Three different types of heart-lung machine systems should be compared in a clinical trial with regard to their impact on blood coagulation and immune system parameters. These were a standard version (A) and two modifications (B) and (C). It should be shown that B and C are 'superior' to A and that the more economical version C is 'not inferior' to the expensive version B. The practical circumstances allowed for a randomized three-armed double blind trial. However, the prior knowledge was not sufficient to plan a study with fixed sample size or an usual sequential design, such that we startet a two-phase adaptive trial. There are four different sources of statistical multiplicity in this trial.

- The first one is the simultaneous consideration of blood coagulation and immune system. Here, we decided to treat both questions separately without special adjustment.

- Each of these two physiological categories is described by several variables (multiple endpoints). This problem is taken up by the application of so-called stable multivariate tests (Läuter, Glimm and Kropf, 1996).

- We have multiple comparisons between the three groups. More precisely, there are two tests of superiority of one treatment with respect to another one and one test of non-inferiority of a treatment. The three comparisons are carried out as tests with a priori ordering of hypotheses. The test of non-inferiority is transformed into a test of a contrast of all three treatments. This latter problem is extended to a series of modified hypotheses similar as in Bauer, Röhmel, Maurer, and Hothorn (1998).

- For the two-phase adaptive design, the methodology of Bauer and Köhne (1994) is used. The paper describes, how these basic techniques are combined. We utilize proposals by Kropf, Hothorn and Läuter (1997) to carry out multiple comparisons with multiple endpoints and modify an approach of Bauer and Kieser (1999) to treat multiple hypotheses in a two-phase design. The adaptation of the multivariate tests after phase I is used to reduce the laboratory costs in phase II (if necessary). Critical assumptions of the statistical methods and the complications resulting from the combination of techniques are discussed. The results of the trial are considered as given after phase I together with the conclusions for phase II. This allows for a trade-off between the expected gain and the related costs for the investigation of the remaining unanswered partial questions.

References:

1. Bauer, P.; Kieser, M. (1999). Combining Different Phases in the Development of Medical Treatments within a Single Trial. Statistics in Medicine 18, 1833-1848.
2. Bauer, P.; Köhne, K. (1994). Evaluation of Experiments with Adaptive Interim Analysis. Biometrics 50, 1029-1041.
3. Bauer, P.; Röhmel, J.; Maurer, W.; Hothorn, L. A. (1998). Testing Strategies in Multi-Dose Experiments Including Active Control. Statistics in Medicine 17, 2133-2146.
4. Kropf, S.; Hothorn, L. A.; Läuter, J. (1997). Multivariate Many-to-One Procedures with Application in Pre-Clinical Trials. Drug Information Journal 31, 433-447.
5. Läuter, J.; Glimm, E.; Kropf, S. (1996). New Multivariate Tests for Data with an Inherent Structure. Biometrical Journal 38, 5-22. Erratum: Biometrical Journal 40, 1015.