Integrating Artificial Intelligence and Machine Learning with Causal Inference for Early Detection & Management of Cardiac Rejection

Author

Roman G. Haynatzki, University of Nebraska Medical CenterFollow

Graduation Date

Spring 5-10-2025

Document Type

Dissertation

Degree Name

Doctor of Philosophy (PhD)

Programs

Biomedical Informatics

First Advisor

John R. Windle, M.D.

MeSH Headings

Heart Transplantation, Graft Rejection, Artificial Intelligence, Machine Learning, Risk Assessment, Causal Inference

Abstract

Cardiac rejection remains a challenge for heart transplantation patients, significantly impacting their outcomes. Traditional AI/ML approaches have demonstrated promising predictive performance but are often criticized for their “black-box” nature and lack of causal interpretability. In response, this dissertation presents a novel decision-support framework that integrates advanced AI/ML techniques with causal inference methodologies inspired by Judea Pearl’s ladder of causation.

Using a comprehensive dataset that combines clinical records from the UNMC heart failure registry, EPIC electronic health records, and UNOS transplant data, our approach includes gradient-boosted models enhanced with SHAP value interpretability alongside a structural causal model framework. This dual strategy facilitates the distinction between mere correlations and actionable causal relationships, enabling counterfactual reasoning and transparent decision-making. Expert clinical insights were incorporated throughout the modeling process, ensuring that the system reflects real-world practice and can be seamlessly integrated into patient care.

Key findings include the observation that, among patients categorized as low risk by our causal inference engine, none (0 out of 14) experienced cardiac rejection, a result that underscores the efficacy of causal modeling in risk stratification. This outcome not only validates the model’s predictive accuracy but also highlights the clinical value of understanding underlying causal mechanisms.

In conclusion, the integration of AI/ML with causality-driven methods provides a transparent and interpretable tool for early detection of cardiac rejection. This framework bridges the gap between predictive performance and clinical interpretability, offering a robust platform for personalized immunosuppressive management and ultimately, improved patient outcomes.

Comments

2025 Copyright, the authors

Recommended Citation

Haynatzki, Roman G., "Integrating Artificial Intelligence and Machine Learning with Causal Inference for Early Detection & Management of Cardiac Rejection" (2025). Theses & Dissertations. 944.
https://digitalcommons.unmc.edu/etd/944