Changes to design aspects of ongoing randomised controlled trials

Abstract

Unplanned changes to the research plan of a randomised controlled trial (RCT) may be a necessary response to unforeseen circumstances. Such changes can help ensure the value and relevance of a trial, but also have the potential to introduce a bias if performed inappropriately.

This thesis addresses three broad aims. The first was to highlight the methodological implications of approaches to making various unplanned changes through a series of simulation studies. The second was to estimate the prevalence of such changes to a contemporary sample of published RCTs registered with the Australian New Zealand Clinical Trial Registry (ANZCTR). The third was to assess, and extend as applicable, recommendations for ensuring such changes are performed appropriately and are well documented.

The simulation studies were performed using data from a large RCT of statin therapy for secondary prevention. The first of these demonstrated how unplanned changes (to various aspects of the design such as the primary endpoint) risks introducing a bias if undertaken with knowledge of treatment allocation, but not if made without knowledge of treatment allocation. The second investigated strategies for reacting to an observed imbalance on baseline prognostic factors in an RCT. It demonstrated that: continuing with original plans of unadjusted analyses, provided valid p-values irrespective of the direction of the prognostic imbalance, and any decisions informed by knowledge of the direction of the prognostic covariate imbalance were prone to bias. The third simulation study investigated strategies for reacting to a lower-than-expected event rate. It showed that switching to an expanded composite endpoint in response to a low pooled event rate does not inflate the type 1 error rate (and is likely to improve the statistical power, provided the expanded composite is sound).

A sample of 181 RCTs that were registered with the ANZCTR and had published a primary result paper were assessed for changes to their research plans, and whether these changes were legitimate or not (that is, made with or without knowledge of treatment allocation). A full audit was conducted on trials with accessible protocols (N=124) and a limited central review on trials without accessible protocols (N=57). The primary results publication was cross-checked with the protocol documents across six methodological aspects for the full audit, and trialists were contacted to resolve queries as necessary. The publication was checked against the registry record for the limited central review, over a subset of three methodological aspects. A key finding of this study was that it was often not possible to reliably assess whether changes had been made or whether changes were made in a blinded manner, based on review of documentation alone. After clarification was sought from the participating trialists, changes were found to be relatively common but typically made in a blinded manner. Improvements to the way unplanned changes to RCT research plans are documented are needed.

A set of recommendations to supplement existing guidelines relating to the documentation of RCT research plans was developed. One key recommendation was that trialists should adopt and implement the principles of the recently developed CONSERVE 2021 Statement to appropriately perform and document substantive changes. A second was that strategies that oblige trialists to upload full protocols (and protocol amendments) to clinical trial registries warrant further investigation and further consideration should be given to including all key methodological aspects in registry records.

The work undertaken provides valuable guidance for trialists (and stakeholders) on how to make and document a methodologically justifiable change to a planned RCT design/analysis and highlights the circumstances under which a change can lead to bias. This is important for ensuring that methodically unsound changes to RCTs are recognised and avoided, and that RCTs that have undergone methodically sound changes are not incorrectly dismissed as potentially biased.