Estimands
From TrialTree Wiki
Estimands
Overview
An estimand is a precise description of the treatment effect to be estimated in a randomized controlled trial (RCT). It defines what is being measured, for whom, and under what conditions. The concept was formalized in the ICH E9 (R1) addendum to improve clarity and consistency in the interpretation of clinical trial results.
Motivation
Estimands help align trial objectives with study design, data collection, and analysis by clearly articulating the target population, the outcome of interest, the way intercurrent events are handled (such as treatment discontinuation or death), and the statistical summary measure. This framework facilitates transparent and consistent decision-making, particularly when comparing results across trials or interpreting complex studies involving multiple outcomes or treatment strategies.
Components of an Estimand
According to the ICH E9 (R1) guidance, an estimand comprises five key attributes. The first is the treatment condition(s), which refers to the interventions or strategies being compared, such as Treatment A versus placebo. The second is the population, defining the group of individuals to whom the estimand applies—commonly all randomized participants. The third is the variable or endpoint, which represents the clinical outcome of interest, such as change in blood pressure at 12 weeks. The fourth is the handling of intercurrent events, which are post-randomization occurrences that affect either the interpretation or the existence of measurements (e.g., treatment discontinuation or use of rescue medication). The fifth is the summary measure, which is the statistical quantity used to summarize the comparison between treatment groups, such as a mean difference, risk ratio, or hazard ratio.
Estimation and Estimators
Once an estimand is clearly defined, it must be paired with an appropriate estimation strategy. An estimator is the statistical method used to generate an estimate of the treatment effect based on observed trial data. The selected estimator should align with the estimand's attributes, especially the strategy for addressing intercurrent events.
Common estimators include differences in means, odds ratios, risk differences, hazard ratios, and regression coefficients. Depending on the nature of the data and estimand, statistical methods such as generalized linear models (GLMs), mixed-effects models, multiple imputation, inverse probability weighting, G-computation, and Bayesian methods may be applied. These choices influence the assumptions required for valid inference, the robustness of results to protocol deviations or missing data, and the overall interpretability of the estimated treatment effect.
Examples
In a study examining systolic blood pressure, the objective might be to assess the effect of Drug X compared to placebo after 12 weeks. The estimand would define the treatment conditions as Drug X versus placebo, the population as all randomized patients, the outcome variable as systolic blood pressure at 12 weeks, and intercurrent events such as treatment discontinuation would be handled using a treatment policy strategy. The summary measure could be the mean difference in systolic blood pressure between the groups. A suitable estimator in this case might be linear regression with treatment group as a covariate.
In another example, a trial evaluating the effect of a diabetes drug on achieving HbA1c levels below 7% at six months may specify a binary outcome. Here, treatment discontinuation could be handled using a hypothetical strategy, with the summary measure being the risk difference. The estimator might involve logistic regression combined with multiple imputation to address missing data due to the intercurrent event.
Intercurrent Events and Strategies
Intercurrent events are critical to the definition of an estimand and must be addressed explicitly. Several strategies exist for this purpose. The treatment policy strategy includes all observed outcomes, regardless of whether an intercurrent event occurred. The hypothetical strategy considers what the outcome would have been if the intercurrent event had not occurred. The composite strategy integrates the intercurrent event into the definition of the outcome itself. The while on treatment strategy analyzes only the data collected prior to the event. Lastly, the principal stratum strategy focuses on the subgroup of participants for whom the intercurrent event would not occur, which may require assumptions or modeling.
Relevance to Trial Design and Analysis
Clearly defining the estimand at the design stage ensures coherence between the research question, data collection, statistical analysis, and interpretation of trial results. This alignment helps avoid post hoc ambiguity and ensures that conclusions are valid and transparent. Estimands support reproducibility, facilitate meta-analysis, and strengthen regulatory and clinical decision-making by making the target of estimation explicit and interpretable.
Related Pages
References
- ICH E9(R1) Addendum on Estimands and Sensitivity Analysis in Clinical Trials to the Guideline on Statistical Principles for Clinical Trials
- Hughes, M. D., & Daniels, M. J. (2021). Understanding Estimands: A Practical Guide. Clinical Trials.
- Meyer, R. D., et al. (2020). Statistical Considerations in Estimand Implementation. Pharmaceutical Statistics.