Although some of the limitations of specific performance measures are inherent, many measures could be substantially improved with better data sources and analytic methods. More broadly, performance measurement could be improved by the development of measures for often overlooked domains of quality. Although others have outlined a more comprehensive research agenda for improving performance measurement (Leatherman et al., 2003; McGlynn et al., 2003b), we describe below a few obvious areas in which progress is needed.

Getting to Better Data

The usefulness of many performance measures is limited by the quality or availability of appropriate data sources. Billing or other administrative data are ubiquitous, relatively inexpensive to use, and adequately robust for many performance measures. However, they often lack sufficient clinical specificity to define relevant patient subgroups (i.e., the denominator of patients appropriate for a given process measure), to conduct adequate risk adjustment, and to detect and discourage physicians from gaming measures. Although data obtained from medical records can often meet these needs, clinical data for performance measurement are very expensive and not widely available.

Future research should address how to meet the minimum data quality needs for various performance measures in the most cost-efficient manner possible. As a start, researchers might identify those measures for which current administrative data sets are sufficient. As described earlier, risk adjustment may not be as important as commonly assumed, particularly for outcome measures applied to relatively homogenous populations. To identify such instances, researchers could use existing clinical registries to highlight procedures or conditions for which adjusted and unadjusted mortality rates are sufficiently correlated.

Where better data are needed, future research could also explore the merits of two alternative approaches. The first would be to improve the accuracy and detail of administrative data by adding a small number of “clinical” variables to the billing record. These could include either specific process of care variables, laboratory values, or information most essential for risk adjustment purposes. With the latter, for example, Hannan and colleagues noted that risk adjustment models derived from administrative data for CABG would approximate the reliability of those from clinical data with the addition of only three variables (ejection fraction, reoperation, and left main stenosis). Similarly, information about laboratory values and

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