The 131st Annual Meeting (November 15-19, 2003) of APHA

The 131st Annual Meeting (November 15-19, 2003) of APHA

3063.0: Monday, November 17, 2003 - 9:12 AM

Abstract #60526

Using Pharmacy and Inpatient Diagnoses to Predict Medical Costs for Medicare Populations

Yang Zhao, PhD1, Randy P. Ellis, PhD2, Arlene S. Ash, PhD3, and John Haughton1. (1) Research Department, DxCG, Inc, 25 Kingston St., Suite 200, Boston, MA 02111, 317-651-3154,, (2) Department of Economics, Boston University, 270 Bay State Rd., Boston, MA 02460, (3) Dept. of Medicine, Section of General Internal Medicine, Boston University School of Medicine, 720 Harrison Ave., Suite 1108, Boston, MA 02118

Research Objective: Inpatient encounter data have been used to reimburse health plans participated in the Medicare + Choice program since the Center for Medicare and Medicaid Services (CMS) implemented a health-status based risk adjustment in 2000. Recently, CMS announced that it will expand its payment formula to use diagnoses from outpatient professional claims in 2004. The reliability and completeness of diagnoses from outpatient encounters are uncertain for some healthcare organizations. This paper explores risk adjustment models that use widely available outpatient pharmacy claims, alone or in combination with inpatient diagnoses to assess disease burden and financial risks in the Medicare population.

Study Design: This study examines 700,000 over-age-65 individuals with pharmacy benefit provided in employer sponsored Medicare supplemental insurance in 1998-2000. People present in all three years contribute two observations. Inpatient encounter data are used to create disease profiles that characterize each individual’s health status, while pharmacy claims are collapsed into pharmacy profiles. We use demographics, pharmacy profiles, and disease profiles generated from year-1 information to predict the next year’s total health care expenditures. R-squares and the predicted vs. actual year-2 costs for biased groups are used to evaluate the predictive accuracy. Finally, we explore the use of prior year pharmacy profiles to predict future drug costs only.

Results: About 85% of the individuals in our sample have at least one prescription in any given year, and one-quarter of the total medical costs are from drugs alone. The pharmacy-based model in our over-age-65 population explains 7.7% of future individual total cost variation. Adding inpatient diagnoses to pharmacy claims, we are able to better distinguish among the high-cost individuals. A model using both data sources performs significantly better than the Rx-only model, increasing the R-square value by 18% up to 9.1%. Pharmacy costs alone are very predictable, with over 45% of the individual variation explained by our prior year pharmacy profile.

Conclusions: Comprehensive classification systems that use pharmacy and diagnoses to identify medical problems and to assess disease burden are useful for population health management. Pharmacy claims, used alone, or in combination with inpatient diagnoses are able to discriminate people with their expected costs.

Learning Objectives:

Keywords: Healthcare Costs, Medicare

Presenting author's disclosure statement:
I have a significant financial interest/arrangement or affiliation with any organization/institution whose products or services are being discussed in this session.
Relationship: I am an employee at DxCG, Inc., which licenses software that implements the Diagnostic Cost Group and RxGroups risk adjustment models.

Health Care Management Issues (Health Services Research Contributed Papers #1)

The 131st Annual Meeting (November 15-19, 2003) of APHA