CLAI
about us
Our goal at CLAI is to build clinical AI that augments human intelligence: AI that understands time, place, and the limits of its own data. We integrate clinical knowledge and phenomenological theory with state-of-the-art machine learning and large language models, and we test our methods across health systems before we trust them. We pursue this goal through four research foci:
1- Temporal phenotyping. Patients change over time, and so should our models of them. We develop methods that read the electronic health record as a sequence of events, from transitive sequential pattern mining (tSPM) and MLHO to recovery trajectories after acute illness and phenotype networks that map how diseases follow one another.
2- Agentic clinical AI. We build autonomous multi-agent systems that read clinical notes, critique each other, and optimize their own prompts while running on local hardware, so patient data stay within the institution. We also study how these systems fail, including optimization instability when the conditions they look for are rare.
3- Data fidelity and generalization. The meaning of a diagnostic code varies from one hospital to the next. We measure diagnostic signal quality with tools such as the Signal Fidelity Index, evaluate fairness and bias, and validate models across federated networks of dozens of hospitals.
4- Environmental health informatics. We link high-resolution environmental exposures, such as extreme heat and air pollution, to clinical records to study how the environment shapes health in aging adults, people living with dementia, and patients of community health centers.
CLAI is based in the Department of Medicine at Massachusetts General Hospital and Harvard Medical School. Its director, Hossein Estiri, is an associate member of the Broad Institute of MIT and Harvard.
Funding
CLAI’s research is supported by the National Institutes of Health, through the National Institute on Aging, the National Institute of Allergy and Infectious Diseases, the National Institute of Environmental Health Sciences, and the National Institute on Minority Health and Health Disparities, and by NVIDIA. Content on this site is solely the responsibility of the authors and does not necessarily represent the official views of the National Institutes of Health.