Quantifying Effect Sizes in Clinical Trials

Introduction: This thesis discusses the importance of the target effct size (ES) used in clinical trials of health interventions. It investigates the common methods of eliciting the target difference and whether target values are optimistic or unrealistic. Regression to the mean (RTM) is shown for trials in sequence, which is assessed through simulations and an adjustment developed to adapt for this bias. Research Question: Investigating currently used methods for eliciting the target difference and optimal methods for adjusting for RTM. Methods: Firstly, a review of the Health Technology Assessment (HTA) journal trial reports of parallel-group randomised controlled trials (RCTs) was performed. The standardised observed and target ES were compared for various clinical areas and elicitation methods. Second, performing simulations of trials in sequence to investigate the effect of RTM. A mathematical solution was evaluated to confirm these simulated results. Results: A review of 107 HTA reports showed the median standardised target ES is 0:30 (mean= 0:30), and the median standardised observed ES is 0:11 (mean= 0:19). Use of previous research was the most common method of elicitation. Simulations showed RTM occurs for trials in sequence, an adjustment method has been developed and proven mathematically, which depends only on the power of the first trial. Conclusions: This thesis demonstrates the most common method of target difference elicitation is the use of previous research. This method leads to RTM of the observed ES. An adjustment based on the power used in the initial trial power and the progression criteria in pilot studies has been developed and tested. If trialists adopt this adjustment then trial sample sizes, though slightly inflated, would potentially provide more realistic estimates of the target ES.