Which estimand is each trial interpretation answering?

First add the treatment conditions: interpretations conflict only when all estimand attributes match and the conclusions about that same treatment effect are incompatible.

They conflict only if the population, treatment conditions, outcome, time point, summary measure, and intercurrent-event strategy match. In a precision oncology trial, an effect under routine treatment changes answers a different question from efficacy assuming continued targeted therapy.

The supplied descriptions can establish whether estimand attributes match, but choosing between interpretations still requires the decision context and availability of outcomes after intercurrent events.

Suppose discontinuation follows adverse behavior: one supports an assignment-policy decision, while the other addresses efficacy under sustained exposure, so they need not conflict.

Two interpretations conflict only after their population, treatment conditions, outcome, time horizon, summary measure, and intercurrent-event strategy match. In a time-to-event analysis, that includes checking whether both report the same quantity, since a hazard ratio and a fixed-horizon risk difference can support different decisions even when derived from the same trial. Only then does comparing their effect estimates test a genuine disagreement. Otherwise, the practical task is to map each estimand to its decision, such as expected benefit under routine discontinuation patterns versus efficacy under continued treatment.

In a repurposing trial, a treatment-policy effect in the eligible population informs expected routine-use benefit, while a hypothetical continued-adherence effect informs whether the compound merits further efficacy testing.

Once the estimand attributes match, put each estimate beside its denominator and uncertainty interval. For a risk, report events over the relevant population at the specified time point, with an interval for the corresponding effect measure. Different point estimates may still be statistically compatible if their intervals overlap substantially, while narrow intervals around incompatible effects support a genuine disagreement.

They conflict only if those attributes match. Suppose a diet trial estimates assignment regardless of later adherence, while another estimates the effect under continued adherence: those are different treatment-effect questions, even with the same outcome and population.

For an economic evaluation, the practical test is whether each estimand supplies the treatment effect needed by the model’s decision context. An effect in trial participants under a hypothetical no-discontinuation strategy may support an efficacy comparison, while reimbursement under routine use may require the treatment-policy effect in the eligible population. If those effects feed different cost-effectiveness estimates, the apparent conflict can change the preferred option without either interpretation being internally inconsistent.

The practical test is whether both estimates would inform the same treatment decision for the same target population. Suppose two analyses use the same outcome and intercurrent-event strategy, but one reports a risk ratio in trial participants while the other reports a standardized risk difference in the treatment-eligible population. They can support different prescribing choices because baseline risk and population weighting change absolute benefit, even without a contradiction about relative effect.

The uncertainty most likely to flip the conclusion is whether an intercurrent event is both treatment affected and prognostic. Handling it with a treatment-policy versus hypothetical strategy can then change the effect’s direction, not merely its precision.

After naming the attributes, tie each estimand to the decision it is meant to support. Suppose the decision concerns effectiveness after treatment discontinuation: a treatment-policy estimand may fit, while a hypothetical estimand answers what would happen without discontinuation and usually needs different assumptions. The next check is whether the planned analysis and data collection can estimate the chosen effect, especially outcomes recorded after the intercurrent event. Which decision matters here: expected effect under actual use, or efficacy under a specified counterfactual condition?