OnCo
ideasIdea

Version control and locked reference sets for AI algorithms used as companion diagnostics

AI is starting to decide which patients get which cancer drug. Every change to the software should be tested against a fixed public set of cases before it is used on patients.

AI-based scoring of HER2, PD-L1 and other markers is entering clinical use. Software updates can shift positivity rates silently. Regulators (FDA's predetermined change control plans, EU AI Act) are building frameworks. A concrete requirement for oncology companion diagnostic algorithms: every version must report performance on a locked public reference set, changes must be logged with effect on positivity rates, and laboratories must record the algorithm version in each patient report.

Hypothesis
Version reporting will reveal at least one clinically meaningful drift (positivity change above five percentage points) in a deployed algorithm within two years, which would otherwise have gone undetected.
Rationale
Software versioning is routine in engineering and absent in diagnostic pathology reporting; drift has been documented in deployed medical AI.
What would test it
Implement version logging and reference-set testing for two deployed pathology algorithms across ten laboratories; monitor positivity rates by version.
Maturity
early clinical
Who has to act
regulator
Cost to try
Small (under $1M)
Years to first evidence
2
Bottlenecks it attacks

Connected

9top

Pages like this

not linked directly; found by shared links