Identification of physicians with unusual performance in screening colonoscopy databases: a Bayesian approach.

BACKGROUND

The adenoma detection rate (ADR) is an important surrogate measure of performance quality for screening colonoscopies.

OBJECTIVE

To demonstrate how physicians with unusual performance concerning the adenoma detection rate may be identified in screening colonoscopy databases.

DESIGN

Bayesian random-effects modeling and Winsorization of potential outliers were applied to develop a robust model for the majority of providers. Divergence was assessed with adjustment for multiple testing. The steps in the analysis were visualized by using funnel plots. Additionally, minimum requirements for the number of colonoscopies with 1 or more detected adenomas were derived.

SETTING

Data from 422 physicians offering screening colonoscopy and participating in a quality assurance program in Bavaria, Germany, were used.

PATIENTS

A total of 69,738 asymptomatic individuals 55 to 79 years of age.

INTERVENTION

Screening colonoscopy.

MAIN OUTCOME MEASUREMENTS

Physician-specific ADRs.

RESULTS

The overall ADR in the sample was 26%. From an initial model, 62 physicians (15%) were identified as potential outliers. A model with normally distributed random effects was then chosen as the robust null model. Of the potential outliers, 10 (16%) were confirmed as physicians with unusual performance at a false discovery rate of 5%. For all of them, the observed ADR was lower than expected, and together they accounted for 1.4% of all included colonoscopies.

LIMITATIONS

Analysis of routine data.

CONCLUSION

The applied statistical approach appears suitable to identify unusual performance in screening colonoscopy databases. Its application may help to evaluate and improve the quality of colonoscopy in population-based colorectal cancer screening programs.