tSPM+
Transitive sequential pattern mining. tSPM+ turns a patient's record into temporal sequences of clinical events and keeps the sequences that predict an outcome. It powers our Long COVID phenotyping across health systems.
Tools we build and release so other teams can test, extend, and reuse our methods.
Transitive sequential pattern mining. tSPM+ turns a patient's record into temporal sequences of clinical events and keeps the sequences that predict an outcome. It powers our Long COVID phenotyping across health systems.
Machine Learns Health Outcomes. MLHO builds interpretable outcome models from high-dimensional records through iterative feature selection, and it runs inside federated networks of hospitals.
A multi-agent system that writes, tests, and revises its own prompts for detecting clinical concepts in notes, running on local hardware so patient data stay inside the institution.
The infrastructure behind our phylophenome work, which maps how diseases follow one another across a population and stores those relationships as a temporal knowledge base.
A measure of how faithfully a diagnostic code reflects the condition it names at a given hospital and time, so that models trained on codes can be checked before they are moved to a new site.
Completeness checks for clinical research data repositories. The DQe-c paper was selected as a best paper in Clinical Research Informatics by the IMIA Yearbook of Medical Informatics.
More code from the group lives on GitHub.