Strategy | Explanation | Truncation-by-deatha | Treatment discontinuationb |
---|---|---|---|
Treatment policy | Measures the effect of the original decision to undertake a treatment, where the intercurrent event (e.g. discontinuation) is taken to be part of the treatment strategy. Cannot be used for terminal events, such as mortality. | Not applicable; relevant outcome data does not exist. | Recommended strategy, as it most closely links to the objective of evaluating the effect of treatment if introduced into a healthcare system. |
Composite | The outcome definition is modified to incorporate the intercurrent event, e.g. ‘requirement for ventilation’ is modified to ‘requirement for ventilation or death’. | Recommended strategy for patient benefit perspective, as it ensures death equates to a poor outcome. Care is required to ensure the outcome remains interpretable/clinically meaningful. | Not recommended, as the outcome becomes less interpretable/clinically meaningful. |
Hypothetical | Measures the effect of treatment in a hypothetical setting where the intercurrent event would not occur, e.g. the treatment effect if there was no discontinuation. | Not recommended; applies to a hypothetical setting which will never exist (no patients die), and so is difficult to interpret. | Recommended in a secondary estimand for discontinuation due to external factors (e.g. supply issues/lack of PPE), to evaluate the effect of treatment in settings where there was no supply issues/lack of PPE. |
While alive/while on treatment | Uses data prior to the occurrence of the intercurrent event; e.g. for ICU days, the number of days a patient was in ICU before they died would be used. | Recommended strategy for healthcare systems perspective, as it provides the real-world resource savings due to treatment. | Not recommended, as estimand becomes less interpretable/clinically meaningful. |
Principal stratum | Measures the effect of treatment in the (unknown) subpopulation of patients for whom the intercurrent event would not occur. | Not recommended, as interest for COVID-19 trials is likely to be a treatment effect in the entire population of patients, rather than in an unknown subpopulation. | Not recommended, as interest for COVID-19 trials is likely to be a treatment effect in the entire population of patients, rather than in an unknown subpopulation. |