Skip to content

DAB Pipeline

A modular, reproducible pipeline for detecting algorithmic bias in opioid use disorder treatment data.

Open In Colab   License: MIT   NIH AIM-AHEAD

What is the DAB Pipeline?

The Detecting Algorithmic Bias (DAB) Pipeline evaluates whether machine learning models for predicting opioid use disorder (OUD) treatment outcomes perform consistently across demographic groups. It uses the CTN-0094 clinical trial dataset and supports multiple modeling strategies and outcome definitions.

We define algorithmic bias as a disparity in model performance between different demographic groups — for example, a model that predicts relapse accurately for White patients but not for Black or Hispanic patients.

Quick Start

python3 run_pipelineV2.py --data data.csv --outcome ctn0094_relapse_event -d ./results

Get started How it works

Key Features

  • Fairness-First Design

    Every evaluation includes the demographic composition of the training cohort, making bias auditing a first-class output.

  • Clinical Outcomes

    8 pre-defined OUD outcomes across binary, count, and time-to-event endpoint types, derived from the CTN-0094 trial.

  • Modular Architecture

    Swap out models or outcome definitions without touching the core pipeline logic. External datasets must conform to the CTN-0094 schema required by PSM.

  • Propensity Score Matching

    Constructs balanced cohorts via PSM across a spectrum of majority/minority demographic ratios.

Built On

This pipeline is described in Odom et al. (2026, Drug and Alcohol Review) and extends the work of Luo et al. (2024), replicating their OUD return-to-use prediction model and adding a fairness analysis layer to evaluate how model performance shifts across demographic compositions.

Read more

Team

Role Name Institution
Clinical Lead Prof. Laura Brandt City College of New York
Computational Lead Prof. Gabriel Odom Florida International University
Data Scientist Ganesh Jainarain City College of New York
Data Scientist Aaron Marker Stony Brook University
Funding NIH AIM-AHEAD 1OT2OD032581-02-267

Citation

If you use this pipeline in your research, please cite:

@software{jainarain2026dab,
  author    = {Jainarain, Ganesh and Marker, Aaron and Odom, Gabriel J. and Brandt, Laura},
  title     = {DAB Pipeline: Detecting Algorithmic Bias in OUD Treatment Data},
  year      = {2026},
  publisher = {GitHub},
  url       = {https://github.com/CTN-0094/Pipeline}
}