OnCo
ideasIdea

A public registry of every AI model used in cancer care

Like a trial registry, every AI tool used on real patients would be listed publicly with what it is for, what data it was trained on, how well it performed and which version is running where.

There is no public record of which AI models are deployed in which hospitals, for what indications, at what versions. The proposal is a mandatory registry (regulator-run or accredited) with a standard model card: intended use, training data summary (sites, years, demographics), validation results by subgroup, version history, deployment sites, and links to post-market performance reports. FDA's list of cleared AI devices is a partial precedent but lacks deployment and performance data.

Hypothesis
A public registry will make independent scrutiny possible, reveal the share of deployed models without external validation, and correlate with faster withdrawal of underperforming models.
Rationale
Trial registration transformed accountability in clinical research; AI in care is at the stage trials were before registration, with selective reporting and unknown deployment.
What would test it
Pilot voluntary registration with 20 hospitals and their vendors for one year; measure completeness and the number of models found to lack external validation; then assess mandatory adoption.
Maturity
speculative
Who has to act
regulator
Cost to try
Small (under $1M)
Years to first evidence
2
Bottlenecks it attacks

Connected

6top

Pages like this

not linked directly; found by shared links

fronts

1
AI & Computation

ideas

4
A liability framework for clinical AI: safe harbour for clinicians, liability for makersMandatory post-market performance reporting for cancer AIPatients told which AI is used in their care, in plain languageRules for retiring cancer AI when performance drops or the standard of care moves

bottlenecks

1
AI that is built but not validated or deployed