Monitoring fairness in machine learning models that predict patient mortality in the ICU

TL;DR

This paper introduces a fairness monitoring method for ICU mortality prediction models, emphasizing the importance of evaluating performance across diverse patient groups to ensure equitable healthcare outcomes.

Contribution

It presents a novel approach to monitor fairness in clinical machine learning models, highlighting the limitations of traditional accuracy metrics.

Findings

Fairness analysis reveals disparities across race, sex, and diagnoses.

Documentation bias affects clinical measurement accuracy.

Fairness monitoring offers deeper insights than standard metrics.

Abstract

This work proposes a fairness monitoring approach for machine learning models that predict patient mortality in the ICU. We investigate how well models perform for patient groups with different race, sex and medical diagnoses. We investigate Documentation bias in clinical measurement, showing how fairness analysis provides a more detailed and insightful comparison of model performance than traditional accuracy metrics alone.