The American Statistical Association (ASA) Biopharmaceutical Section Regulatory-Industry Statistics Workshop (ASA-FDA) is held in Washington D.C. annually. Originally a meeting among only the FDA statisticians, it was later expanded to all statisticians from the industry, academia and the FDA. The result is one of the most relevant conferences for statistical practitioners in the biopharmaceutical industry.
One of the biggest buzzwords at this year’s ASA-FDA conference was “Estimands.” For those who have not heard the term before, it is defined as follows:
Estimand: Is the target of estimation to address the scientific question of interest posed by the trial objective. Attributes of an estimand include the population of interest, the variable (or endpoint) of interest, the specification of how intercurrent events are reflected in the scientific question of interest, and the population-level summary for the variable.
This term was first defined in the ICH E9 (R1) addendum document (currently in its Step 2 draft version, dated 16 Jun 2017). What motivated this addendum is that over the years, there have been some protocols and new drug applications (NDAs) presented to the FDA that lacked clarity with respect to the study objectives and the related treatment effect parameters of interest, and how they are connected with the study design, conduct, analysis and interpretation as a whole. Specifically, intercurrent events, defined as events that occur after treatment initiation and either preclude observation of the variable or affect its interpretation (e.g., rescue medications, discontinuation of treatment, switching treatment, terminal events such as death, etc.) were often mishandled as simple “missing data” problems. Without carefully considering what impact each potential intercurrent event may have on the treatment effect in the planning stage, the treatment effect may not be estimated appropriately and/or may be misinterpreted in the clinical study report.
The structured framework in this guidance aims to align planning, design, conduct, analysis and interpretation, by starting with a clear trial objective and then defining a suitable estimand as the target of estimation for the particular trial objective. An estimand should specify the following four attributes:
- Population: the patients targeted by the scientific question;
- Variable (or endpoints): an measurement of some kind obtained from/for each patient, that is required to address the scientific question;
- Handling of intercurrent events: specifies how to account for intercurrent events to reflect the scientific question of interest;
- Population-level summary for the variable: provides, as required, a basis for a comparison between treatment conditions.
In the simplest case in which there are no intercurrent events, an example of an estimand might be the difference in mean systolic blood pressure between active drug and placebo in the change from baseline to Week 4 in the full analysis set.
Intercurrent events, on the other hand, are not as easy to anticipate and to plan for ahead of time. In ICH E9 (R1), 5 strategies for addressing intercurrent events are proposed. Below, examples for each strategy are provided where the intercurrent event is the use of rescue medication.
Strategy | Handling of Intercurrent Event | Example |
Treatment policy strategy (ITT principle) | Value of the variable of interest is used regardless of whether or not the intercurrent event occurs. | Compare “Drug X + rescue medication as needed” versus “Placebo + rescue medication as needed.” |
Composite strategy | The intercurrent event is integrated with one or more other measures of clinical outcome as the variable of interest. | Success declared if change from baseline is above a pre-specified threshold and no use of rescue medication occurred. |
Hypothetical strategy | Assume the intercurrent event did not occur. | Base analysis on assumptions of what the measurements would have been had rescue medication not been available to subjects |
Principal stratum strategy | Adjust the target population in which the intercurrent event would not occur. | Based on study entry criteria, select a target population of subjects who would not require rescue medication over the assessment period. Consider enrichment designs as well as run-in and randomized withdrawal designs to identify the target population. |
While on treatment strategy | Look at response to treatment prior to the occurrence of the intercurrent event. | Take average of measurements prior to use of rescue medication. |
Ultimately, one strategy may work well for one study but not the next. It is important to think through which intercurrent events may occur in the trial ahead of time, and how they may impact the interpretation of the treatment effect.
Lastly, ICH E9 (R1) provides a distinction between sensitivity analysis and the supplementary analysis. Sensitivity Analysis is a series of analyses targeting the same estimand, with differing assumptions to explore the robustness of inferences from the main estimator to deviations from underlying modeling assumptions and limitations in the data. Supplementary Analysis is a general description for analyses that are conducted in addition to the main and sensitivity analysis to provide additional insights into the understanding of the treatment effect. The term describes a broader class of analyses than sensitivity analyses, often performed on different estimands.
The strategies and statistical methods outlined here are not novel. The importance of ICH E9 (R1) is that it provides a framework for more careful planning of the statistical endpoints and analyses that will be used to estimate treatment effects. Thoughtful planning is necessary for the success of any clinical trial. Even a highly efficacious drug needs a strong foundation from a study protocol that includes clear study objectives and with the estimands properly planned in order to support an unchallengeable conclusion of “The drug works!” The structured framework presented in ICH E9 (R1) is not meant to be a gold standard that every trial must follow, but it certainly challenges us to dig deeper to consider the trial endpoints and objectives in a more systematic and complete way at the study design and planning stage.