Adaptive designs for trials aiming to optimise implementation strategies and the effect of an additional interim analysis: a simulation study

Trials that aim to optimise the implementation of an intervention are often complex, requiring multiple, combined strategies and requiring uptake on multiple levels. Previous implementation trials have optimised implementation strategies using multi-arm cluster randomised control trials (cRCTs) but can be inefficient and waste resources. Adaptive designs may potentially improve efficiency of these trials, but under what design features and trial properties is unknown. A simulation study was performed to assess under what conditions, if any, one or two interim adaptive designs offer increased efficiency compared to a 'fixed trial' approach for implementation cRCTs. A four-arm cRCT was simulated with varied trial properties, with a fixed design or an adaptive design (varying by number of interim analyses and timing of interim analyses) and modelled using Bayesian hierarchical models. The adaptive design allowed for stopping early for futility and dropping an arm for futility. The power, type 1 error rate, and adaptive design decisions were compared between the designs across trial properties. Both one and two interim adaptive designs offered power gains and a lower type 1 error rate compared to the fixed designs across most trial properties. A high intra-class correlation (ICC) of 0.2 led to adaptive trials dropping effective arms or incorrectly stopping for futility more frequently, with the incorrect decisions being compounded when two interim analyses were used. The rate of these incorrect decisions was reduced when the first interim analysis in the two interim designs was delayed. Adaptive designs can offer improved efficiency, more power, and reduce resource wastage for trials compared to fixed designs.