Considerations on the Use of Estimands Within the Integrated Summary of Effectiveness and Integrated Summary of Safety, and Benefit/Risk Analysis

On the ISE side, the use of pooled analysis techniques is well established. There is an expectation that the increasingly common use of pre-defined estimands in the individual clinical trials to be combined in an ISE will alleviate an important source of heterogeneity. In fact, before implementing the estimand framework, pooled analyses were used to combine clinical studies based on the sam e endpoint definition, with little or no attention to intercurrent events that could have been managed in different ways, leading to a heterogeneous impact on
the analysis results. Through the estimands, the intercurrent events are incorporated in the treatment effect definition, opening the space for more transparent and hopefully homogeneous pooled analyses.

The combined studies might have implemented different strategies to take care of intercurrent events, for example, some studies may have used the treatment strategy and others the hypothetical or the composite strategies. One of the objectives of an ISE might be defining a target estimand implementing a common
strategy for all studies. An example of a primary efficacy estima nd based on the treatment policy strategy, reported in the statistical analysis plan of an ISE on a new in vestigational medicinal product (IMP) for asthma, is described below.

Difference between Investigational Medicinal Products (IMP) and placebo in the change from baseline t o Week 24 of forced expiratory volume (FEV)₁ regardless of adherence (i.e., whether the patient is able to remain on treatment) and whether additional or alternative medication is used. This estimand is specified by the following four attributes:

• Treatment: treatment condition of interest is IMP regardless of whether the patient continues with it until week 24 and whether additional or alternative medication is used; alternative treatment condition is placebo regardless of whether the patient continues with it until week 24 and whether additional or alternative medication is used.
• Population: patients with asthma.
• Variable: change from baseline to week 24 in FEV1. How to account for intercurrent events: already covered in treatment attribute.
• Population-level summary for the variable: difference between treatments in mean changes from baseline at week 24.

A table providing the details for the actual implementation of this target estimand at the study level would be needed as well as a method to account for missing data in the statistical analysis.

In some cases it is not possible to implement a common target estimand. This could happen in the example above because in some of the studies, patients prematurely stopped the study treatment and/or patients
taking alternative medications were withdrawn from the study. In this case a reasonable approach could be to group the studies based on the primary estimand originally d efined at the study level. Another alternative might be the identification of treatment effects in strata based on potential experience of intercurrent events. Estimands based on Principal Stratum strategies might be useful for this purpose, i.e., to estimate the treatment effect in the subset of patients in whom the intercurrent event( s) of interest would (or would not) occur, for example, the subset of patients who would not use rescue medication regardless of which treatment arm they were assigned.