What Changed at Aidoc

Mercy and Aidoc released a white paper, commissioned by AVIA, reporting on five months of AI deployment across the Mercy health system. The review documents more than 50,000 new clinical findings surfaced in that window and tracks the associated changes in speed of diagnosis and time to treatment. The scale is the notable part: this is a multi hospital deployment reporting aggregate output rather than a single site pilot.

Our read

A five month figure from a named health system gives buyers something to model against, which vendor accuracy claims do not. The number to interrogate is what counts as a new clinical finding, since that definition determines whether 50,000 represents meaningful catches or a high volume of flags a radiologist would have reached anyway. AVIA commissioning the work makes it more independent than a vendor case study and less independent than peer review, which is roughly where its evidentiary weight should sit.

The FDA cleared Aidoc's comprehensive abdomen CT triage solution, powered by its self developed CARE foundation model, bringing 11 newly cleared indications together with three previously cleared into a single 14 indication workflow. The company states this is the first FDA clearance of a double digit set of acute indications powered by one foundation model. In the FDA reviewed pivotal study the new indications reported a mean sensitivity of 97 percent and mean specificity of 98 percent, and the company reports roughly an order of magnitude reduction in false alerts compared with leading single condition tools.

Our read

The operationally decisive number here is not the sensitivity, it is the false alert rate. Triage AI fails in practice when radiologists learn to ignore it, and single condition tools deployed side by side compound that problem because each one alerts independently. A reported order of magnitude reduction in false alerts, if it holds in a buyer's own case mix, is what determines whether the tool survives past month three. Two things to establish in evaluation: that the cleared indications match the acute findings actually driving delays in your emergency department, and how the platform governs models it did not build, since aiOS also hosts third party AI and the governance layer is doing work across all of them.