Disparity implications of machine-learning-based MTM eligibility criteria

Area of Science:

• Health Services Research
• Health Equity
• Health Informatics

Background:

• Medicare Medication Therapy Management (MTM) programs offer clinical and economic advantages.
• Racial/ethnic minority groups encounter barriers to MTM program enrollment.
• The Enhanced MTM demonstration allows flexible beneficiary identification, but concerns about algorithmic bias persist.

Purpose of the Study:

• To determine if health cost-based MTM eligibility varies by race/ethnicity.
• To evaluate if machine learning models replicate existing disparities in predicted MTM eligibility.

Main Methods:

• Analysis of 2019 Medicare administrative data for a 10% random sample of fee-for-service beneficiaries.
• Assessment of top-quartile medication and healthcare costs from Medicare and healthcare system perspectives.
• Application of six machine learning algorithms to predict eligibility and assess racial/ethnic disparities.

Main Results:

• Black and Hispanic beneficiaries had significantly lower adjusted odds of top-quartile costs compared to non-Hispanic White beneficiaries.
• Machine learning models accurately reflected these disparities in predicted MTM eligibility.
• Observed disparities in medication and healthcare costs were consistently reproduced by predictive models.

Conclusions:

• Cost-based MTM eligibility determined by predictive algorithms may perpetuate racial/ethnic disparities in program access.
• Further research is needed to develop strategies for mitigating bias in MTM eligibility determination.
• Ensuring equitable access to MTM services requires addressing algorithmic biases.