Case Studies

Rating Episode Grouper Output

Client: Medical Groups, 2006

Creative Statistical Inference Builds Consensus for Clinical Improvement Efforts

As episode grouping software gained acceptance amongst health plans and large purchasers of health care, clinicians often pointed with understandable skepticism to the software’s complexity and to the accuracy of the claim data analyzed. As a result, improvement initiatives were often delayed, and in some instances, abandoned amidst calls for transparent data processing software and comprehensive claim validation. Such requests, though commonsensical, were costly to fulfill or exposed proprietary methods to competing software developers. Clearly, an alternate approach to build consensus was required.

This project examined the statistical profiles of episode grouper output to identify specific clinical episodes of apparently questionable composition. Conversely, clinical conditions that were very well-described by cost distributions known to derive from stable underlying systems or processes were recognized to have greater validity and reliability. Examples of these two classes of clinical episodes are shown below:

Class 1 - Hypothyroidism:
Cost per episode is log-normally distributed with
very strong goodness-of-fit statistics.

Class 2 - Simple tonsillitis, adenoiditis, and pharyngitis:
Poor goodness-of-fit. Spike in low cost episodes suggests heterogeneity
or data issue.

By focusing improvement efforts on clinical conditions exhibiting the expected cost distributions, concerns surrounding software complexity and data accuracy were alleviated.