Randomized phase II trials (RP2T): Selection design or poor man's phase III?

Abstract

6595 Background: With the increasing number of new anticancer agents, RP2Ts have been used more often, aiming to select active agents to be tested further. RP2Ts are typically underpowered for statistical comparisons of endpoints between treatment arms. We sought to investigate how frequently formal statistical testing (as opposed to “picking the winner”) has been used in RP2Ts. Methods: We systematically searched PubMed for RP2Ts on systemic cancer therapy, using the medical subject headings “drug therapy” and “neoplasms,” and limiting the search to years 1995/1996 and 2005/2006, and to publication type “randomized controlled trials.” For each study, relevant data that were collected independently by two investigators included reporting of P values and confidence intervals (CIs) for primary (PE) and other endpoint(s) results. Results: 288 articles were retrieved and 107 were eligible: 22 were published in 1995/96 and 85 in 2005/06. Median number of arms and patients per RP2Ts were 2 and 94, respectively. PE was response rate in 71 (66.4%), a control arm was present in 55 (51.4%) and 48 (45%) were partially or entirely sponsored by for-profit organizations. For the PE, P values were reported in 50 (47%), and CIs in 69 (65%) RP2Ts. Either P values or CIs for the PE were reported in 85 (79.4%; 95% CI: 70.8% - 86.1%) cases. No temporal trend was identified in such practice (17 of 22 RP2Ts from 1995/1996 and 68 of 85 RP2Ts from 2005/2006). Likewise, source of funding and use of a control group were not associated with reporting P values/CIs for PE. Conclusions: Our study suggests that formal statistical comparisons are frequently performed between arms of RP2Ts. Further studies are needed in order to identify factors that are associated with such practice. [Table: see text]