Background: Traditionally, when treatment estimates are reported in randomised controlled trials (RCTs), they are intention-to-treat (ITT) estimates. However, translating ITT estimates can be challenging, as they do not always provide the information that patients and clinicians want. For example, when non-adherence and poor participant retention occur in a trial, ITT estimates reflect the effect of being offered the intervention, rather than the benefits of adhering to the intervention. Moreover, when translating evidence into practice, it can often be of interest to understand what role mediators may play in facilitating the effects of an intervention. Estimating the effects of actually adhering to an intervention and of potential mediators requires alternatives to ITT estimates, but these alternatives can only be implemented without bias by using innovative causal inference and mediation methods. Motivated by issues encountered in a melanoma surveillance trial (MEL-SELF), we aim to conduct a methodological scoping review to identify and summarise causal inference and mediations methods that can be used in RCTs.
Methods: We searched MEDLINE and EMBASE for articles in which authors discussed causal inference and mediation methods in RCTs. One reviewer will undertake full-text screening and data extraction and a second reviewer will check extractions.
Results: 745 unique articles were retrieved from the database searches. After title and abstract screening, 117 studies were included for full-text screening. Preliminary findings indicate considerable development in applying causal inference methods to obtain estimates that can account for nonadherence or provide mediator effects, with these methods often comparing favourably to traditional, but biased approaches. As well as applying these methods in simple scenarios (e.g., where participants either adhere or don’t adhere, or where there is only a single mediator), more complex scenarios are also explored (e.g., where participants partially adhere or where there are multiple mediators). There has also been increased application of novel approaches such as combining causal inference methods with machine learning techniques such as random forest and stacking. Selection of included studies is currently underway, and final results will be presented at the conference.
Conclusion: This scoping review will summarise evidence on causal inference and mediation methods and their use in RCTs. Use of these methods will allow estimation of potentially more informative estimates than ITT in cancer-related trials, as they will assist in generating evidence that end-users are most interested in: what is the effect of an intervention if that actually adhere to it.