Add estimand clinical trials perspective to estimation section

estimation.qmd

@@ -73,6 +73,69 @@ Model paramaters and other estimands are often symbolized using
 lower-case Greek letters: $\alpha, \beta, \gamma, \delta$, etc.
 :::
 
+#### Estimands in Clinical Trials
+
+In clinical trial research,
+the concept of an estimand has taken on particular importance following the ICH E9(R1) addendum.
+An **estimand** is a precise definition of a treatment effect reflecting a clinical question related to a trial objective [@nejm2024estimand].
+A clear specification of the estimand ensures that the clinical question of interest aligns with the statistical analysis plan.
+
+##### Intercurrent Events
+
+**Intercurrent events** are events that occur after treatment initiation that affect either the interpretation or availability of data associated with the clinical question of interest [@nejm2024estimand].
+Examples include treatment discontinuation due to side effects,
+use of rescue therapy,
+death (when death is not the primary endpoint),
+or withdrawal of consent.
+These events pose challenges for estimating treatment effects and require explicit strategies for how to incorporate them into analyses.
+
+##### Estimand Strategies
+
+Different strategies for handling intercurrent events lead to different estimands [@nejm2024estimand]:
+
+**Treatment policy strategy** (ITT analysis):
+
+The intention-to-treat (ITT) principle holds that treatment effects should be estimated from all participants assigned to a group,
+regardless of whether they adhered to their assigned treatment.
+This approach includes all participants regardless of intercurrent events and represents the "real-world effect."
+The ITT effect quantifies the causal effect of treatment assignment and should remain the primary effect estimate in randomized trials.
+
+**Hypothetical strategy** (trial-product or efficacy estimand):
+
+This strategy involves imagining a scenario in which intercurrent events (such as treatment discontinuation) wouldn't occur,
+and estimates the treatment effect in that hypothetical scenario.
+Investigators discard observed endpoints after discontinuation and impute counterfactual values.
+However, this approach is problematic because:
+
+- It generates estimates that haven't been — and probably couldn't be — observed
+- It assumes treatment benefits would be the same in participants who discontinued and those who remained on treatment, which is unlikely
+- It often inflates the treatment effect relative to the ITT estimate
+
+When reporting hypothetical strategies,
+authors must clarify that the resulting estimates do not correspond to an observed treatment effect and are likely to be larger in magnitude than the ITT effect [@nejm2024estimand].
+
+**Composite variable strategy**:
+
+This approach incorporates the intercurrent events into the estimate of the treatment effect,
+recognizing that these events are informative.
+For example,
+a composite variable could be defined as a participant meeting the endpoint criteria while continuing to receive treatment.
+
+##### Reporting Recommendations
+
+@nejm2024estimand recommend that clinical trial reports should:
+
+- Present ITT analyses as the primary results and generally the only analyses in the abstract
+- Fully report data on side effects and adverse events
+- If other estimand strategies are presented, clearly describe them using ICH terminology
+- Place non-ITT estimands in supplementary appendices, not alongside ITT results
+- Include detailed methods sections describing intercurrent events and their handling
+
+By precisely defining the estimand before collecting data,
+researchers ensure that the trial design,
+conduct,
+and analysis address a meaningful clinical question [@pohl2021estimands; @lawrance2020estimand; @nejm2024estimand].
+
 ### Estimates
 
 \index{estimate} \index{estimated value}

references.bib

@@ -151,6 +151,19 @@ @article{lawrance2020estimand
   url={https://link.springer.com/article/10.1186/s41687-020-00218-5}
 }
 
+@article{nejm2024estimand,
+  title={What's in an Estimand?},
+  author={Troxel, Andrea B and Gatsonis, Constantine A and Hogan, Joseph W and Hubbard, Rebecca A and Hunter, David J and Normand, Sharon-Lise T},
+  journal={New England Journal of Medicine},
+  volume={394},
+  number={1},
+  pages={6--9},
+  year={2025},
+  doi={10.1056/NEJMp2513633},
+  url={https://www.nejm.org/doi/full/10.1056/NEJMp2513633},
+  note={Perspective}
+}
+
 @article{efron1978assessing,
   title={Assessing the accuracy of the maximum likelihood estimator: Observed versus expected Fisher information},
   author={Efron, Bradley and Hinkley, David V},