Abstract:
|
This paper focuses on a two-stage empirical Bayesian estimator of healthcare quality applied to count data with a high proportion of zeroes. A two-stage estimator to risk- and reliability-adjust measures are gaining popularity in federal and state policy. The first stage model accounts for differential risk among the patients, creating a risk-adjusted rate which may be subject to large standard errors due to small sample sizes in certain groups; therefore, at the second-stage, risk-adjusted rates are pulled toward a prior under a Bayes framework to create a reliability-adjusted rate. We focus on the case where the quality measure is the count of potentially preventable hospitalizations, where many Medicaid beneficiaries have zero counts. We compare two likelihood models: normal and negative binomial to profile state performance on the quality of home- and community-based services (HCBS) received by Medicaid beneficiaries. The model validation and diagnostic show that the three models have similar performance when all states have large sample sizes, but the negative binomial model outperforms the normal model when a subpopulation is considered.
|
ASA Meetings Department
732 North Washington Street, Alexandria, VA 22314
(703) 684-1221 • meetings@amstat.org
Copyright © American Statistical Association.