The current implementation of external quality assurance focuses on identical, albeit simple univariate procedures. This approach proves to be inefficient, exhibiting low specificity. Definitions of performance indicators were classified by the local steering committee and applied to Bavarian obstetric data augmented with multivariate and confidence interval analyses as well as statistical simulations. A highly selected group of units with multivariate extreme values was audited. Blind application of federal regimes without further in-depth analysis of concomitant data is inefficient and may lead to wrong conclusions. Exclusive univariate analyses of performance indicators are not enough. Multivariate analyses have a higher specificity. Classification reveals an imbalance in the indicator set with respect to varying degrees of coverage as well as a bias towards outcome. Deficiencies in the organisational processes crucial for quality of care are often only revealed by dedicated cause and effect analyses within the hospital, thus constituting an essential prerequisite for desired changes in health care policies. It is unlikely that mere publication of performance indicators in the impending quality report will generate lucidity in spite of abundant structure. The installation of standardised programmes with identical computation rules may appear convincing as long as no doubt is cast upon the efficacy of the performance indicators in respect to their intended usage as well as upon the expected feedback effects. The experience so far however teaches us that these elements may be essential yet although not sufficient.     «

The current implementation of external quality assurance focuses on identical, albeit simple univariate procedures. This approach proves to be inefficient, exhibiting low specificity. Definitions of performance indicators were classified by the local steering committee and applied to Bavarian obstetric data augmented with multivariate and confidence interval analyses as well as statistical simulations. A highly selected group of units with multivariate extreme values was audited. Blind applicati...     »