Disentangling estimands and the intention‐to‐treat principle. (2nd December 2016)

• Record Type: Journal Article
• Authors: Leuchs, Ann‐Kristin; Brandt, Andreas; Zinserling, Jörg; Benda, Norbert
• Other Names: Rosenkranz Gerd guestEditor.
• Is Part Of: Pharmaceutical statistics. Volume 16:Number 1(2017)
• Page Start: 12
• Page End: 19
• Publication Date: 2016-12-02
• Subjects: causality -- clinical trials -- ICH E9 -- missing data -- nonadherence -- selection bias; Pharmacy -- Statistical methods -- Periodicals; Pharmacy -- Statistics -- Periodicals; 615.10727
• DOI: 10.1002/pst.1791
• Languages: English
• ISSNs: 1539-1604

Abstract

Abstract : Randomized controlled trials (RCTs) aim at providing reliable estimates of treatment benefit. Missing data and nonadherence to treatment are distinct problems that can substantially impede this task. In practice, the fact that the handling of missing data due to nonadherence affects the question that is addressed is often ignored. Estimands allow precisely predefining the question of interest. Estimands are definitions of that which is being estimated with regard to population, endpoint, and handling of postrandomization events (eg, nonadherence). Depending on the situation, different estimands are of relevance. Therefore, it is important that the intention‐to‐treat (ITT) principle, which is considered the gold standard for analyzing RCTs, does not restrict an RCT's primary objective to only one of several relevant estimands. Although much ambiguity is involved around what is considered to constitute the ITT principle, many associate ITT with completely following up all patients and including all data of all randomized patients as allocated into the analysis. This would restrict primary objectives to estimating the effect of treatment policy and is certainly not warranted in all situations. To maintain the advantage of having the clear recommendation to follow the ITT principle while allowing for various relevant estimands as primary objective, we argue that the appropriate way forward is to define ITT as including all randomized patients into the analysis set and each patient is to be allocated to his or her randomized treatment. This definition comprises the actual intent of ITT and can be fully implemented also in settings where complete follow‐up is impossible.