Sample sizes in pilot cluster randomised trials from 2010-2020

Poster presented at ICTMC 2022

Over the past 10 years there has been a significant increase in the number of registered and published pilot cluster-randomised controlled trials (cRCTs). However, there is minimal guidance regarding appropriate sample sizes for pilot cRCTs, despite the fact that pilot studies are often used to estimate parameters for main trials and all pilot trials should justify their choice of sample size.

We aimed to conduct an audit of pilot cRCTs from 2010-2020 to document typical sample size for these trials, their justification, and explore trends across time and types of trial.

We performed a search of PubMed and Web of Science  databases between 2010-2020. The search terms “pilot” or “feasibility” were used, alongside specific terms to identify cRCTs.

From included studies we collated key information including planned and actual sample size (number of participants and clusters), number of study arms, whether and how the sample size was justified, and whether an ICC was estimated. To explore differences across trial types we also collected data regarding cluster type, therapeutic area and type of funding.

The literature search produced 3168 records of which 178 studies met the inclusion criteria and were analysed.

Almost all studies (90%) had two arms, and none had more than four. Most studies were publicly funded (76%), and clusters were predominantly healthcare related (56%), constituting primary care (12%), secondary care (17%) ,social care (16%) and healthcare professionals (11%).

We found a median of 4 planned clusters per arm (IQR: 3,6; range: 2,150: ) and median planned participants of 75 per arm (IQR 40,200; range:20,3000 ).Studies randomising education clusters had the largest sample size median of 175 per arm, (IQR 75,305) and those randomising health professionals (median 44 per arm, IQR: 30,96) had the smallest.

Most studies did not justify their choice of sample size. Where this was reported, the most common justification was to estimate parameters for a main trial, in order for it to be adequately powered. The majority did not report an ICC estimate: a key parameter for designing a main cRCT.

The research suggests that 4 clusters per-arm is the most commonly reported sample size for pilot cRCTs but there was a very wide range. There is little consistency in the justification for the chosen sample size. While most pilot cRCTs are intended to provide estimates of key parameters for a main trial, few reported the ICC, an important parameter in cRCT design. Future work will explore whether and when 4 clusters per-arm is an appropriate sample size.