Practical guidance on accessible statistical methods for different estimands in randomised trials
Lead Research Organisation:
UNIVERSITY COLLEGE LONDON
Department Name: MRC Clinical Trials Unit
Abstract
Randomised controlled trials (RCTs) involve assigning patients at random to different treatments. Because of the random assignment, patient factors (such as age, sex, and disease stage) will, on average, be the same between treatment groups, meaning that any differences in outcomes (such as mortality) between the groups will likely be due to the treatment. Because of this, RCTs are considered the gold standard for generating high-quality evidence on whether new treatments are safe and effective.
However, during most RCTs patients will experience unplanned events which mean they have to stop treatment early, switch from one treatment to another, or receive additional treatments outside of the trial. This can complicate the interpretation of results, as it is not always clear how such unplanned events are handled in the calculation of the treatment effect. For instance, the effect of starting a patient on the treatment, regardless of whether they continue taking treatment or switch to something else, could be calculated. Alternatively, the effect of actually completing a course of treatment could also be calculated. Most trials do not explicitly state which treatment effect they have calculated, which can lead to misinterpretations.
New international guidelines have recently been developed by drug regulators and the pharmaceutical industry to highlight the importance of clear reporting of the treatment effect's interpretation. However, the guidance does not provide the statistical methods for calculating these treatment effects. Whilst some statistical methods have been proposed for calculating different treatment effects, these are often written in an overly technical manner, are published in academic journals unfamiliar to those running trials, and do not provide the computer code required to implement such methods. Thus, many trials use inappropriate methods to calculate treatment effects. As such, there is urgent need for guidance on appropriate, accessible, statistical methods to calculate treatment effects in RCTs.
However, during most RCTs patients will experience unplanned events which mean they have to stop treatment early, switch from one treatment to another, or receive additional treatments outside of the trial. This can complicate the interpretation of results, as it is not always clear how such unplanned events are handled in the calculation of the treatment effect. For instance, the effect of starting a patient on the treatment, regardless of whether they continue taking treatment or switch to something else, could be calculated. Alternatively, the effect of actually completing a course of treatment could also be calculated. Most trials do not explicitly state which treatment effect they have calculated, which can lead to misinterpretations.
New international guidelines have recently been developed by drug regulators and the pharmaceutical industry to highlight the importance of clear reporting of the treatment effect's interpretation. However, the guidance does not provide the statistical methods for calculating these treatment effects. Whilst some statistical methods have been proposed for calculating different treatment effects, these are often written in an overly technical manner, are published in academic journals unfamiliar to those running trials, and do not provide the computer code required to implement such methods. Thus, many trials use inappropriate methods to calculate treatment effects. As such, there is urgent need for guidance on appropriate, accessible, statistical methods to calculate treatment effects in RCTs.
