feasible for use during practice hours. Their relationship to improved patient outcomes remains unevaluated.
Large data sets refer to claims-based administrative data bases such as those for Medicare Part A and Part B claims. Roos et al. (1989) distinguish three types of data bases and the kinds of studies that are feasible with each. A Level 1 data base contains only hospital discharge abstracts and will permit aggregate studies of, for instance, in-hospital mortality rates and lengths of stay, by geographic region or over time. A Level 2 data base contains, in addition, unique patient-identifying numbers. It can be used to study, for instance, short-term readmissions and volume and outcome relationships at a hospital-specific level. A Level 3 data base (the most comprehensive) will also have information from health program enrollment files, including when eligibility begins and ends. This data base permits the highest quality longitudinal studies, short- and long-term outcomes studies, and population-based (system-wide coverage) studies. Studies can include outcomes for intervention-free individuals and for poor outcomes or other complications that are not recorded as part of the hospital stay.
Weiner et al. (1989) have provided examples of quality-of-care indicators that might be developed from ambulatory care data bases. These include system measures such as the rate of hospitalizations, of readmissions, and “avoidable disease” or disease first diagnosed at an advanced stage. Other examples include (1) preventive-care indicators, such as the percentage of eligible persons receiving a recommended number of periodic screening tests or exams within a given time period and the documented incidence of newly diagnosed disease versus the expected incidence; (2) diagnostic indicators, such as the number or proportion of patients who receive unnecessary diagnostic tests or procedures; and (3) treatment indicators, such as the percentage of patients with a given diagnosis who receive the appropriate medication, the percentage of patients undergoing ambulatory surgical procedures who experience complications including hospitalizations, and the percentage of all visits to the patient’s primary provider.
All large administrative data bases have several theoretical advantages for quality assessment. First, the accuracy of various types of data (e.g., medications, previous hospitalizations, and numbers of physician visits and medical conditions treated) is unaffected by errors in patient or practitioner