Group 130: Auditing Algorithmic Bias in Critical Healthcare AI

An official NSRI research team auditing algorithmic bias, geographic disparities, and fairness gaps in multi-modal clinical machine learning models trained on large-scale electronic health records.

Target Dataset Focus: MIMIC-IV and eICU Collaborative Research Databases

Clinical machine learning models are rapidly shifting from retrospective research to active bedside software.

However, if these models are trained on data reflecting systemic inequalities, they risk automating and magnifying biases against historically underrepresented demographic and geographic groups.

This research group will conduct a systematic meta-analysis auditing multi-modal clinical AI models (predictive health informatics, critical care survival curves, and diagnostic imaging models). Our primary objective is to map "fairness gaps" - tracking how validation metrics like AUROC, sensitivity, and false-positive rates fluctuate when applied across diverse racial, socioeconomic, and global cohorts.

Students in this group will gain direct exposure to medical data science challenges and solutions, learn how to evaluate algorithmic equity, and contribute to a peer-reviewed manuscript aimed at open-access health informatics journals.

We are looking for dedicated students split into two primary tracks:

- Data & ML Track: Comfortable with Python, data preprocessing, and understanding statistical machine learning metrics. - Public Health Track: Strong passion for public health and experience in literature synthesis and scientific manuscript drafting.