Abstract
BACKGROUND: Estimation of comparative treatment effects between randomized groups is well-supported in randomized trials. By contrast, treatment group-specific inferences are challenging, as patients are selectively chosen for enrollment, and such inferences are formally discouraged by the CONSORT guidelines. The present study is the first-large scale assessment of the proportion of phase III oncology trials that present treatment group-specific inferences. METHODS: Published phase III randomized oncology trials were screened from ClinicalTrials.gov. Treatment group-specific inferences were defined by the presence of 95% CI or standard error for treatment-specific outcomes. RESULTS: A total of 774 phase III trials enrolling 568,080 patients were included. Treatment group-specific inferences were present in 58% of trials (446 of 774), and appeared to be increasing over time (adjusted odds ratio for the publication year, 1.11; 95% CI, 1.06 to 1.17; p < 0.0001). Of the remaining 328 trials, 49 (6%) described group-specific outcomes with measures of variability, such as interquartile range, and 279 (36%) provided point estimates only (e.g., median) for group outcomes. INTERPRETATION: The majority of published phase III oncology trials present treatment group-specific inferences. However, this inference lacks statistical support, as patients are not randomly sampled from the underlying population, and conflicts with CONSORT guidelines. While ongoing methodological efforts to improve the transportability of treatment group-specific inferences are promising, conventional attempts to generalize treatment-specific outcomes from randomized trials may be misleading. Instead of inference, treatment group-specific outcomes should be described using measures of variability.