Formal Statistical Testing and Inference in Randomized Phase II Trials in Medical Oncology

Abstract

Objectives: With the growing number of new anticancer therapies, randomized phase II trials have been used more often in oncology. Although the primary objective of such trials is not to formally compare results between arms, this practice seems frequent. We sought to quantify the frequency of use of formal statistical testing or inference through the use of P values and confidence intervals (CIs) in randomized phase II trials. Methods: We searched PubMed for randomized phase II trials assessing systemic cancer therapies published in the years 1995/1996 and 2005/2006. For each study, 2 reviewers independently abstracted data, including reporting of P values and CIs for the primary endpoint. Results: We retrieved 288 articles, 107 of which were eligible for analysis. The median number of patients per trial was 94, the primary endpoint was response rate in 71 (66.4%) cases, and a control arm was present in 55 (51.4%) trials. Either P values or CIs for the primary endpoint were reported in 85 (79.4%; 95% CI, 70.8%-86.1%) cases. Year of publication, source of funding, and use of a control group were not associated with this practice. Conclusions: Formal statistical comparisons between arms of randomized phase II trials are frequently undertaken in medical oncology. The extent to which such a practice abrogates phase III testing is unknown.