Fairness in Classifying and Grouping Health Equity Information

Area of Science:

• Machine Learning
• Health Informatics
• Medical Ethics

Background:

• Predicting antimicrobial treatment is crucial in wound care.
• Social determinants of health, like language and gender, can influence healthcare access and outcomes.
• Evaluating fairness in machine learning models is essential to prevent health disparities.

Purpose of the Study:

• To assess the balance between fairness and performance in machine learning classifiers for predicting antimicrobial treatment.
• To analyze the impact of grouping language codes on classifier fairness and performance.
• To investigate the role of social determinants of health (gender, language) in model fairness.

Main Methods:

• Utilized structured data from community nursing wound care electronic health records.
• Employed various common statistical learning-based classifiers.
• Evaluated classifier fairness using gender and language as predictors.
• Analyzed the effect of different language code groupings on model outcomes.

Main Results:

• K-Nearest Neighbors demonstrated superior fairness metrics across various language code groupings.
• Classifier performance remained consistent despite different language code groupings.
• Grouping more variables generally enhanced fairness metrics without compromising performance.

Conclusions:

• Machine learning models can be optimized for fairness in healthcare predictions.
• Language and gender data are important considerations for equitable AI in nursing.
• Variable grouping strategies can improve fairness in predictive models for antimicrobial treatment.