Selected WorkNAFLD-THESIS-EARLY-LIVER-INJURY

NAFLD Thesis: Early Prediction of Liver Injury in NAFLD Patients Prescribed Common Medications

Research into whether NAFLD patients on common drugs are at elevated risk of drug-induced liver injury, using real-world EHR data, survival analysis, and a rule-based early warning system.

Hybrid MLExpert SystemsClinical DataTime SeriesFeature EngineeringBigQuery
Cover
Stack

Tech Stack

PythonBigQuerypandas / NumPystatsmodelsscikit-learnFastAPIOMOP CDM
SectionOverview

Overview

A data-driven study exploring whether NAFLD patients prescribed certain drugs face a meaningfully higher risk of drug-induced liver injury, and whether that risk can be flagged early from routine clinical data.

SectionThe Problem

The Problem

NAFLD is common and largely silent. When patients with pre-existing liver vulnerability are prescribed hepatotoxic drugs, injury signals emerge gradually in routine blood tests, but no system connects the dots at prescribing time.

SectionThe Solution

The Solution

Built a longitudinal data pipeline on real-world EHR data, engineered clinically grounded lab features, and derived transparent rules for early risk flagging, paired with a scoring engine to assess causality when injury is suspected.

SectionResults

Results & Impact

NAFLD patients on certain drugs show significantly elevated liver injury rates. The rule engine flags high-risk prescriptions and generates monitoring recommendations grounded in the observed data.