ResearchWith an overall theme of clinical evidence generation and evidence synthesis with real-world data (RWD), my research interests span: Federated learning across health systemsPrivacy-preserving distributed algorithms that combine evidence across hospitals and research networks without sharing patient-level data. Bias correction for error-prone EHR dataMethods that use surrogate and algorithm-derived phenotypes to reduce bias from misclassified outcomes in EHR-based studies. Evidence synthesis and meta-researchMethods for more timely and reliable systematic reviews and meta-analyses, covering preprints, publication bias, and fragility. News
Show older news
Publications* Equal contribution 
PDA (Privacy-Preserving Distributed Algorithms) in action: ten principles for high-quality multi-site clinical evidence generationJournal of the American Medical Informatics Association (JAMIA), 2026
Sets out 10 best-practice principles for multi-site studies that use privacy-preserving distributed algorithms (PDA), from study preparation and protocol development to analysis and reporting. 
Trial design–aware funnel plot for publication bias assessment with noninferiority or equivalence objectivesJournal of Clinical Epidemiology, 2026
Trial design–aware funnel plots that mark the decision boundaries of noninferiority and equivalence trials, to assess publication bias and inverse publication bias in meta-analyses. 
Unlocking multi-institutional insights into disease progression with PEAL as a lossless, one-shot federated learning solutionnpj Digital Medicine, 2026
PEAL fits multi-level mixed-effects models with spline trends from a single round of summary statistics, matching pooled-data results; applied to systemic sclerosis cohorts at two centers. 
An alternative method for assessing the fragility of survival analysis results: a proof-of-concept study based on the log-rank testAmerican Journal of Epidemiology, 2026
A fragility index for survival data that counts the fewest changes in event or censoring status needed to overturn the statistical significance of a randomized trial result. 
Incorporating preprints in systematic reviews: a preliminary study of a novel method for rapid evidence synthesisJournal of the American Medical Informatics Association (JAMIA), 2025
A two-stage method for including preprints in meta-analyses: weight them by a confidence score, then adjust for publication or reporting bias; shown on two COVID-19 treatments. 
Leveraging AI for Meta-Analysis: Evaluating LLMs in Detecting Publication Bias for Next-Generation Evidence SynthesisCochrane Evidence Synthesis and Methods, 2025
Tests GPT-4o and Llama 3.2 Vision on detecting publication bias from funnel plots in simulated meta-analyses; neither model detected it consistently without fine-tuning. 
Retiring the Term “Weighted Mean Difference” in Contemporary Evidence SynthesisCochrane Evidence Synthesis and Methods, 2025
Clarifies how the term "weighted mean difference" relates to the mean difference, why it persists, and recommends clearer reporting in evidence synthesis. 
Risk of neuropsychiatric and related conditions associated with SARS-CoV-2 infection: a difference-in-differences analysisNature Communications, 2025
In EHR data from 25 RECOVER institutions, children and youths with SARS-CoV-2 infection had higher risks of neuropsychiatric conditions, including anxiety, than matched COVID-19-negative peers. 
Unlocking efficiency in real-world collaborative studies: a multi-site international study with one-shot lossless GLMM algorithmnpj Digital Medicine, 2025
COLA-GLMM fits generalized linear mixed models across sites in one communication round without loss of accuracy; validated on eight international databases for COVID-19 mortality risk factors. 
DisC2o-HD: Distributed causal inference with covariates shift for analyzing real-world high-dimensional dataJournal of Machine Learning Research, 2025
A distributed algorithm for average treatment effects in high-dimensional multi-site data that accounts for covariate shift across sites and approximates the pooled estimator. 
Advancing Interpretable Regression Analysis for Binary Data: A Novel Distributed Algorithm ApproachStatistics in Medicine, 2024
ODAP-B, a one-shot distributed algorithm for relative risks of rare binary outcomes, gives less biased estimates than two-step meta-analysis, shown with data from eight medical centers. 
International Benchmarks for the Practice Environment Scale of the Nursing Work Index: A Meta-AnalysisNursing & Health Sciences, 2024
A meta-analysis of 160 publications from 38 countries that provides international benchmark values for the Practice Environment Scale of the Nursing Work Index. 
Communication-Efficient Distributed Estimation of Causal Effects With High-Dimensional DataStat, 2024
A communication-efficient estimator of the average treatment effect for multi-site data, even when covariates outnumber patients at each site, that approximates the pooled estimator at a fast rate. 
