Related Experiment Video
Updated: Jul 15, 2025
Urinary Tract Infection in a Small Animal Model: Transurethral Catheterization of Male and Female Mice
Published on: December 1, 2017
Measuring and Reducing Racial Bias in a Pediatric Urinary Tract Infection Model
Abstract:
Clinical predictive models that include race as a predictor have the potential to exacerbate disparities in healthcare. Such models can be respecified to exclude race or optimized to reduce racial bias. We investigated the impact of such respecifications in a predictive model - UTICalc - which was designed to reduce catheterizations in young children with suspected urinary tract infections. To reduce racial bias, race was removed from the UTICalc logistic regression model and replaced with two new features. We compared the two versions of UTICalc using fairness and predictive performance metrics to understand the effects on racial bias. In addition, we derived three new models for UTICalc to specifically improve racial fairness. Our results show that, as predicted by previously described impossibility results, fairness cannot be simultaneously improved on all fairness metrics, and model respecification may improve racial fairness but decrease overall predictive performance.
Insights
Respecifying clinical predictive models can reduce racial bias but may decrease overall performance. Achieving fairness across all metrics simultaneously is not always possible, highlighting trade-offs in healthcare AI.
Area of Science:
- Health Informatics
- Clinical Decision Support
- Algorithmic Fairness
Background:
- Clinical predictive models using race can worsen healthcare disparities.
- Respecification of models offers a potential strategy to mitigate racial bias.
- The UTICalc model, designed to reduce urinary tract infections in children, is examined for bias.
Approach:
- The UTICalc model was respecified by removing race and introducing new features.
- Two versions of the UTICalc model were compared using fairness and predictive performance metrics.
- Three additional models were developed to specifically enhance racial fairness.
Key Points:
- Model respecification can improve racial fairness but may compromise overall predictive accuracy.
- Simultaneously improving all fairness metrics is challenging, as indicated by impossibility theorems.
- Trade-offs exist between enhancing fairness and maintaining predictive performance in clinical models.
Conclusions:
- Excluding race and adding features to UTICalc showed effects on racial bias and performance.
- The study confirms that achieving perfect fairness across all metrics is often unattainable.
- Careful consideration of trade-offs is crucial when modifying clinical models for fairness.
Related Concept Videos
01:26Urine Studies II: Urine Culture and Sensitivity Test
01:30Urinary Tract Infection III: Diagnostic Studies and Interprofessional Care