|
Activity Number:
|
90
- Novel Statistical Methods for COVID Pandemic and Other Current Health Policy Issues
|
|
Type:
|
Contributed
|
|
Date/Time:
|
Monday, August 9, 2021 : 10:00 AM to 11:50 AM
|
|
Sponsor:
|
Health Policy Statistics Section
|
|
Abstract #318180
|
|
|
Title:
|
Fair Regression for Multiple Undercompensated Groups
|
|
Author(s):
|
Samson Mataraso* and Anna Zink and Sherri Rose
|
|
Companies:
|
Stanford University and Harvard University and Stanford University
|
|
Keywords:
|
regression;
health policy;
fairness
|
|
Abstract:
|
Risk adjustment is used to distribute payments to insurers and mitigate losses from high cost enrollees. Underpredicting spending for certain groups can be detrimental to their care as insurers may change the services they cover to make these patients less likely to enroll in their health plan. We predicted health care spending using ordinary least squares and net compensation penalized regression, which prioritizes fairness for undercompensated groups by penalizing underpredicted spending for patients in these groups. We simulated a population with nine covariates for each patient who were then assigned to anywhere from zero to five undercompensated groups. Costs were calculated as a function of the covariates and group assignments. The predictive ratio for OLS (R2 = 0.78) is 1.26 for individuals who do not belong to any undercompensated group (Gc) and an average of 0.87 for the undercompensated groups. For the penalized regression (R2 = 0.75), the predictive ratio is 1.17 for Gc and an average of 0.98 for the undercompensated groups. We have shown that net compensation penalized regression can reduce undercompensation while making only minimal sacrifices in overall fit.
|
Authors who are presenting talks have a * after their name.
