estimandr

ICH E9(R1) Estimand Framework for R

Every clinical trial submitted to FDA or EMA now requires a formally documented estimand. In practice, statisticians write these in Word documents — disconnected from the analysis code, impossible to validate, and guaranteed to drift from what is actually implemented.

estimandr closes that gap. It represents every component of an ICH E9(R1) estimand as a structured, computable R object: population, variable, intercurrent events (with all five strategies as first-class types), population-level summary, and analysis method. It validates that your analysis is consistent with your stated estimand, generates SAP paragraph text automatically, and produces regulatory-ready submission tables and reports.

Installation

remotes::install_github("repro-stats/estimandr")

Quick start

library(estimandr)

# 1. Define the five estimand components
pop <- est_population(
  description = "Adults aged >= 18 with moderate-to-severe atopic dermatitis",
  criteria    = list(
    inclusion = c("Age >= 18 years", "Baseline EASI >= 16", "Randomised"),
    exclusion = c("Prior biologic therapy within 12 weeks")
  ),
  flag_var = "ITTFL"
)

var <- est_variable(
  name         = "Change from baseline in EASI score at Week 16",
  type         = "continuous",
  timepoint    = "Week 16",
  adam_dataset = "ADEFF",
  adam_var     = "CHG",
  direction    = "decrease",
  scale        = list(name = "EASI", range = c(0, 72), mcid = 6.6)
)

disc <- define_ice(
  name     = "Treatment discontinuation",
  code     = "DISC",
  strategy = ice_hypothetical(
    assumption = "Patients would have continued to respond had they remained on treatment",
    estimation = "MMRM using all available pre-discontinuation data under MAR"
  )
)

sm <- est_summary_measure(
  type        = "difference",
  description = "Difference in LS mean change from baseline (Active - Placebo)"
)

an <- est_analysis(
  method      = "MMRM",
  description = "MMRM with treatment, visit, treatment x visit, and baseline",
  formula     = "CHG ~ TRT01P * AVISIT + BASE + BASE:AVISIT",
  software    = "mmrm::mmrm()"
)

# 2. Build the estimand
primary <- define_estimand(
  id              = "primary",
  label           = "Primary efficacy estimand",
  population      = pop,
  variable        = var,
  ices            = list(discontinuation = disc),
  summary_measure = sm,
  analysis        = an
)

# 3. Add a sensitivity estimand
sensitivity <- define_estimand(
  id    = "sensitivity_tp",
  label = "Treatment policy sensitivity estimand",
  role  = "sensitivity",
  population      = pop,
  variable        = var,
  ices            = list(discontinuation = define_ice(
    "Treatment discontinuation",
    strategy = ice_treatment_policy("Real-world effect")
  )),
  summary_measure = sm,
  analysis        = an
)

tree <- estimand_tree(primary = primary, sensitivity = list(sensitivity))

# 4. Validate consistency
check_consistency(primary)

# 5. Generate outputs
cat(sap_text(primary))             # SAP paragraph
estimand_table(tree)               # Regulatory submission table
estimand_report(tree,              # Full HTML report
  output   = "outputs/estimand_spec.html",
  study_id = "TRIAL-001",
  sponsor  = "Example Pharma Ltd"
)

Key functions

Function	Purpose
define_estimand()	Define a complete estimand
est_population()	Specify the target population
est_variable()	Specify the outcome variable
define_ice()	Define an intercurrent event
ice_treatment_policy()	Treatment policy ICE strategy
ice_hypothetical()	Hypothetical ICE strategy
ice_composite()	Composite ICE strategy
ice_while_on_treatment()	While-on-treatment ICE strategy
ice_principal_stratum()	Principal stratum ICE strategy
est_summary_measure()	Population-level summary measure
est_analysis()	Statistical analysis method
estimand_tree()	Primary/secondary/sensitivity hierarchy
check_consistency()	Validate estimand-analysis alignment
sap_text()	Generate SAP paragraph text
estimand_table()	Regulatory submission table
estimand_report()	Full HTML estimand report
estimand_template()	Built-in templates for common scenarios

Templates

Pre-built templates for six common trial scenarios:

estimand_templates()        # list all available templates
estimand_template("superiority_continuous")
estimand_template("non_inferiority_binary")
estimand_template("time_to_event")
estimand_template("oncology_os")
estimand_template("rwe_effectiveness")

Integration with bayprior

estimandr is designed to pair with bayprior for Bayesian estimand analyses. Define your estimand first, then link it to a bayprior elicitation for the primary analysis:

library(bayprior)
# elicit prior, then reference it alongside your estimand

Regulatory background

• ICH E9(R1) Addendum on Estimands and Sensitivity Analysis (November 2019)
• FDA Guidance: E9(R1) Statistical Principles for Clinical Trials (May 2021)
• EMA Reflection Paper on Estimands (March 2020)

License

MIT © ReproStats