Project
NeuroRx AI: Predicting Antidepressant Response
Overview
Developed a machine learning model to predict patient responses to antidepressants, utilizing clinical trial data to guide personalized treatment plans.
Problem
Approximately 30% of individuals with depression do not respond to initial antidepressant treatments, leading to prolonged suffering and increased healthcare costs. Personalized treatment strategies can enhance efficacy and reduce trial-and-error prescribing.
Solution
Achieved an AUC score of 0.83, identifying key predictors such as prior treatment resistance and baseline depression severity.
Resources
📊 Data Sources:
STAR*D Dataset: A large, publicly available clinical trial for depression treatments. Includes demographics, treatment regimens, and QIDS/HAM-D scores.
MIMIC-IV: ICU EHR data—useful for augmentation with co-morbidities or lab results.
eICU Collaborative Research Database: Additional ICU patient data.
dbGaP - Genetic Data: If you want to explore pharmacogenomics, dbGaP includes genotype/phenotype data.
🔐 Tip: Apply for access to MIMIC and dbGaP (takes 1–2 weeks with CITI training).
Category:
Domain: Psychopharmacology / Mental Health AI
Industry: Digital Health, Precision Psychiatry, PharmaTech
Technical Specialization: Predictive Modeling, Clinical Decision Support, Explainable AI
