Absence of optimism bias in industry-sponsored cancer trials.

Abstract

2020 Background: Randomized controlled trials (RCTs) in oncology power their studies to detect expected effect sizes. Prior studies have shown that there is optimism bias, the a priori overestimation of treatment effect size among cooperative-group-supported RCTs. However, it is unknown whether such bias is present among industry-supported trials. Methods: All published phase 3 clinical oncology RCTs were identified through ClinicalTrials.gov. Only superiority-design RCTs assessing a therapeutic intervention to improve disease-related outcomes were included. We compared the ratio of observed to expected hazard ratio (OEHRR) between trial subgroups using the Mann-Whitney U-test; comparisons of median OEHRR to a hypothetical median of 1 was performed using the Wilcoxon Signed Rank test. Results: We identified 140 phase 3 trials with available hazard ratio (HR) data. Of these, 123 trials (88%) were industry-sponsored, and 38 trials (27%) were cooperative-group-supported. For all trials, the median OEHRR was 1.099 (IQR = 0.855-1.291), demonstrating evidence of optimism bias when compared to a hypothetical median OEHRR of 1 (p = 0.018). In the subgroup analysis, compared to non-industry-sponsored trials (median OEHRR 1.253, IQR 1.061-1.334), industry-supported trials (median OEHRR 1.061, IQR 0.829-1.274) had a significantly lower OEHRR (p = 0.022) and did not demonstrate optimism bias (p = 0.15). Similarly non-cooperative group trials (median OEHRR 1.208, IQR 1.019-1.317) had a significantly lower OEHRR (p = 0.005) and did not demonstrate optimism bias (p = 0.562) compared to cooperative group trials (median OEHRR 1.208, IQR 1.019-1.317), which did demonstrate optimism bias (p < 0.001). Conclusions: Cooperative group trials, which represent a minority of trials, suffer from optimism bias. In contrast, industry-funded trials, which account for the majority of trials, do not demonstrate evidence of optimism bias, and have very close concordance between observed and expected effect size. These findings suggest that the powering and design of industry-funded trials better models the outcomes eventually observed. The reasons for this are likely complex and multifactorial, but may include financial constraint considerations, as industry-supported trials may not be as financially-limited as cooperative group studies. Therefore, industry-supported studies may be able to power trials with sufficient participants to reflect the estimated effect size.