USING ADMINISTRATIVE DATABASES TO EVALUATE THE QUALITY OF MEDICAL-CARE - A CONCEPTUAL-FRAMEWORK

Health care is consuming an ever larger share of national resources in the United States. Measures to contain costs and evidence of unexplai ned variation in patient outcomes have led to concern about inadequacy in the quality of health care. As a measure of quality, the evaluatio n of hospitals through analysis of their discharge databases has been suggested because of the scope and economy offered by this methodology . However, the value of the information obtained through these analyse s has been questioned because of the inadequacy of the clinical data c ontained in administrative databases and the resultant inability to co ntrol accurately for patient variation. We suggest, however, that the major shortcoming of prior attempts to use large databases to perform across-facility evaluation has resulted from the lack of a conceptual framework to guide the analysis. We propose a framework which identifi es component areas and clarifies the underlying assumptions of the ana lytic process. For each area, criteria are identified which will maxim ize the validity of the results. Hospitals identified as having unexpe ctedly high unfavorable outcomes when our framework is applied will be those where poor quality will most likely be found by primary review of the process of care. We outline the following criteria for the sele ction of disease-outcome pairs to be studied in large administrative d atabase analysis: (1) disease entities or clinical states selected sho uld be well defined and easily diagnosed; (2) if diagnostic classifica tion systems are used, disease groups should be homogeneous as to the clinical states they contain; (3) the disease entities should be preva lent; (4) a plausible link should exist between the quality (process) of care and the frequency of the outcome; (5) types of care which conf orm to acceptable practice standards but still lead to variation in th e outcome of interest should be excluded from the analysis; (6) the ou tcomes should be prevalent; (7) constraints of the ICD-9 coding system should be understood such that only those disease-outcome pairs least affected by these limitations are selected for analysis; (8) constrai nts of the structure of the database should be considered when the ana lysis is performed; and (9) disease-outcome pairs should be chosen whe re there is agreement on the processes of care that affect the outcome of interest, either favorably or unfavorably.