[Skip to Content]
Access to paid content on this site is currently suspended due to excessive activity being detected from your IP address 54.166.74.94. Please contact the publisher to request reinstatement.
[Skip to Content Landing]
Views 107
Citations 0
Letters
June 27, 2012

Limitations of Administrative Databases—Reply

Author Affiliations

Author Affiliations: Center for Comprehensive Access and Delivery Research and Evaluation, Iowa City VA Medical Center, Iowa City, Iowa (mary-vaughan-sarrazin@uiowa.edu).

JAMA. 2012;307(24):2589-2590. doi:10.1001/jama.2012.6636

In Reply: Dr Haut and colleagues provide another example of systematic coding bias in administrative data (undercoding of diagnostic procedures that do not affect reimbursement) that, if not recognized, may lead to invalid inferences regarding impact of procedure use on patient outcomes. We agree with Haut et al that, despite this and other limitations, the use of administrative data should not be abandoned.

Administrative data are a source of information regarding real-world clinical practices across geographic regions, hospitals, and important patient subgroups (eg, race, payer) and have variations that are difficult to assess using data from randomized clinical trials, surveys, or other primary data sources. Thus, creative approaches to enhancing administrative data are needed. As Haut et al note, the lack of complete clinical information in administrative data can be addressed, in part by supplementing administrative data with information available through electronic health records or existing registries, such as the SEER cancer registry. Often the administrative data can be significantly improved with only 1 or 2 additional variables. For example, Hannan et al1 showed that the inclusion of 3 additional variables obtained from a clinical registry (ejection fraction, reoperation, and >90% narrowing of the left main coronary artery) significantly improved the discrimination of coronary bypass surgery risk-adjustment models based on administrative data. As another example, Pine et al2 showed that the predictive validity of risk adjustment models for 6 common diagnoses derived from administrative data and electronically available laboratory data were similar to models derived from data painstakingly abstracted from patients' medical records.

First Page Preview View Large
First page PDF preview
First page PDF preview
×