Research

My research asks how real-world data and AI can make clinical evidence more reliable: how studies are designed, what they measure, how they are run, and how their results are judged.

How to read the status lines

  • Established work
  • Current research
  • Proposed research
Depression care as a set of possible pathsSix stages from symptoms to outcomes. A dashed path returns from engagement to recognition after relapse, and any stage can end in leaving observed care.Symptomsemerge or worsenRecognitionscreening or visitDiagnosisrecorded, or notTreatment starttherapy or drugEngagementkept up, or notOutcomessymptoms, functionrelapse, returnmay leave careAny step can be skipped, repeated or left

Conceptual illustration

Depression care is a set of paths, and each data source sees part of it

Mental Health Care and Outcomes

Question
How do adolescents and new mothers with depression move through care, and how can routine data measure that path and where it leads?
Status
In development
Approach
Longitudinal cohorts in EHR and claims data, care-cascade measures with explicit time windows, NLP and large language models for clinical notes

With collaborators in mental health research, I am planning studies of adolescent and maternal and postpartum depression that follow recognition, treatment start, engagement and outcomes in real-world data, and testing whether language models can recover the symptoms and outcomes that only clinical notes record.

Read the full direction

RequestGLP-1 receptor agonists and glycemic control

Identify adults with type 2 diabetes2 who started a GLP-1 receptor agonist1 between 2019 and 2023. Exclude anyone with prior insulin use3. Baseline HbA1c is the most recent value on record5; the outcome is HbA1c six months later6. Also include the adolescent endocrinology cohort4 so the sample is large enough.

  1. Missing decision
  2. Ambiguous definition
  3. Undefined window
  4. Internal contradiction
  5. Potential error
  6. Observability

Synthetic example

Reviewing a data request before any data are pulled

AI-Assisted Research Design and Data Quality

Question
Can an AI assistant find the missing decisions, unclear definitions and contradictions in a real-world data request before anyone builds the dataset?
Status
Proposal in development
Approach
A set of real-world data requests reviewed three ways (expert reviewers, a structured template, an AI copilot) and scored against a reference list of problems

I am developing a University of Florida seed-funding proposal to test whether an AI copilot can review researchers' data requests as well as expert reviewers and structured templates do. The idea is at an early stage and there are no results yet.

Read the full direction
A one-time count versus recruitment opportunity over timeSynthetic curves over 24 months. A one-time database count is flat. Patients already in the record who remain eligible decline over time, while newly eligible patients accumulate, so the true opportunity first falls below and later rises above the one-time count. A database that is refreshed with a delay shows fewer patients than are really available; the shaded band is the difference.0612182401×Months after the trial opensEligible patients, relative to the countRecruitment opportunityOne-time query countWhat the data showExisting patients still eligibleA one-time count versus recruitment opportunity over time0612182401×Months after the trial opensRecruitment opportunityOne-time query countWhat the data showExisting patientsNot yet visible in the data

The shaded band is the part of the opportunity a delayed data refresh cannot yet see.

Synthetic example

From a one-time count to recruitment opportunity over time

Clinical Trial Design and Feasibility

Question
How many patients could a trial actually enroll, and when, once eligibility over time, new patients and the age of the data are taken into account?
Status
Proposal submitted
Approach
Eligibility tracked over time in EHR data, incident patients and data freshness in the estimate, calibration against observed trial accrual

A UF-FSU CTSA pilot proposal, From Population to Participant: Calibrating Real-World Data for Trial Feasibility and Accrual, has been submitted. It would turn one-time database counts into time-varying estimates of recruitment opportunity and check them against real accrual.

Read the full direction

Selected foundations

Earlier, published work in clinical prediction, clinical NLP, trial eligibility and healthcare quality measurement. Each directly informs one of the directions above.

All publications

  1. Alzheimer's & Dementia, 2023 · First author

    Early prediction of Alzheimer's disease and related dementias from electronic health records

    Machine learning models trained on OneFlorida+ EHR data to predict Alzheimer's disease and related dementias up to five years before diagnosis, comparing knowledge-driven and data-driven feature sets across four computable phenotypes.

  2. BMC Medical Informatics and Decision Making, 2019 · Co-author

    De-identifying clinical notes across institutions with deep learning

    How well deep learning de-identification models trained on one institution's notes transfer to another's, and what it takes to adapt them, using the 2014 i2b2/UTHealth corpus and a new corpus of University of Florida Health notes.

  3. 2025 · Analysis and visualization

    Visualizing healthcare quality trends in Texas Medicaid

    Small-multiple and bubble charts of potentially preventable admissions, visits and readmissions across Texas Medicaid service areas, prepared for the 2024 Texas Quality Forum.