Table 2.

Three statistical approaches for comparing the effect of randomized treatments on functional outcomes “truncated due to death”

Statistical approach	Target patient population	Advantages	Disadvantages
Survivors analysis	Patients who survive in each treatment group	- Simple to compute - If mortality is independent of treatment, provides an estimate of the causal effect of the treatment on the always survivors	- If the effect of the treatment is to sustain frail patients, then the estimated treatment effect is a biased estimate of the causal effect of the treatment on the always survivors and may be misleading - Randomization is not guaranteed to be preserved within the subset of survivors - Does not include all randomized patients, violating the intention to treat principle for randomized trials
Survivor average causal effect (SACE)	Always survivors—that is, patients who would survive regardless of which treatment they receive	- Estimates the causal effect of the treatment on the functional outcome among always survivors - Randomization is preserved within the subset of always survivors so the comparison of randomized treatment groups is unbiased - When the intervention does not affect mortality, SACE is identified from the observed data (ie, all survivors are always survivors) without making any untestable assumptions	- Comparison of the functional outcome is made among a subset of patients (ie, those who would survive regardless of which treatment they received) that is not directly identifiable - Requires eliciting expert opinion on assumptions that are not testable within the observed data (eg, there are no patients who would survive under control but who would die under intervention) - Does not include all randomized patients, violating the intention to treat principle for randomized trials
Composite endpoint	All randomized patients	- Simple to compute (both quantiles of and the rank statistic for the distribution of the composite outcome) - All randomized patients are included in the analysis, consistent with the intention to treat principle - Randomization is preserved so the comparison of randomized treatment groups is unbiased - Provides a hypothesis test for comparing the distribution or a specific quantile (eg, median) of the composite endpoint across treatment groups that has a causal interpretation - The distribution of the composite endpoint in each treatment group can be examined to identify if treatment groups differ in mortality and/or the functional outcome	- Requires eliciting expert opinion on ordering mortality and the functional outcome - The size of the rank statistic may be difficult to interpret