recall. Second, the use of these data bases is unobtrusive; patient consent for individual studies is not required, and no bias is introduced from individuals’ knowing that they are being studied. Unobtrusiveness may also contribute to their acceptability to practitioners and health care facilities. Third, assessors can create and test different statistical models or approaches to risk adjustment. They can also alter study designs or use several different study designs to test findings; for instance, they can use both cohort and case-control designs to examine the effect of different intervention periods.

Fourth, the same files can be applied in different ways, for instance, tracking outcomes of surgery, computer modeling of readmission, examining changes in complication rates over time, or studying outcomes of care for patients in different geographic areas. Fifth, investigators can accurately assess risks as well as benefits associated with treatment, especially for areas of medical uncertainty. The data bases can provide inputs for clinical decision making by allowing calculation of the probability of complications of treatment or of mortality at varying lengths of time after treatment. Sixth, the use of administrative data bases is relatively inexpensive in comparison to methods that require large-scale primary data collection.

An important strength of Level 3 data bases is that they contain population data, and thus they permit some assessment of population access and outcomes. Comparative studies should be able to identify possible areas of underuse.


Administrative data bases have considerable drawbacks for quality assessment. First, data bases may exclude important information such as certain events, information on location of service and provider, or costs, and may assemble the elements in ways that complicate linkage to other files.

Second, the precision of the coding schemes (primarily the ICD-9-CM4 and CPT systems) is of great concern, particularly for medical conditions that encompass a broad range of clinical severity and contain important clinical subgroups, such as congestive heart failure and diabetes mellitus. The ICD-9-CM coding system does not distinguish procedures performed on the right side of the body from those performed on the left. For this reason, a data base with ICD-9-CM codes will not allow a reviewer to determine whether a second hip replacement, for instance, is a reoperation or a new operation. Of equal concern is the poor ability of data bases to distinguish the order of events during a single episode of care (e.g., a pulmonary embolus that was present at the time of admission versus one that developed after surgery). Although administrative data bases record the occurrence of events such as x-rays and diagnostic tests, the results of these

The National Academies | 500 Fifth St. N.W. | Washington, D.C. 20001
Copyright © National Academy of Sciences. All rights reserved.
Terms of Use and Privacy Statement