In October 2014, ICH have released a concept paper E9 (R1) i.e. “Choosing Appropriate Estimands and Defining Sensitivity Analyses in Clinical Trials”. Recently in August 2017, ICH has also released draft guideline addendum for public consultation in EC Europe, MHLW/PMDA Japan, FDA US, Health Canada, MFDS Republic of Korea. Deadline for comments is expected by March 2018 from most of regulatory areas. Let us discuss here “what are Estimands?”, “why this addendum was required?”, and “What is aim of this addendum?”.
What are Estimands? How is it different from Estimate and Estimator? We know that estimate of population parameter, calculated using estimator, is a single value of a statistic. For example, sample average x is an estimate of population mean and average method is an estimator. Estimate is a numerical estimate of estimand and estimand relates to a scientific question of interest. For example, in a hypertension study, mean difference in change in systolic blood pressure (SBP) from baseline to week 16 between treatments in subjects dosed with study drug and completed study could be an estimand. In clinical trials, we can say that “An estimand reflects what is to be estimated to address the scientific question of interest posed by a trial.” In defining an appropriate ‘estimand’ for trial, and in determining a strategy for statistical analysis to derive estimated effects, a number of choices and assumptions need to be made. This leads to sensitivity analysis.
What lead to addendum for ICH E9?
- Clinical trials are planned to estimate true treatment effect if protocol is followed religiously.
- Non adherence to protocol and missing data makes it difficult to estimate treatment effect in presence of different patterns of non-adherence. For example, use of prohibited medications, dropout from study, non-adherence to study drug are different reasons that could affect the overall estimate of treatment effect.
- Handling of non-adherence to study protocol is handled differently by different trialists. There exist a confusion and controversy on definition and appropriate selection of estimands in clinical trials.
- There could be multiple ways to handle a same non-adherence. For example, following could be two estimands for a same trial.
“compare experimental drug X and placebo in terms of improving endpoint Y at time Z for all randomised patients, without regarding adherence to randomised treatment” or
“compare experimental drug X and placebo in terms of improving endpoint Y at time Z for all randomised patients if all patients had remained in the trial and received treatment as planned without rescue medication until time Z”
- At present, while sensitivity analyses are presented, they are rarely based on a systematic consideration of the various choices and assumptions made and are rarely discussed in terms of their relative importance for decision making. There are no clear regulatory standards to follow in defining an appropriate set of sensitivity analyses or for the joint interpretation of these “sensitivity analyses” with the primary analysis.
What ICH aims to discuss or clarify in addendum?
- To identify and harmonise around factors like outcome population, treatment adherence, time period of interest etc., which may be used to define and describe different estimands.
- To describe a frame work in which choice of appropriate estimand can be made and agreed between sponsor and regulator based on clear descriptions.
- One type of approach in defining estimand cannot be applicable to full range of experimental situation faced by drug developers. Considerations related to trial design differ according to therapeutic area.
- To discuss the importance and relevance of current approach in clinical trials that is to perform supportive analysis to investigate treatment effect on range of different estimands.
- To discuss a framework for sensitivity analysis which would cover 1) sensitivity analysis may be based on different mathematical assumptions, different methods or even different estimands 2) handling of missing data 3) interpretation of sensitivity analysis.