Healthcare Decision-Making

A new study led by researchers from ADAPT at Munster Technological University (MTU) has introduced a novel framework, Probabilistic Causal Fusion (PCF), which integrates Causal Bayesian Networks (CBNs) and Probability Trees (PTrees) to improve healthcare decision-making.  The research, recently published on arXiv, an open-access repository for scientific papers, highlights how this approach can provide clinicians with not only predictive insights but also a deeper understanding of the causal relationships influencing patient outcomes.

The study titled “Integrating Probabilistic Trees and Causal Networks for Clinical and Epidemiological Data”, is co-authored by Sheresh Zahoor (MTU), Pietro Liò (University of Cambridge), Gaël Dias (Université Caen Normandie), and Mohammed Hasanuzzaman (Queen’s University Belfast). The researchers tested PCF on real-world medical datasets, including MIMIC-IV, the Framingham Heart Study, and Diabetes data. The results showed that PCF performs on par with traditional machine learning models while offering additional causal reasoning capabilities.  This allows clinicians to simulate hypothetical scenarios, such as the impact of lifestyle changes or medical interventions, before making critical decisions.

One of the key innovations of PCF is its use of sensitivity analysis and SHapley Additive exPlanations (SHAP) to improve interpretability.  Sensitivity analysis helps quantify the influence of causal parameters on key health outcomes, such as Length of Stay (LOS), Coronary Heart Disease (CHD), and Diabetes, while SHAP highlights the most important predictive features.

By bridging the gap between traditional clinical intuition and data-driven AI models, PCF provides a new tool to support more informed and evidence-based medical decision-making.

The full study is available here: arXiv link.