An Analysis of Unobserved Selection in an Inpatient Diagnostic Cost Group Model
- Resource Type
- Authors
- Emmett B. Keeler; Dana P. Goldman; Nasreen Dhanani; Glenn Melnick; Hongjun Kan
- Source
- Health Services & Outcomes Research Methodology. 4:71-91
- Subject
- Selection bias
Actuarial science
Health economics
business.industry
Health Policy
media_common.quotation_subject
Public Health, Environmental and Occupational Health
Sample (statistics)
Group model
Risk adjustment
Health administration
Simultaneous equations model
Medicine
business
health care economics and organizations
Selection (genetic algorithm)
media_common
- Language
- ISSN
- 1387-3741
The study assesses unobserved selection bias in an inpatient diagnostic cost group (DCG) model similar to Medicare's Principal Inpatient Diagnostic Cost Group (PIP-DCG) risk adjustment model using a unique data set that contains hospital discharge records for both FFS and HMO Medicare beneficiaries in California from 1994 to 1996. We use a simultaneous equations model that jointly estimates HMO enrollment and subsequent hospital use to test the existence of unobserved selection and estimate the true HMO effect. It is found that the inpatient DCG model does not adequately adjust for biased selection into Medicare HMOs. New HMO enrollees are healthier than FFS beneficiaries even after adjustment for the included PIP-DCG risk factors. A model developed over an FFS sample ignoring unobserved selection overestimates hospital use of new HMO enrollees by 28 percent compared to their use if they had remained in FFS. Models that better captures selection bias are needed to reduce overestimation of Medicare HMO enrollees' resource use.