Analysis Component: Population-level Summary (Estimator)

[ASL-rmarshall]

There may be multiple population-level summaries associated with an estimand:

• A single "main estimator" to calculate the estimate for the estimand.
• As many "sensitivity estimators" as needed to test the validity of the assumptions defined for the estimand.
• Any additional "supplementary estimators" that are deemed necessary to more fully investigate and understand the trial data.

While "main" and "sensitivity" estimators will have a tight association with the estimand for which they are defined, a single "supplementary" estimator may have relevance for multiple estimands. For example, a "summary of <baseline characteristic> by treatment" supplementary estimator may have relevance for any estimand that references the specified baseline characteristic.

Each estimator will be defined as a combination of properties :

• Population-defining properties:
  • Reference to a defined subject population (analysis set)
  • Additional subsetting criteria
• "Classification variables"/dimensions/grouping factors - variables used to subdivide the population for analysis. This will usually include treatment (unless treatment is handled separately)
• "Statistical unit" - what is being summarised (e.g. subjects vs events)
• Method - a specification of which statistical calculation will be used for the summary
• Model (if needed) - though this might be represented as a "role" played by each of the classification variables.
• Measure(s) - a specification of the output value(s) obtained through performance of the method.

Each of these properties may need to be defined explicitly or may be inherited from the associated estimand.