Leveraging error-prone algorithm-derived phenotypes: Enhancing association studies for risk factors in EHR dataJournal of Biomedical Informatics, 2024
Combines several error-prone EHR phenotypes with a small validated sample to reduce bias from phenotyping error, improving efficiency by up to 30% over using a single phenotype. 
Evaluating site-of-care-related racial disparities in kidney graft failure using a novel federated learning frameworkJournal of the American Medical Informatics Association (JAMIA), 2024
A federated framework with counterfactual modeling that assesses how site of care contributes to racial disparities in kidney graft failure across 73 U.S. transplant centers. 
Confidence score: a data-driven measure for inclusive systematic reviews considering unpublished preprintsJournal of the American Medical Informatics Association (JAMIA), 2024
A data-driven "confidence score," estimated with a survival cure model, for weighting preprints in systematic reviews and meta-analyses; validated on 146 COVID-19 therapeutics preprints. 
Real-World Effectiveness of BNT162b2 Against Infection and Severe Diseases in Children and AdolescentsAnnals of Internal Medicine, 2024
Estimates the real-world effectiveness of the BNT162b2 vaccine against infection and severe disease in children and adolescents in PEDSnet during the Delta and Omicron periods. 
Quantifying and correcting bias due to outcome dependent self-reported weights in longitudinal study of weight loss interventionsScientific Reports, 2023
A framework to detect self-reported weights that are missing not at random and correct the resulting bias, applied to the Keep It Off trial of financial incentives for weight-loss maintenance. 
Food insecurity and binge eating: A systematic review and meta-analysisInternational Journal of Eating Disorders, 2023
A systematic review and meta-analysis finding that food insecurity is associated with higher odds of binge eating and binge-eating disorder in adults. 
Adding caplacizumab to standard of care in thrombotic thrombocytopenic purpura: a systematic review and meta-analysisBlood Advances, 2023
A meta-analysis of trials and observational studies on adding caplacizumab for immune TTP: faster response and fewer refractory cases, with more bleeding in randomized trials. 
Distributed learning for heterogeneous clinical data with application to integrating COVID-19 data across 230 sitesnpj Digital Medicine, 2022
dCLR, a one-round distributed conditional logistic regression that accounts for between-site heterogeneity, applied to 14,215 COVID-19 patients from 230 clinical sites. 
A cost-effective chart review sampling design to account for phenotyping error in electronic health records (EHR) dataJournal of the American Medical Informatics Association (JAMIA), 2022
An outcome-dependent sampling design that uses EHR-derived phenotypes to choose which charts to review, making validation more efficient than sampling charts uniformly. 
Identifying Clinical Risk Factors for Opioid Use Disorder using a Distributed Algorithm to Combine Real-World Data from a Large Clinical Data Research NetworkAmerican Medical Informatics Association (AMIA) Annual Symposium Proceedings, 2020
Applies the one-shot distributed algorithm ODAL to OneFlorida data to identify risk factors for opioid use disorder, with better estimates than meta-analysis. 
An augmented estimation procedure for EHR-based association studies accounting for differential misclassificationJournal of the American Medical Informatics Association (JAMIA), 2020
An augmented estimator that combines a small chart-reviewed sample with error-prone EHR phenotypes, reducing bias in risk-factor associations while keeping variance low. 
Robust-ODAL: Learning from heterogeneous health systems without sharing patient-level dataPacific Symposium on Biocomputing (PSB), 2020
A privacy-preserving, communication-efficient distributed algorithm that accounts for heterogeneous sites, outperforming ODAL and meta-analysis on OHDSI claims data. 
Inherited Thrombophilia and the Risk of Arterial Ischemic Stroke: A Systematic Review and Meta-AnalysisJournal of the American Heart Association, 2019
A meta-analysis of 68 studies finding that adults with arterial ischemic stroke more often carry inherited thrombophilias such as factor V Leiden and protein C or S deficiency. 
Identification of Rare Adverse Events with Year-varying Reporting Rates for FLU4 Vaccine in VAERSAmerican Medical Informatics Association (AMIA) Annual Symposium Proceedings, 2018
Uses a likelihood ratio test on VAERS reports to identify rare adverse events with year-varying reporting rates for the quadrivalent FLU4 influenza vaccine. Honors & Awards
Teaching
|
