Clinical Benefit Scales and Trial Design: Some Statistical Issues

Abstract

Recently developed clinical-benefit outcome scales by the European Society for Medical Oncology and the American Society of Clinical Oncology allow standardized objective evaluation of outcomes of randomized clinical trials. However, incorporation of clinical-benefit outcome scales into trial designs highlights a number of statistical issues: the relationship between minimal clinical benefit and the target treatment-effect alternative used in the trial design, designing trials to assess long-term benefit, potential problems with using a trial endpoint that is not overall survival, and how to incorporate subgroup analyses into the trial design. Using the European Society for Medical Oncology Magnitude of Clinical Benefit Scale as a basis for discussion, we review what these issues are and how they can guide the choice of trial-design target effects, appropriate endpoints, and prespecified subgroup analyses to increase the chances that the resulting trial outcomes can be appropriately evaluated for clinical benefit.

Figure 1.

Simulated distribution of differences in observed median survival for a statistically significant 600-patient trial when the true medians are 9→13.5 months (A), 9→12 months (B), 9→11 months (C), 9→10.5 months (D), or 9→10 months (E). Shaded areas (darker to lighter) represent grade 4 to grade 1 clinical benefit (Table 1). The probabilities of a statistically significant trial are >99% (A), 90% (B), 62% (C), 41% (D), and 22% (E). (100 000 trials simulated using exponential distributions with 600 patients enrolled over 36 months and the analysis performed when 513 events are observed.).