Die Statistische Auswertung Prärandomisierter Versuchspläne

Summary

A statistical model will be introduced for the evaluation of the pre-randomized designs (PRD) proposed by M. Zelen. Treatment and selection effects will be defined by this model. These effects can be estimated properly in the simple PRD. The approach suggested by Zelen yields correct estimates for these effects only when the sample characteristics which describe the outcome of the trial fulfill certain requirements. Examples and counterexamples are given.

In the double PRD, not all effects are estimable. It can be shown that the effect which is estimated by the procedure suggested by Zelen cannot be meaningfully interpreted as a therapy effect. This is shown by an example.

In the double Zelen design therapy effects are only estimable meaningful under additional restrictive conditions. These assumptions cannot be verified by statistical methods. If these assumptions are only slightly violated it is possible to perform a conservative test for a lower bound of the therapy effect.