Eisenhandler J, Hughes J, Averill RF, Goldfield NI, Gay JC, Bao M; AcademyHealth. Meeting (2003 : Nashville, Tenn.).

Abstr AcademyHealth Meet. 2003; 20: abstract no. 453.

3M Health Information Systems, Clinical Research, 100 Barnes Road, Wallingford, CT 06492 Tel. (203)949-6662 Fax

RESEARCH OBJECTIVE: To examine the extent to which the addition of computerized pharmacy data to an existing global risk-adjustment system would improve its ability to predict total health care costs, and to examine whether pharmacy data could substitute for outpatient data. STUDY DESIGN: We used computerized pharmacy data to modify Clinical Risk Groups (CRGs), a proprietary patient classification system that forecasts subsequent year total spending based on computerized diagnosis and procedure codes from both inpatient and outpatient encounters. We allowed selected medications recorded in a pharmacy benefits management database to trigger the addition of specific diagnosis codes. For example, oral hypoglycemics or insulin would trigger a diagnosis of diabetes, and sustained use of inhaled bronchodilators would indicate asthma. We examined the change in the number of individuals categorized as having a chronic disease or a recent significant acute illness when pharmacy data was added. We also examined predictive performance for subsequent year costs by calculating R² values, both with and without pharmacy data added to various combinations of inpatient and outpatient data. For the latter analysis, we capped an individual's spending at $50,000, which reduces the impact of very expensive cases on R². POPULATION STUDIED: We used computerized data for a commercial insurance population of 309,196, one-third of which was less than 20 years old, and 85% less than 50. PRINCIPAL FINDINGS: The proportion of enrollees classified by the risk adjustment system as 'healthy' (no recorded chronic disease or recent significant acute illness) varied by the sources of data as follows: inpatient only, 98%; inpatient plus outpatient, 80%; inpatient plus pharmacy, 66%; and pharmacy plus inpatient and outpatient, 59%. We also found that those enrollees whose risk classification increased after the addition of pharmacy data tended to be more expensive than average for their original risk group, but less expensive than the average for the group they moved into.The effect of pharmacy data on predictive performance as measured by R² was also substantial but less impressive. With a spending cap set at $50,000 (accounting for 13.5% of all spending), we obtained the following R² values: for inpatient data only, 10.8; inpatient and pharmacy data, 15.2; inpatient and outpatient, 18.1; outpatient and pharmacy, 18.2; and inpatient, outpatient, and pharmacy, 19.0. CONCLUSIONS: The addition of computerized pharmacy data to a patient classification system increases the identification of individuals who are being treated for chronic illness. The effect is much greater when pharmacy data are added to inpatient data only, but still important when added to the combination of inpatient and outpatient data. For predicting total spending, the addition of pharmacy data adds little to a combination of inpatient and outpatient data, but substantially improves predictive performance when added to inpatient data only. Pharmacy data may provide a useful supplement for predicting costs when outpatient data are unavailable or inadequate IMPLICATIONS FOR POLICY, DELIVERY OR PRACTICE: Computerized pharmacy data may be more readily available than outpatient encounter data in some situations, and can help identify individuals with chronic illness and improve predictions of overall health costs.

Publication Types:

• Meeting Abstracts

Keywords:

• Chronic Disease
• Costs and Cost Analysis
• Health Care Costs
• Health Expenditures
• Humans
• Inpatients
• Outpatients
• Pharmacies
• Risk Adjustment
• classification
• economics
• hsrmtgs

Other ID:

• GWHSR0003750

UI: 102275429

From Meeting Abstracts