ABSTRACT
I. BACKGROUND Germ cell tumors are the most commonplace of the cancers afflicting young men between the ages of 15 and 35 (Motzer and Bosl, 1987). The first goal of treatment is achieving complete response, that is, complete disappearance of the tumor. For disseminated germ cell tumors, multiple-drug chemotherapy, including cisplatin, has been used to achieve a complete response in 70 to 80% of patients, yet approximately 10 to 15 % of patients relapse, usually within the first 2 years (Bosl and Geller, 1989). While long-term follow-up of these patients is of interest, there is a strong relationship between cure of this disease and complete response, so that complete response is often taken as the major endpoint in clinical trials. The acute and chronic toxicities of cisplatinbased treatment for germ cell tumors are well recognized and include decreased white blood cell and platelet counts, decreased serum magnesium, decreased renal function, nausea and vomiting, mouth sores, and occasional hearing loss (Bosl et al., 1988). Treatment with less toxic drugs is desirable, but not at the risk of decreasing the number of cures. For this reason, goodrisk patients, that is, those likely to achieve complete response, have been identified in prognostic factor studies (e.g., Bosl et aI., 1983). In these
This trial was designed according to the proposals of DeMets and Ware (1982) and Lan and DeMets (1983). As the data accumulated, at most five analyses were planned. Assuming that the five analyses would take place after successive groups of 24 patients per arm were evaluated, the sequence of nominal significance levels to be used were 0.0033, 0.0274, 0.0584, 0.0872, 0.1118. The first hypothesis test (based on 24 patients per arm) would be undertaken at the 0.0033 level, and if the difference in complete response proportions in the two arms was not extreme enough to stop the trial, a second group of 24 patients per arm would be accrued and a second analysis undertaken (based on 48 patients per arm) at the 0.0274 level; and so on. Accrual of the required number of patients was anticipated within 3 years. If the complete response proportion in the experimental treatment arm turned out to be 0.8 and the standard treatment arm had response proportion 0.9 as anticipated, the power of this design to detect this difference was 0.75.
