Davis Receives $1.6M Grant for AI Model Monitoring

A four-year National Library of Medicine grant will fund methods to track and correct accuracy gaps in clinical AI models across patient groups.

Updated on Nov. 3, 2026 in Artificial Intelligence

Davis Receives $1.6M Grant for AI Model Monitoring

Sharon Davis, Assistant Professor of Biomedical Informatics, has received a four-year, $1.6 million research grant from the National Library of Medicine, part of the National Institutes of Health. The project addresses how clinical prediction models can lose accuracy over time and across different patient populations.

Why it matters

Clinical AI models can degrade as medical practice, information systems and patient populations change, and they may perform poorly for certain clinical, geographic or sociodemographic groups. Methods to monitor and correct these gaps aim to keep such tools reliable for the patients who use them.

The project uses VA and Vanderbilt Health data to study prediction models for surgical complications, hospital-acquired acute kidney injury and postpartum hemorrhage, and synthetic patient data is used to monitor small groups such as rural hospital patients. Whether the team's methods will restore accuracy both overall and within subgroups remains untested.

The players

Sharon Davis, PhD

Davis is an Assistant Professor of Biomedical Informatics who leads research on monitoring and updating clinical AI models for accuracy across patient populations.

National Library of Medicine

The National Library of Medicine is the world's largest biomedical library and a component of the National Institutes of Health that funds health informatics research.

Department of Veterans Affairs

The Department of Veterans Affairs is the federal agency that operates the VA health system, whose data and national acute kidney injury risk model are used in the research.

The details

Davis and colleagues previously built data-driven methods that detect drift in deployed models and recommend updates, an approach that has helped maintain a national Department of Veterans Affairs risk model for acute kidney injury. The new grant extends that work to track how performance gaps between patient groups change over time in both traditional machine learning models and large language models.

November 2, 2026: Announcement of the four-year research grant award.

Biotech
Nov. 3
Vanderbilt Team Launches MutexaGPT for Enzyme Design

Roadmap

The project extends earlier VA work on deployed model drift from monitoring overall accuracy to monitoring subgroup gaps, a shift toward fairness-aware maintenance of clinical AI. It also brings large language models into the same monitoring framework used for traditional machine learning tools.

Patients whose care relies on clinical prediction models may benefit if monitoring methods keep those tools accurate for their own demographic and clinical group. Hospital teams using these tools could eventually receive updates when a model's performance begins to drift.

The takeaway

Clinical AI tools need ongoing checks, not just one-time validation, to stay accurate for every patient group they serve. Hospitals and clinicians should ask how a prediction tool is monitored after deployment.