Design and analysis of trials evaluating proportionate interventions and trials with intervention induced clustering

Introduction: Individually randomised controlled trials (iRCTs) of complex interventions commonly induce clustered outcomes in the intervention arm only, termed partially nested trials (pnRCTs). In addition, iRCTs are increasingly used to evaluate interventions delivered proportionate to individual need. This can result in only some of the intervention arm having clustered outcomes due to post randomisation allocation to clusters, termed within-arm pnRCTs. Research question: What elements need to be considered in the design, analysis and reporting of complex intervention trials with continuous outcomes, with a particular focus on proportionate interventions and intervention induced clustering in one trial arm? Methods: Firstly, a systematic review of trials of proportionate interventions was performed. Simulation of pnRCTs and within-arm pnRCTs were used to investigate appropriate analysis methods. Sample size formulae for such RCTs were identified and summarised. Finally, a review of publicly funded iRCTs with clustering was undertaken. Results: Proportionate interventions commonly induced within-arm partial nesting. Appropriate analysis methods were identified and demonstrated for pnRCTs, although with few clusters, small cluster sizes, and small intracluster correlation coefficient (ICC) there was no optimal method. Accounting for non-random clustering in within-arm pnRCTs was not possible; however, under realistic scenarios ignoring clustering can provide valid statistical inference. Sample size formulae for pnRCTs require an ICC estimate. From 15 publicly funded iRCTs the median healthcare provider ICC was 0.009. To improve transparency an additional Consolidated Standards of Reporting Trials-nonpharmacologic treatments item related to reporting ICC is suggested. Conclusions: This thesis demonstrates the extent of clustering in both proportionate intervention trials and publicly funded iRCTs. Appropriate analysis methods are demonstrated for pnRCTs. For within-arm pnRCTs it is typically recommended to ignore the clustering. Sample size methods are summarised and empirical ICCs provided. This work provides practical guidance for design, analysis and reporting for continuous outcomes in RCTs with intervention induced clustering in one trial arm.