Background

Outcomes for patients treated by the same surgeon tend to be more similar than those under the care of another surgeon due to surgeon experience, practice, and infrastructure. This phenomenon is referred to as the clustering effect. While the impact of clustering is widely recognized for cluster randomized controlled trials, its potential impact upon traditional randomized controlled trials (RCTs) evaluating skill-based interventions, such as surgical interventions, is less familiar. Within a traditional RCT, clustering can occur at the level of the surgeon, the center, or both. Failure to account for clustering may produce a biased estimate of the treatment effect (i.e., difference between groups) and an underestimate of the standard error, thereby increasing the risk of making a Type I error. To manage clustering efficiently, it is essential to inflate the sample size to account for within-cluster dependency, employ stratified randomization by surgeon and center, and include surgeon, or surgeon nested within center, as a random effect in multi-center trials.

Methods

We conducted a systematic electronic search of five high-impact orthopedic journals as ranked by Web of Science. To be eligible, studies had to meet the following criteria: published in English, involve multiple centers or multiple surgeons within a single center, include a skill-based intervention, be conducted on live human subjects, and be a RCT. A study was considered to account for clustering if it clearly described the sample size calculation and specifically mentioned surgeon-level clustering effects, such as the intraclass correlation coefficient, variance inflation factor, or design effect. Studies were also deemed to address clustering in their analysis if they used a mixed/random-effects model with surgeons as a random effect or explicitly stated that surgeon-level clustering effects were accounted for using a specified strategy.

Results

A total of 864 studies were included in title/abstract screening, of which 184 full text articles were assessed for eligibility. Eligibility and data extraction were completed by two independent reviewers where disagreements were resolved by a third individual. Of the 49 eligible studies, 25 were single center with multiple surgeons (median 4 surgeons per center, range 2 to 10) and 23 were muti-center RCTs. Six (6%) stratified their randomization by surgeon at a single centre but no studies accounted for clustering in the sample size calculation and only one study adjusted for surgeon by including surgeon as a random effect in a mixed-effects regression analysis.

Conclusions

The clustering effect in orthopaedic surgical trials is frequently overlooked, leading to potential biases in treatment effect estimates. It is crucial to emphasize the importance of accounting for clustering effects among researchers in this field to enhance the validity and reliability of trial outcomes.