Skip to content

Validation

pyrbmi maintains strict numerical parity with R's rbmi package, the reference implementation for regulatory submissions.

Validation Strategy

Our validation approach follows FDA/EMA guidance on software validation for clinical trials:

  1. Unit Tests: All functions have unit tests with ≥90% coverage
  2. R Parity Tests: Numerical output compared against R rbmi on identical datasets
  3. Reference Datasets: Tests use publicly available or synthetic datasets with known properties
  4. Tolerance Criteria: Statistical results match within machine precision (1e-10 relative)

R Parity Testing

The R parity test suite runs weekly against the latest R rbmi release:

# Run R parity tests locally (requires R + rpy2)
uv run pytest tests/test_vs_r/ -v

Test Coverage

Component R Test Coverage
MAR imputation ✅
J2R imputation ✅
CR imputation ✅
CIN imputation ✅
LMCF imputation ✅
Rubin's rules pooling ✅
MMRM covariance structures ✅

Automated Validation

The validation workflow runs:

  • Weekly: Against latest R rbmi from CRAN
  • On PR: When R-related code changes
  • Results: Published as workflow artifacts

See R Parity Report for the latest validation results.

Validation Report

The full validation report (v0.7.0 milestone) will include:

  • Numerical comparison tables
  • Tolerance analysis
  • Edge case documentation
  • Regulatory submission notes

Last updated: [Placeholder for automated report]