Table 1 Attributes of estimands and graphical display of statistical methods
Objective | Attributes | Graphical display | |||
|---|---|---|---|---|---|
Population | Variables (endpoints) | Strategies for dealing with death | Effect measures | ||
Time-to-deterioration (TTD) analysis (method for evaluating composite variables) | |||||
• Evaluate time to deterioration (death or deterioration) | • All participants | • Time to first event (death or deterioration) | • Composite endpoint • Death and deterioration are equally treated | • Event probability at a specified time • Hazard ratio • Median (or mean) time to event | • Describe the proportion of deterioration by survival curves |
Prioritized composite outcome approach (method for evaluating composite variables) | |||||
• Evaluate “win” by multiple aspects (multiple endpoints) | • All participants | • Composite endpoint by death and deterioration • “Win” defined by generalized pairwise comparisons | • Composite endpoint • Requires eliciting expert opinion on ordering death and QOL | • Win ratio [27] • Net benefit [15] | • Describe the combination of two step charts of cumulative probability of death and that of QOL deterioration |
Semi-competing risk analysis (method for evaluating composite variables) | |||||
• Evaluate the cumulative probability of QOL deterioration having occurred in the presence of death | • All participants | • Time to deterioration | • Competing risks | • Sub-distributional hazard ratio • Cause specific hazard ratio | • Describe the cumulative probability of QOL deterioration (cumulative incidence function) |
Linear mixed model for repeated measures (MMRM) (method for evaluating QOL itself) | |||||
• Evaluate the magnitude of worsening QOL | • All participants | • QOL at the time of primary interest • QOL at every visit | • Assuming missing at random for death • Implicitly impute data beyond death | • Mean difference in QOL at the time of primary interest • Difference in slopes of QOL trajectories over time | • Describe a trajectory of mean QOL over time using a line chart with a measure of uncertainty |
Principal stratification for survivor average causal effect (SACE) (method for evaluating QOL itself) | |||||
• Evaluate the magnitude of worsening QOL | • Participants who would not die regardless of which treatment they received (“always survivors”) | • QOL at the time of primary interest • QOL at every visit | • Death in the target population does not occur • The target population is not directly identifiable | • Mean difference in QOL at the time of primary interest • Difference in slopes of QOL trajectories over time | • Describe a trajectory of QOL over time by line chart with a measure of uncertainty |