Surfacing risk and patterns from longitudinal health data.

Mental Health · Active

Suicide risk prediction & phenotyping

Using the Maryland Suicide Data Warehouse and large-scale claims data to identify high-risk clinical trajectories, characterize decedent subtypes with deep embedded clustering, and surface temporal condition patterns associated with suicide death.

Mental Health · Active

Digital monitoring & youth mental health

County-level evaluation of digital monitoring tools (e.g., GoGuardian Beacon) used in U.S. K-12 schools to identify students at risk of self-harm, using difference-in-differences and quasi-experimental designs.

Veteran Health

Veteran mental health & multimodal data

Models that predict suicidal ideation in U.S. veterans from acoustic and linguistic features of speech; mobile health platforms for multimodal patient data collection and delivery of behavioral interventions (PTSD, Gulf War Illness).

Oncology

Real-world oncology evidence

Integrated registries for immune checkpoint inhibitors, NLP pipelines for adverse-event extraction from clinical notes, and precision-medicine infrastructure (G-DOC Plus) for multi-omics + clinical data analysis.

Federated Learning

Federated learning for medical AI

Contributed to the Federated Tumor Segmentation (FeTS) Challenge and Intel/Penn federated learning consortium — enabling cross-institution model training on rare cancers without centralizing PHI.

Social Determinants

Social determinants & health disparities

Augmenting clinical prediction models (e.g., LACE for readmission) with social determinants of health, and quantifying racial / socioeconomic disparities in postpartum depression using population-level discharge data.