Fair regression for health care spending

Area of Science:

• Health economics
• Computer science
• Statistics

Background:

• Health insurance risk adjustment formulas predict healthcare spending to ensure fair payments.
• Current formulas underpredict spending for certain groups, leading to undercompensation and potential enrollment discrimination.
• This impacts healthcare access for vulnerable populations.

Purpose of the Study:

• To develop improved risk adjustment methods addressing fairness for undercompensated groups.
• To integrate fairness considerations directly into regression models.
• To propose a novel fairness metric for evaluating risk adjustment formulas.

Main Methods:

• Developed new fair regression methods for continuous outcomes by incorporating fairness into the objective function.
• Utilized concepts from statistics, computer science, and health economics.
• Applied methods to IBM MarketScan Research Databases and conducted simulation studies.

Main Results:

• New fair regression methods demonstrated significant improvements in group fairness, achieving up to 98% enhancement.
• Overall model fit experienced only a minor reduction, approximately 4%.
• Proposed a novel fairness metric, emphasizing the need for a suite of metrics.

Conclusions:

• The developed fair regression methods offer substantial improvements in risk adjustment fairness for undercompensated groups.
• These methods can mitigate incentives for insurers to avoid enrolling specific populations.
• This research contributes to more equitable health insurance markets and improved healthcare access.