Table 1 Examples of challenges in determining the estimand based on the method of analysis in several fictional trials

Time to non-fatal myocardial infarction will be analysed using a Cox regression model. Patients who die before experiencing an event will be censored at the time of death.

Hazard ratio in the hypothetical setting where patients do not die.

This is because the Cox model assumes that censored patients are still alive and at risk of non-fatal myocardial infarction.

Quality of life (using EQ-5D) will be collected at the beginning of every chemotherapy cycle, until disease progression or death. Quality of life will be analysed using a repeated-measures mixed-model, with treatment, time point, and their interaction included as fixed effects, and patient as a random effect. Intention-to-treat analysis (all available data analysed according to allocated treatment group) will be used.

Difference in the means in hypothetical setting where patients do not die and do not experience disease progression.

This is because data is set to missing after disease progression or death, but the repeated-measures mixed-model implicitly imputes what the missing data would have been had patients been alive and not progressed.

Difference in the mean blood pressure at 6 months will be analysed using a complier-average causal effect (CACE) analysis. All randomised patients will be included in the analysis.

Difference in the means in the subset of patients who would have complied under both treatment conditions.

This is because while complier-average causal effect analysis is undertaken on the full trial population, it only applies to the subset of participants who would comply under either treatment strategy.