Skip to contents

Specify an ICH E9(R1) estimand and apply an intercurrent-event strategy

Source: R/estimand.R
makeEstimand.Rd

makeEstimand() records the analysis target using the five attributes of the ICH E9(R1) estimand framework (treatment condition, population, variable / endpoint, intercurrent-event handling, and population-level summary). It does not change the estimation; it documents the target for a statistical analysis plan and travels with the results for reproducibility and regulatory clarity.

applyIntercurrentEvent() implements the data handling for the intercurrent-event (ICE) strategy. An ICE is coded as one of the event-type values in the data (the observed time is the earliest of the outcome, competing, censoring, and ICE times). Supported strategies:

"treatment policy"

The ICE is ignored and follow-up continues to the outcome – the default intent-to-treat target. The data are returned unchanged (the ICE code, if any, is treated as a competing/censoring event exactly as supplied).

"hypothetical"

The ICE is recoded as censoring (event type 0). concrete's inverse-probability-of-censoring weighting then targets the risk that would be observed if the ICE had not occurred, under the assumption that the ICE acts as conditionally independent censoring given the measured covariates – a strong, untestable assumption that should be accompanied by a censoring sensitivity analysis.

"composite"

The ICE is recoded as an occurrence of the event of interest TargetEvent, defining a composite endpoint.

The "while on treatment" and "principal stratum" strategies are not supported: the former is a different (descriptive, non-extrapolated) estimand that an IPCW estimator does not target, and the latter requires latent stratification beyond the current scope.

Usage

makeEstimand(
  treatment = "as randomized (intent-to-treat)",
  population = "as enrolled",
  endpoint = "cause-specific absolute risk",
  summary = "risk difference",
  strategy = c("treatment policy", "hypothetical", "composite"),
  intercurrent = NULL
)

# S3 method for class 'ConcreteEstimand'
print(x, ...)

applyIntercurrentEvent(
  Data,
  EventTime,
  EventType,
  Intercurrent,
  strategy = c("treatment policy", "hypothetical", "composite"),
  TargetEvent = 1,
  intercurrent = "intercurrent event",
  Verbose = TRUE
)

Arguments

treatment

character: the treatment condition / intervention contrast.

population

character: the target population.

endpoint

character: the variable / endpoint (e.g. cause-specific absolute risk, RMST) and its horizon.

summary

character: the population-level summary (e.g. risk difference, risk ratio, RMST difference).

strategy

one of "treatment policy", "hypothetical", "composite".

intercurrent

character or NULL: a label for the intercurrent event.

x

a "ConcreteEstimand" object.

...

ignored.

Data

a data.table/data.frame of one row per subject.

EventTime

character: name of the observed-time column.

EventType

character: name of the event-type column (0 = censoring, positive integers = events).

Intercurrent

numeric: the event-type value in EventType that codes the intercurrent event.

TargetEvent

numeric: for the "composite" strategy, the event-of- interest code that the intercurrent event is merged into.

Verbose

logical: report what was recoded.

Value

makeEstimand() returns a "ConcreteEstimand" object.

applyIntercurrentEvent() returns a copy of Data with EventType recoded for the chosen strategy, carrying the estimand in attr(., "Estimand").