Site-variable allocation ratios in randomized controlled trials: implications for sample size, recruitment efficiency, and statistical analysis.

Pavel S Roshanov

DOI: https://doi.org/10.1101/2024.11.03.24316666

2024-11-04

Abstract:Introduction: In multicentre randomized trials, some sites face logistical constraints that specifically affect their ability to recruit into one arm of the trial more than other arms. Often these are greater limits on their ability to deliver one of the study interventions. This paper proposes the use of allocation ratios that differ by site to increase recruitment capacity in asymmetrically constrained sites. Methods: Simulations of randomized trials assessed the impact of several allocation ratios (1:1 to 1:5) - and variation of ratios across sites - on sample size and recruitment capacity, and evaluated several adjustment approaches for time-to-event, binary, and continuous outcomes to prevent bias from site-variable allocation ratios. Results: Deviating from 1:1 allocation increases recruitment capacity within sites facing asymmetric constraints faster than it increases sample size requirements. For instance, a 1:3 ratio increased sample size by 35% but doubled the hypothetical recruitment capacity with fewer sites. The bias in treatment effect estimates that occurs when the baseline risk or outcome mean differ between sites allocated with different ratios was readily prevented with simple covariate adjustment or stratification by site or allocation ratio. Conclusions: Site-variable allocation ratios may relieve recruitment bottlenecks caused by asymmetric constraints in trial procedures that affect some of the sites in a trial. Accounting for the variation in allocation ratios during analysis is necessary to ensure unbiased treatment effect estimates. This strategy is particularly relevant for trials with low marginal costs for participant recruitment and follow-up, such as many large pragmatic trials embedded in routine care.

Epidemiology