PS04: On the Proper Specification of Clinical Questions and Its Vital Implications in the Estimand Framework

The estimand framework for the first time provided a systematic approach to ensure alignment among clinical trial objectives, trial conduct, data collection, statistical analyses, and interpretation of results. Per the ICH E9 (R1), an estimand is a precise description of the treatment effect being estimated and consists of 5 attributes namely, treatment condition, target study population, variable/endpoint of interest, population level summary and intercurrent events (ICEs). The framework has provided a step-by-step approach in defining an estimand. The first and most pivotal step is to clearly define the clinical question of interest. The question will then drive the objective, design and conduct of the study. Besides, a clear and accurate specification of the clinical question will guide the proper selection of methods for handling ICEs, especially in complicated situations such as non-inferiority trials that involves evaluation of multiple trials simultaneously that may differ in several aspects such as standard of care and use of rescue medications. However, in most clinical trial protocols and statistical analysis plans, there appears to be more emphasis in first outlining ICEs and associated strategies without a proper definition of the clinical question of interest. This is not desirable because treatment differences evaluated using different ICE strategies will answer different clinical questions of interest. In cases where the clinical question of interests is not adequately and precisely pre-specified, variations in the ICEs handling will occur, leading to inconsistent and misinterpreted study results. We contend that we should follow the original intent of the estimand framework which recommends that studies should first have a clearly defined clinical question of interest. Consequently, the pivotal step then would be to determine the most relevant clinical questions of interest that trials need to address to adequately evaluate the safety and efficacy of the drug to be approved for marketing. The choice of the clinical question/s of interest likely changes from indication to indication or even from study to study, taking into considerations the clinical implications to the patients and public health in general. Despite these variations, we contend that there are a few fundamental clinical questions that should be included as minimum requirements in all studies intended to support regulatory approval. First and foremost, in line with the FDA's mission of improving the public health, we need to ensure that any drug entering the market for the intended indication or as part of a treatment regimen, should not negatively affect the overall public health. This might be reflected closest as the effect of the drug in the trial regardless of how it is taken, i.e., a treatment policy estimand. Often a treatment effect evaluated using this estimand must be used in conjunction with input from outside sources including clinical judgement to ensure that it truly does not harm public health. Obviously, this alone does not warrant drug approval, especially when statistical superiority is not shown. Minimally, we also need the new drug to demonstrate the added benefits in the regimen that the regiment could not achieve otherwise, such as when the regimen benefits could be solely driven by other medical interventions subjects might have been exposed such as the use of other medications. These two questions sometimes could be answered by the same estimand, such as when there are no alternative treatments. We argue that these two estimands should be the minimum requirements for every study intended to support regulatory approval. The actual requirements may very well be more stringent. Another estimand that could be of interest is to see things from the perspective of treating doctors and their patients, in generating information that could be better used for treatment management. It seems natural to have the estimand that assures the drug contribution, or a variant of it, to be the primary estimand, while the estimand that assures no harm to public health, and the estimand for treatment management, as supplemental estimands. In this session we will invite leaders in the field to provide examples that follow these principled approaches to derive estimand that closely follows the targeted clinical questions, to fully realize the potential of the estimand framework that will inspire the improvements in trial design, conduct, data collection, analysis and reporting to elevate the utility of clinical studies.