Algorithmic Fairness of Machine Learning Models for Alzheimer Disease Progression

Area of Science:

• Neurology
• Artificial Intelligence
• Biomedical Data Science

Background:

• Machine learning (ML) models offer promise for early Alzheimer disease (AD) detection and management.
• However, potential biases in these models could worsen existing health disparities.

Purpose of the Study:

• To evaluate the algorithmic fairness of ML models predicting longitudinal AD progression.
• Assessing fairness across demographic groups is essential for equitable healthcare applications.

Main Methods:

• Prognostic study using data from the Alzheimer Disease Neuroimaging Initiative (ADNI).
• Logistic regression (LR), support vector machines (SVM), and recurrent neural networks (RNN) were used to predict progression to mild cognitive impairment (MCI) and AD.
• Fairness was assessed across sex, ethnicity, and race using metrics like equal opportunity and equalized odds.

Main Results:

• Models demonstrated lower prediction sensitivity for Hispanic and Black participants compared to non-Hispanic White participants.
• Significant disparities in true positive rates were observed across racial and ethnic groups for MCI and AD progression.
• Fairness metrics were generally met for sex, with some exceptions in specific transition stages.

Conclusions:

• While ML models show aggregate accuracy, they often fail to meet fairness criteria across demographic groups.
• The findings underscore the critical need to incorporate fairness considerations into the development and deployment of ML models for AD progression prediction.
• Ensuring algorithmic fairness is paramount to prevent the exacerbation of health disparities in AD care.