Sample size calculations for randomized controlled trials with repeatedly measured continuous variables as primary...

Randomized controlled trials (RCTs) with repeatedly measured continuous variables as primary outcomes are common. Although statistical methodologies for calculating sample sizes in such trials have been extensively investigated, their practical application remains unclear. This study aims to provide an overview of sample size calculation methods for different research questions (e.g., key time point treatment effect, treatment effect change over time) and evaluate the adequacy of current practices in trial design.

We conducted a comprehensive search of PubMed to identify RCTs published in core journals in 2019 that utilized repeatedly measured continuous variables as their primary outcomes. Data were extracted using a predefined questionnaire including general study characteristics, primary outcomes, detailed sample size calculation methods, and methods for analyzing the primary outcome. We re-estimated the sample size for trials that provided all relevant parameters.

A total of 168 RCTs were included, with a median of four repeated measurements (interquartile range 3-6) per outcome. In 48 (28.6%) trials, the primary outcome used for sample size calculation differed from the one used in defining the primary outcomes. There were 90 (53.6%) trials exhibited inconsistencies between the hypotheses specified for sample size calculation and those specified for primary analysis. The statistical methods used for sample size calculation in 158 (94.0%) trials did not align with those used for primary analysis. Additionally, only 6 (3.6%) trials accounted for the number of repeated measurements, and 7 (4.2%) trials considered the correlation among these measurements when calculating the sample size. Furthermore, of the 128 (76.2%) trials that considered loss to follow-up, 33 (25.8%) used an incorrect formula (i.e., N∗(1+lose rate) for sample size adjustment. In 53 (49.5%) out of 107 trials, the re-estimated sample size was larger than the reported sample size.

The practice of sample size calculation for RCTs with repeatedly measured continuous variables as primary outcomes displayed significant deficiencies, with a notable proportion of trials failed to report essential parameters about repeated measurement required for sample size calculation. Our findings highlight the urgent need to use optimal sample size methods that align with the research hypothesis, primary analysis method, and the form of the primary outcome.