Frozen model and validation population

Define frozen model and applicability

A frozen model includes coefficients, intercept or baseline survival, preprocessing and input definitions. Independence from development data must be explicit. Plan precision for several performance domains rather than relying on a minimum event count alone.

Flow diagram

Frozen model and validation populationF1.S1Identify the exact modelF1.S2Define independent validationsettingF1.S3Compare population and case mixF1.S4Specify validation sample sizeF1.S5Freeze untouched evaluation planF1.N1External validation specificationElementOriginal modelValidationdefinitionDifferenceModel versionIntended usePopulationPrediction horizonCase mixF1.N2Assumptions and rationaleF1.N3Method and protocol evidenceF1.S0 Start flowFrozen model and validationpopulationF1.S999 End flowEnd flow

Steps

Identify the exact model

Obtain the exact frozen model, preprocessing, coefficients, intercept or baseline survival and required input definitions. Record model version and any unavailable components.

Define independent validation setting

Document geography, time, sites and independence from development.

Compare population and case mix

Record eligibility and differences from the original target population.

Specify validation sample size

Plan precision for calibration, discrimination and key subgroups.

Freeze untouched evaluation plan

Keep model updating separate from the original-model validation.

Frozen model and validation population

End flow

Tables and supporting information

External validation specification

External validation specificationElementOriginal modelValidationdefinitionDifferenceModel versionIntended usePopulationPrediction horizonCase mix

Record the frozen model before examining validation performance.

Assumptions and rationale

External validation requires independence appropriate to the claim; a random split of development data is not external validation.

Method and protocol evidence

Attach the protocol, methodological sources and versioned software specification supporting these decisions. Record citations and provenance; do not enter participant identifiers.

Sources