OnCo
ideasIdea

Mandatory post-market performance reporting for cancer AI

Once an AI tool is in use, its maker and the hospital would have to report regularly how it is actually performing on real patients, and the reports would be public.

Post-market surveillance of medical devices focuses on adverse events, not performance. For AI, the relevant harm is silent degradation. The proposal requires deployed cancer AI to report standardised performance metrics (sensitivity, specificity, calibration, subgroup results, override rates) per site quarterly to the model registry, with thresholds that trigger regulator review, analogous to pharmacovigilance periodic safety update reports.

Hypothesis
Mandatory performance reporting will detect clinically significant performance degradation in a meaningful share of deployed models within their first two years, which would otherwise have gone unnoticed.
Rationale
Studies of deployed models (for example sepsis prediction tools) found real-world performance far below published claims, discovered only by independent academic audits; systematic reporting would make this routine.
What would test it
Require quarterly reports for all AI in one national screening programme for two years; count detected degradations and actions taken.
Maturity
speculative
Who has to act
regulator
Cost to try
Medium ($1M to $50M)
Years to first evidence
3
Bottlenecks it attacks

Connected

5top

Pages like this

not linked directly; found by shared links

fronts

1
AI & Computation

ideas

3
A public registry of every AI model used in cancer careA standard for monitoring AI performance drift with pause thresholdsEvery AI output logged in the record with input hash, version and clinician response

bottlenecks

1
AI that is built but not validated or deployed