Importance: Blinding of individuals involved in randomized clinical trials (RCTs) can be used to protect against performance and biases, but discrepancies in the reporting of methodological features between registered protocols and subsequent trial publications may lead to inconsistencies, thereby reintroducing bias.
Objective: To investigate inconsistency in blinding as reported in trial registries and publications.
Data sources: An exploratory dataset and a validation dataset were created. The exploratory dataset consisted of RCTs included in systematic reviews of adverse events from the SMART Safety database published between January 1, 2015, to January 1, 2020. The validation dataset was based on a literature search on PubMed for all registered RCTs published within the same time frame.
Study selection: Eligible RCTs for the exploratory dataset included were those that specified drug safety as the exclusive outcome and included at least 1 pairwise meta-analysis involving 5 or more RCTs of health care interventions. The validation dataset included a random selection of RCTs without restriction on outcome.
Data extraction and synthesis: Preferred Reporting Items for Systematic Reviews and Meta-Analyses reporting guidelines were followed during data extraction. RCTs were matched to their registries and information on blinding was extracted from both the journal publication and trial registry. Extraction was performed by 1 author and cross-checked by 2 additional authors, with discrepancies resolved via consensus. The data analysis was conducted between July 2023 and January 2024.
Main outcomes and measures: The primary outcome was inconsistency in blinding reports in the publication and the associated trial registry. Factors associated with the inconsistency were further investigated using multivariable logistic regression. The results were then compared with the validation dataset.
Results: A total of 1340 RCTs were included, with a median (IQR) sample size of 338 (152-772) participants. Of these, 749 (55.90%) were multiregional, 1220 (91.04%) were multicenter, and 835 (62.31%) were prospectively registered. The most frequently studied condition was cancer, representing 472 trials (35.22%). In the exploratory dataset, 1080 trials (80.60%) had inconsistent reporting of blinding in their published trial registry. Higher odds of inconsistency were associated with trials conducted as single-center (OR, 2.84; 95% CI, 1.24-7.74; P = .02) or those focused on cancer (OR, 3.26; 95% CI, 2.04-5.38; P < .001). Evaluation of the 98 RCTs in the validation dataset revealed that 70 (71.43%) had inconsistencies between the published trial and its registries. The occurrence of inconsistencies was significantly higher in the exploratory dataset than the validation dataset (P = .03).
Conclusions and relevance: In this systematic review of RCTs, there were significant inconsistencies in the reporting of blinding between trial publications and their corresponding registries. These findings underscore the importance of maintaining consistency between registered protocols and published trial reports to ensure methodological transparency and minimize bias.