The Estimands Framework and Causal Inference: Complementary not Competing Paradigms Publication type: Article Type Journal: Journal Volume: 00
Thomas Drury, Jonathan W. Bartlett, David Wright, Oliver N. Keene · 2024
Read The Estimands Framework and Causal Inference: Complementary not Competing Paradigms Publication type: Article Type Journal: Journal Volume: 00 by Thomas Drury, Jonathan W. Bartlett, David Wright, Oliver N. Keene (2024) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2412-12380.txt.
Introduction
Clinical trial teams now write down exactly what effect they want to estimate before a study starts, using a guide called the estimands framework. Statisticians working in causal inference also define an estimand, and the overlap confuses both camps. Thomas Drury, Jonathan Bartlett, David Wright, and Oliver Keene compare the two approaches in 2024 with worked examples. They show how each one defines a summary effect for a stated group of patients.
The trial framework, they argue, is structured and plain enough for non-mathematicians, which helps teams talk clearly about goals. The causal approach states its quantity with full mathematical care and draws its assumptions as graphs. The authors urge both sides to treat the tools as partners, not rivals.
Catalog record
| CREATOR | Thomas Drury, Jonathan W. Bartlett, David Wright, Oliver N. Keene |
|---|---|
| YEAR | 2024 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 49 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2412-12380
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
The Estimands Framework and Causal Inference: Complementary not Competing Paradigms Publication type: Article Type Journal: Journal Volume: 00 Thomas Drury, Jonathan W. Bartlett, David Wright, Oliver N. Keene Abstract The creation of the ICH E9 (R1) estimands framework has led to more precise specification of the treatment effects of interest in the design and statistical analysis of clinical trials. However, it is unclear how the new framework relates to causal inference, as both approaches appear to define what is being estimated and have a quantity labelled an estimand. Using illustrative examples, we show that both approaches can be used to define a population-based summary of an effect on an outcome for a specified population and highlight the similarities and differences between these approaches. We demonstrate that the ICH E9 (R1) estimand framework offers a descriptive, structured approach that is more accessible to non-mathematicians, facilitating clearer communication of trial objectives and results. We then contrast this with the causal inference framework, which provides a mathematically precise definition of an estimand, and allows the explicit articulation of assumptions through tools such as causal graphs. Despite these differences, the two paradigms should be viewed as complementary rather than competing. The combined use of both approaches enhances the ability to communicate what is being estimated. We encourage those familiar with one framework to appreciate the concepts of the other to strengthen the robustness and clarity of clinical trial design, analysis, and interpretation. keywords ICH E9(R1), Estimands, Causal Inference, Potential Outcomes † † copyright-year: 2023 † † startpage: 1 † † titlemark: The Estimands Framework and Causal Inference: Complementary not Competing Paradigms † † corresponding: Corresponding author: Thomas Drury Drury et al. 1 Introduction There has been widespread adoption of the estimands framework by the pharmaceutical industry following the publication of ICH E9 (R1) ( ICH-E9-R1 ; Mallinckrodt2020 ; Fletcher2022 ) . There has also been some use of the framework in academia ( Cro2022-he ) . The purpose of the framework is to provide context and clarity on what is being estimated in a clinical trial by formally defining a quantity labeled an estimand that has five components: Population, Treatments, Endpoint, Summary Measure, and Intercurrent Events (with strategies chosen to address them). The framework was designed to provide a common language for different stakeholders to discuss and define effects of interest in clinical trials ( ICH-E9-R1-Training ) . It was also, in part, an acknowledgment that to gain a full understanding of an effect targeted in randomized clinical trials (RCTs), we often need to consider how post baseline events impact the outcomes of interest. Although the framework is a positive step forward in the design, analysis and reporting of clinical trials, a potentially confusing factor is that the term “estimand” has been used to describe a similar idea in the Causal Inference literature for many years ( Hernan2024 ) . A further complication is that what is labelled an estimand in each approach appears to partially overlap, and both relate to effects on patient outcomes under different treatment conditions. In this paper we review how ICH E9 (R1) and the causal inference literature define what effect is being estimated in the context of a randomized trial. We discuss the origins of the two approaches, how they relate, and provide examples illustrating how each can be used to define the effect of interest. Although there are pros and cons to both approaches, they share the overall aim of providing clarity on what is being estimated and we suggest that it may be beneficial to use aspects from both when defining the treatment effect targeted in clinical trials. We also show that, although not explicitly referenced, the ICH E9 (R1) estimands framework relies on concepts developed as part of causal inference methodology. Finally, we discuss our interpretation of some more ambiguous aspects of ICH E9 (R1) which are not clear and comment on some recent opinions from the causal inference literature about the utility of the estimand framework. This paper is organised as follows: in section 2 we outline some examples of clinical trials describing potential effects that could be targeted, section 3 gives a brief history and outline of the ICH E9 (R1) estimands framework, section 4 reviews relevant aspects of causal inference. Section 5 illustrates how the example effects from section 2 could be formulated using both approaches. Section 6 contrasts and discusses several aspects of both the estimands framework and causal inference and finally we provide a conclusion in section 7. 2 Example Clinical Trials Randomised clinical trials provide a rigorous basis for comparing groups of patients assigned different pre-specified treatments. With complete follow-up, randomization guarantees that the observed treatment effect on the outcome is, aside from random error, due to the treatment each individual was assigned. However, this effect of assignment is a combination of the effect of taking the assigned treatment plus any subsequent effects from other events that occur after assignment, for instance, treatment discontinuation or use of additional (or rescue) medications. In many clinical trials, this effect of assignment may not be the most relevant clinical question of interest to all trial stakeholders and other questions are often important, including the effects attributable to specific treatments (or combinations of treatments). Events such as treatment discontinuation or use of additional medication can affect the value and interpretation of the outcome measured. For some clinical questions of interest, this can lead to difficulty using the recorded outcome measures directly to address the research questions in the trial and it is typically necessary to provide more information about how these events are reflected in the effects being targeted. We present three examples based on real clinical trials that exhibit some of the issues discussed above. We do not discuss the trials in detail but use some aspects as motivation to aid the discussion of the similarities and differences between the causal inference and ICH E9 (R1) approaches to defining what is to be estimated. 2.1 Diabetes The PIONEER 1 trial compared oral semaglutide to placebo in adult patients with type 2 diabetes, with a primary endpoint of change from baseline HbA1c ( Aroda2019-ie ) . During the trial, patients could discontinue assigned treatment and/or receive rescue medication if they did not maintain acceptable glycaemic control, both of which would likely change their HbA1c. Overall, more placebo patients needed rescue medication compared to semaglutide, creating a differential impact between the groups. Therefore, comparing change in HbA1c values at the end of the trial considers the combined effect of the semaglutide taken plus any rescue taken and does not address the treatment effect caused by or attributable to taking only semaglutide itself for the duration of the trial versus only placebo for the duration of the trial . 2.2 Nasal Polyps The SYNAPSE trial compared mepolizumab and placebo in chronic rhinosinusitis with nasal polyps after 52 weeks ( Han2021-aj ) . The co-primary endpoints were nasal polyps score and nasal blockage score at the end of the trial. A post baseline event was surgery for nasal polyps and treatment with mepolizumab was intended to reduce the number of patients having to undergo surgery. Although surgery is likely to improve subsequent nasal scores, it represents