Overview
Pulse oximeters measure blood oxygen levels, but studies have shown they overestimate oxygen saturation in patients with darker skin tones — a flaw that went largely unnoticed for decades and contributed to delayed care during COVID-19. This project built a tool to detect and quantify that bias in clinical datasets. It placed at Macathon 2025.
Data
We used publicly available clinical datasets containing paired pulse oximeter readings and arterial blood gas measurements, along with patient race/ethnicity data.
Methods
- Computed hidden hypoxemia rates (true SpO2 < 88% masked by oximeter reading ≥ 92%) stratified by race
- Visualized disparity in readings across groups
- Flagged patient records where the gap between oximeter and true oxygen level exceeded a clinical threshold
Results
We replicated findings from peer-reviewed literature showing Black patients were significantly more likely to have hidden hypoxemia and built a reusable pipeline that can be applied to new clinical datasets.
