The quality of the reported sample size calculations in randomized controlled trials indexed in PubMed

Abstract

Background: There are limited data on the quality of reporting of information essential for replication of the calculation as well as the accuracy of the sample size calculation. We examine the current quality of reporting of the sample size calculation in randomized controlled trials (RCTs) published in PubMed and to examine the variation in reporting across study design, study characteristics, and journal impact factor. We also reviewed the targeted sample size reported in trial registries.

Methods: We reviewed and analyzed all RCTs published in December 2014 with journals indexed in PubMed. The 2014 Impact Factors for the journals were used as proxies for their quality.

Results: Of the 451 analyzed papers, 58.1% reported an a priori sample size calculation. Nearly all papers provided the level of significance (97.7%) and desired power (96.6%), and most of the papers reported the minimum clinically important effect size (73.3%). The median (inter-quartile range) of the percentage difference of the reported and calculated sample size calculation was 0.0% (IQR -4.6%;3.0%). The accuracy of the reported sample size was better for studies published in journals that endorsed the CONSORT statement and journals with an impact factor. A total of 98 papers had provided targeted sample size on trial registries and about two-third of these papers (n=62) reported sample size calculation, but only 25 (40.3%) had no discrepancy with the reported number in the trial registries.

Conclusions: The reporting of the sample size calculation in RCTs published in PubMed-indexed journals and trial registries were poor. The CONSORT statement should be more widely endorsed.

Keywords: Crossover; Factorial; Protocol; Sample size; Trials.