Fit-for-Purpose Reference Standards

GMLP Principle 5. Choose reference standards your validation actually depends on: consensus-informed, expert-driven, and tied to intended use environment.

Overview

Reference standards determine what 'correct' means for your model. A reference standard chosen casually undermines every downstream performance claim. A poorly suited reference standard produces validation that does not predict deployed performance.

We design reference standard selection grounded in consensus methods, appropriate expertise, and a rationale that explicitly connects to intended use.

Our Process

  1. 1

    Intended use connection

    What ground truth makes sense for the deployment context?

  2. 2

    Consensus-method selection

    Inter-rater protocols, adjudication rules.

  3. 3

    Expert qualification

    Reader credentials, experience requirements.

  4. 4

    Rationale documentation

    Defensible written rationale for the chosen standard.

  5. 5

    Sensitivity analysis

    Performance under alternative reference standards.

Frequently Asked Questions

Single expert reads enough?

Almost never. Consensus or adjudicated reads are standard.

What about pathology gold standards?

Used where appropriate; not always feasible. We surface the trade-offs.

Can reference standards evolve?

Yes, with documented change control.

How does this interact with regulatory submission?

Reference standard rationale is directly cited in your submission.

Make reference standards defensible by design.

Send us your current validation setup. We will return a reference standard assessment within four weeks.

Start a Conversation