Trey Brown — Clinical Data Scientist

Featured Work

Production ML
in Clinical Settings

★ Presented at Epic XGM 2026

Healthcare ML · XGBoost · SHAP · Epic Cosmos

HIV Risk Prediction Model

Built and deployed an XGBoost model trained on Epic Cosmos data to proactively identify patients at elevated HIV risk — enabling clinical outreach before symptoms appear. Jefferson Health was the first organization in the Epic ecosystem to translate a Cosmos research model through Slate development into a live Nebula production deployment with an operational OPA.

0.805

AUC (org-wide validation)

114K+

Patients scored

~1,789

High-risk patients flagged

Fairness-aware design Deliberately removed race and SVI features to prevent disproportionate flagging, accepting a 0.053 AUC trade-off for ethical deployment

SHAP in production TreeExplainer integrated into Epic PMA — clinicians see per-patient feature contributions directly in Hyperspace

Deployment pipeline Cosmos → Slate Docker container → Nebula production, solving version conflicts across scikit-learn, XGBoost, and pickle serialization

🌐

Cosmos

Research data

→

🐳

Slate

Docker dev

→

🚀

Nebula

Production

Why this matters

Epic Cosmos is a research environment — getting a model out of it and into a live clinical workflow had never been done before. This project proved the full pipeline is possible and documented the path for other health systems to follow.

NLP · LLM · Epic PMA

Valvular Disease NLP Pipeline

Production GPT-4o inference pipeline that classifies echocardiogram impressions into multi-class valvular severity labels (none / trace-mild / moderate / severe) when structured documentation is absent. Migrated from Epic PMA to Microsoft Fabric using Azure OpenAI and Spark SQL for scale beyond the 15-minute RWB timeout limit.

GPT-4oAzure OpenAIMicrosoft FabricSpark SQL

NLP · Transformers · Research

Brain Edema Detection In Research

Exploring lightweight transformer alternatives (<5GB, CPU-only) to replace high-cost LLM inference for brain edema classification from radiology impressions — targeting Epic Nebula deployment constraints. Current approach uses GPT-4o via Epic's AI Services framework; evaluating fine-tuned DistilBERT and model distillation strategies. Labeled 82K reports for training data.

DistilBERTHugging FacePyTorchGPT-4oEpic PMA

LLM · Clinical NLP · Epic Workflow API

Clinical Note Summarizer

GPT-4o-powered Epic Workflow API model that summarizes clinical notes on demand, surfacing output directly in the patient list via a StoreStructuredText print group in Hyperspace. Built using Epic's native cognitive inference framework (epic.io.workflow_integrations) extracted from Slate Docker.

GPT-4oEpic Workflow APIPythonParcel

Technical Range

Skills

ML & Modeling

XGBoostscikit-learnPyTorch DistilBERT / ClinicalBERTSHAP Cross-validationClass imbalanceFairlearn

NLP & LLMs

GPT-4o / Azure OpenAIHugging Face Transformers Fine-tuningPrompt engineering Clinical text classification

Epic Ecosystem

Epic PMA / NebulaEpic Cosmos Slate DockerWorkflow API Clarity / CaboodleCosmos Data Science

MLOps & Cloud

DockerAzure MLMicrosoft Fabric Spark SQLGitHub Actions

Data & Languages

PythonSQLpandas NumPyJavaScriptHTML / CSS

Credentials & Accreditations

CPHIMS M.S. Health Care Informatics — UCF Epic XGM Speaker 2026 Epic Cadence Epic Prelude Epic Referrals & Authorizations Epic Real Time Eligibility Epic Campaigns Epic Cogito Epic Cosmos — Data Science Epic Cosmos — Data Architecture (in progress)

Experience

Where I've Worked

Apr 2025 – Present

Jefferson Health Network

Clinical Data Scientist

Designing and deploying predictive models in Epic's production environment. Built the first Cosmos-to-Nebula ML deployment in the Epic ecosystem. Presented at Epic XGM 2026.

Jun 2023 – Apr 2025

Lehigh Valley Health Network

Clinical BI Developer, Cogito Team

Partnered with the Medical Director of Data Science to prototype clinical predictive models using JupyterLab. Extracted and preprocessed large-scale patient data from Epic Clarity using SQL to build ML-ready datasets. Built executive-facing dashboards tracking clinical KPIs across patient throughput, staffing, and virtual nursing programs.

Apr 2022 – Apr 2023

University Health

SER Administrator / Epic Analyst

Owned the organization's entire SER master file — the source of truth for provider identity across Epic. Sole author and maintainer of a 3,000-line SQL job that ran daily to automatically provision all internal providers, controlling how provider data flowed in and out of the system. Configured RTE for 4,000+ monthly eligibility checks and resolved X12/EDI claim rejection issues across multiple payers.

Get in Touch