Many multicenter randomized controlled trials do not account for center effect: a methodological review

Multicenter randomized controlled trials are considered the gold standard for evaluating medical interventions, yet they raise important methodological challenges related to center effects. We aim (1) to assess how center effect is taken into account in multicenter randomized controlled trials, during randomization and statistical analysis; (2) to assess whether any subgroup analyses by center have been conducted and their results.

In this methodological review, we searched MEDLINE via PubMed to identify primary reports of multicenter randomized controlled trials published in five major general medical journals in 2024. Selection and extraction were done in duplicate.

Of 188 multicenter randomized controlled trials included, 94 (50%) assessed pharmacological interventions, 76 (40%) nonpharmacological interventions, and 18 (10%) both pharmacological and nonpharmacological interventions. The median number of centers was 32, interquartile range: 16-73. The randomization was reported as stratified or minimized on center in 92 trials (49%), and this seemed more frequent for nonpharmacological intervention trials (n = 57, 75%) than for pharmacological intervention trials (n = 25, 27%). Regarding statistical analysis, 37 trials (20%) reported to account for center in primary analysis. This was also more frequently reported in nonpharmacological intervention trials (n = 23, 30%) than pharmacological ones (n = 12, 13%). Additionally, only 33 trials (36%) accounted for center effect in their primary analysis after stratification or minimization by center in the randomization. Finally, eighteen trials (10%) reported a subgroup analysis by center with a significant interaction test in one.

Only few recent multicenter randomized controlled trials account for center effect in the randomization and statistical analysis. Subgroup analysis by center was rarely conducted.