[Possibilities and limitations of statistical regression models for the calculation of threshold values for minimum provider volumes].

Inadequate statistical procedures are often applied for the derivation of threshold values in various medical research areas. The frequently applied method to establish threshold values on the basis of simple comparisons between arbitrarily defined low-volume and high-volume groups may be misleading because the result depends on the preceding classification. In this paper, the possibilities and limitations of statistical regression models for the calculation of threshold values are described. The features of these models for the selection of minimum volumes for hospitals or physicians are discussed. Simulated data examples are used to demonstrate that the definition of a useful minimum provider volume should not be based upon a calculated value of purely mathematical meaning without clinically assessing the risk curve. In the application of statistical regression models to retrospective observational data it should be noticed that calculated threshold values are only of a hypothesis-generating character. In order to verify that a minimum provider volume leads to the expected quality improvement, a prospective intervention study is required.