Projects & Collaboration
Active Research Programs
We welcome collaborations across institutions. Below is a summary of ongoing and completed projects using the PDA framework.
Clinical modeling and risk prediction
Forecast patient outcomes, powered by shared models not shared data.
- Binary outcomes: ODAL
- Time-to-event: ODAC, ODACH, ODACoR, ODACoRH
- Count / categorical: ODAP, ODACAT
- Zero-inflated outcomes: ODAH
- High-dimensional predictors: ADAP
Causal inference and comparative effectiveness research
Discover real-world treatment effects across systems, with privacy.
- Target trial emulation with binary / survival outcomes
- Self-controlled case series design for vaccine safety outcomes
- Test-negative design for vaccine efficacy outcomes
- Federated counterfactual modeling for
Phenotyping and subtype discovery
Understand disease heterogeneity without centralized access to data.
- Heterogeneous latent class models
- One-shot EM clustering
- Federated latent transfer learning
- Subphenotyping under hospital-specific patterns
Cohort characterization
Summarize patient cohort characteristics.
- DiscoChar: continuous variables
- DiscoChar: categorical variables
- DiscoChar: time-to-event variables
Federated Pharmacovigilance: GLP-1RA vs. DPP4i
Multi-site causal inference comparing cardiovascular outcomes of GLP-1 receptor agonists and DPP-4 inhibitors. Uses negative control outcomes for bias correction and factor model approaches for unmeasured confounding.
COMPASS: ICU Lab Test Utilization
Characterizing patterns of lab test ordering in ICU settings across Penn Medicine. Combines DuckDB pipelines with CCSR diagnostic classification to identify high-frequency testing patterns.
COVID-19 Vaccine Safety Surveillance
Self-controlled case series analysis of adverse cardiac events following mRNA vaccination, implemented as a federated SCCS model deployable across multiple health systems.
ADRD Drug Repurposing Initiative
Leveraging real-world EHR data and federated distributed algorithms to identify drug candidates for Alzheimer’s disease and related dementias. Collaboration across major academic medical centers.
Telemedicine & Elastic Utilization
Causal analysis using T-learner (XGBoost) methods to decompose telemedicine’s effect on follow-up care into behavioral pathways (convenience vs. escalation) across high-virtualizability clinical domains.
Multi-Agent LLM Research Framework
Exploring ring and star topologies for multi-agent AI systems in biomedical research contexts, with focus on agent disagreement mechanisms and mitigating hallucination in evidence synthesis.
Federated Tensor Train Decomposition for CAR Logistic Regression
Novel federated framework incorporating Tensor Train decomposition for high-dimensional covariate adjustment in distributed logistic regression. Deployed across GLP-1, SGLT2, and DPP4 treatment comparisons.