Addressing treatment contamination in the design and analysis of trials of complex interventions

Student thesis: Doctoral ThesisDoctor of Philosophy


In mental health trials there is concern that the control treatment (therapy) might be contaminated. For example, this might be due to it being delivered by a clinician (therapist) who has been trained in the active intervention. It is often suggested by investigators, reviewers and funders of clinical trial proposals that cluster randomisation, with clusters defined at the level at which treatment contamination is thought to occur (e.g. therapists), should be used to prevent contamination. This thesis explores statistical methodologies used to account for both treatment contamination in the control trial arm and non-compliance in the experimental arm when estimating treatment efficacy. It considers trials where treatment receipt is measured on a binary scale and develops methods to accommodate continuous measures of treatment receipt.
The primary research objective was to compare the efficiency of two competing trial design options for evaluating efficacy in the presence of contamination. First, treatment allocation by cluster randomisation together with an estimator of the average treatment effect that accounts for clustered data. Second, allocation at the participant level, me-asurement of treatment receipt in all participants, and use of a randomisation-based estimator to target the complier average causal effect. Monte Carlo simulations under the two options with a binary measure of treatment receipt showed that the cluster randomisation design was more efficient under high levels of contamination, modest intraclass correlation coefficients and small cluster sizes. With a continuous measure of treatment receipt, the design with cluster randomisation was favoured more frequently as the difference between potential (counterfactual) doses became smaller, i.e. when there was greater non-adherence.
The secondary research objective was to develop a novel randomisation-based efficacy estimator in trials with contamination and non-compliance measured on a continuous scale. Efficacy estimators were applied to a trial of a psychological intervention for people with poorly controlled type 2 diabetes. There was some treatment contamination and non-compliance in the trial but little evidence of treatment efficacy according to the analyses. The tertiary objective was to review problems and solutions associated with contamination in published trials of complex interventions in mental health.
A main output from this project is the proposal of a novel valid estimator for evaluating efficacy and a demonstration of its utility. Another output is the provision of an online decision support tool (using the results from the simulations) to help those planning trials choose between the two competing design options for dealing with contamination.
Date of Award2018
Original languageEnglish
Awarding Institution
  • King's College London
SupervisorSabine Landau (Supervisor), Khalida Ismail (Supervisor) & Paul McCrone (Supervisor)

Cite this