By creating realistic, synthetic patient interactions and recruitment scenarios, generative models enable simulation-based training that allows study teams to rehearse outreach and stress-test communication approaches prior to real-world deployment.

This project evaluates how artificial intelligence (AI) can improve and accelerate the matching of lung cancer patients to appropriate clinical trials. Using securely de-identified clinical data—including diagnosis, cancer stage, laboratory results, imaging, prior treatments, and biomarker information—the system is designed to predict trial eligibility with greater speed and accuracy. The study compares advanced AI approaches, including deep learning and large language models, to determine which methods perform best for trial matching, and benchmarks their performance against a leading commercial trial-matching tool to identify strengths, limitations, and areas for improvement. Ultimately, this work aims to support clinicians and health systems in identifying relevant clinical trial options more efficiently, broaden access to innovative therapies, reduce disparities in trial enrollment, and improve outcomes for patients with lung cancer.

Heart failure (HF) remains a major cause of hospitalization, readmission, and mortality. This project aims to improve risk prediction using machine learning applied to electronic health record (EHR) data. The goal is to develop a model that can predict whether patients hospitalized for HF will be readmitted within 3 or 6 months or die within 6 months after discharge, enabling earlier and more targeted clinical interventions. The model will leverage a broad range of EHR data, including demographics, clinical history, laboratory results, and medication use, and will incorporate both clinical and social determinants of health to better reflect real-world patient complexity. A key emphasis is on ensuring robust performance in diverse populations, including substantial representation of Black patients who are disproportionately affected by HF outcomes. By integrating these multidimensional data sources, the project seeks to create a more accurate and equitable predictive tool to support clinicians in identifying high-risk patients, optimizing care delivery, and ultimately reducing preventable readmissions and mortality.

Chronic obstructive pulmonary disease (COPD) remains a leading cause of hospitalization and preventable 30-day readmissions, underscoring limitations in current risk prediction methods. This work focuses on developing a next-generation multimodal approach that integrates chest imaging with electronic medical record (EMR) data to more accurately identify patients at highest risk following discharge. By combining advanced imaging analytics with clinical, social, and utilization data, the project seeks to uncover previously unrecognized patterns associated with readmissions. Machine learning methods are being used to develop and validate a scalable risk stratification tool designed for integration into clinical workflows. The ultimate goal is to enable more precise, proactive, and personalized post-discharge care, thereby reducing avoidable readmissions and improving outcomes for patients with COPD.