CAREER: An Integrated Machine Learning Framework for Longitudinal Healthcare Intelligence: Representation, Alignment, and Robustness

In modern healthcare, analyzing vast amounts of longitudinal patient data, such as diagnoses, prescriptions, and medical procedures, can help healthcare providers understand complex health conditions and make treatment decisions. Artificial intelligence (AI) has great potential to assist in this process by discovering hidden patterns in a patient's medical history to provide personalized care recommendations. However, current AI systems often struggle to capture the complex structure of medical records, may produce recommendations that conflict with established medical knowledge, and can become unreliable when faced with noisy or unexpected data. This project addresses these challenges by developing an integrated framework to make AI in healthcare more reliable and medically accurate. By improving the dependability of machine learning frameworks, the project serves the national interest by advancing health and welfare, and enabling more effective and personalized patient care. The project also supports education by training undergraduate and graduate students to prepare the next generation of healthcare technology innovators. This project will develop a novel, comprehensive machine learning framework to improve healthcare decision support systems using electronic health record data. The research will pursue three complementary activities spanning data representation, algorithm alignment, and system robustness. First, the team of researchers will design deep learning architectures that jointly model patient visit sequences and relational graphs of clinical events to create richer data representations. Second, the project will integrate medical knowledge graphs and large language models into the training process, ensuring algorithmic reasoning aligns with established clinical principles. Finally, the research will evaluate system vulnerabilities through adversarial testing and incorporate contrastive training to maintain predictive consistency under adverse conditions. The framework will be rigorously evaluated across diverse clinical applications. Project outcomes will be disseminated to the broader research community through open-source software releases and integrated into interdisciplinary academic curricula. This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria. NSF Award ID: 2544634 | Program: 01002930DB NSF RESEARCH & RELATED ACTIVIT,01002627DB NSF RESEARCH & RELATED ACTIVIT | Principal Investigator: Zijun Yao | Institution: University of Kansas Center for Research Inc, LAWRENCE, KS | Award Amount: $385,250 View on NSF Award Search: https://www.nsf.gov/awardsearch/show-award/?AWD_ID=2544634 View on Research.gov: https://www.research.gov/awardapi-service/v1/awards/2544634.html