The purpose of a randomized clinical trial is to assess whether there is a causal relationship between an intervention and outcome of interest. While it is commonly believed that the purpose of randomization is to remove bias in treatment assignments or balance baseline characteristics, we will demonstrate that, in reality, randomization is the critical design feature that enables us to conclude that observed statistical associations between interventions and patient responses are causal relationships.
Using easily understood examples, interactive activities and minimal mathematics, we will demonstrate how randomization is essential to objective assessments of causality, and we will discuss the implications this has for study design, conduct and analysis. For example, we will explain why the Intention to Treat (ITT) principle is essential to inference based on randomization and not merely a means to conservatively estimate the “real-world” treatment effect (another popular misconception). We will address how missing data or incomplete follow-up compromise our ability to draw causal conclusions and explain why there cannot be a statistical solution to the problem of missing data; the only remedy is to minimize missingness to the greatest extent possible and to employ sensitivity analyses to try to contain the problem. As a result, it is critical that clinical coordinators and investigators understand the importance of ensuring that data are as complete as possible, and it is imperative that studies be designed to ensure that collection of efficacy, laboratory, adverse event, or other assessments be continued until the planned end of follow-up even after participants discontinue their treatment.
We will address the impact of non-adherence to assigned treatment on the study results and the conclusions that may be drawn. Even when subjects are non-adherent to their assigned treatment, the intention-to-treat analysis still provides valid causal inference regarding the effect of assignment to treatment. While clinical investigators may have a strong desire to assess the effect of receipt of treatment under conditions of full adherence, we will show that unless we force subjects to adhere, it is not possible to conduct a trial to measure this effect directly, and no statistical procedure can recover this effect from a trial without full adherence. In particular, alternatives to ITT such as “per-protocol” and “as-treated” analyses do not measure this effect. Instead, they merely subvert the randomization and therefore cannot yield valid causal conclusions.
We will also discuss the role of imbalances in baseline risk factors and why observed (or unobserved) imbalances do not invalidate trial results. We will consider the role of baseline adjustment in the analysis of trial data and the desirability of allocation procedures, such as “minimization” that seek to ensure balance.
Additional topics include: the impact of errors in randomization and other protocol violations; assessments of safety including the analysis of adverse events; analysis of post randomization subgroups such as survivors or responders; and implications for the conduct and analysis of non-inferiority trials.
Anyone involved in the design or implementation of randomized clinical trials with an interest in ensuring the quality and integrity of those trials. This includes clinicians, clinic coordinators, statisticians, data managers, and others. No formal statistical background is required, and both statisticians and clinicians can expect to benefit.
To help participants develop a deeper understanding of