Getting a medical AI product CE certified is hard. Keeping it certified, with every update and in every hospital, is harder. And certification is only one part of the work that happens out of sight.
Every AI product starts with an idea. Hospitals already collect enormous amounts of data, and somewhere in that data there are patterns that can help predict what will happen to a patient. When a model shows it can find those patterns and make reliable predictions, that is a big achievement.
It is also just the tip of the iceberg. Below the surface sits everything needed to get it safely into a hospital and keep it there. And for a hospital, the tip is all it should see. If the product feels simple, the work underneath has been done well.
In this article, I will take you below the surface, to the work the hospital never has to deal with.
Making it work outside the lab
Every hospital stores its data in its own way. The same lab result can sit in a different place in two hospitals; a procedure code can mean something different from one hospital to the next, or exist in only one of them. To deal with this, we standardize all incoming data, build processing logic that can handle these differences, and check everything that comes in before our model uses it. On top of that, we recalibrate our models for each hospital, within our certified change procedures, to account for differences in patient population, procedure types and ways of working.
Keeping it safe, every single time
Every recalibration is also a change to a medical device, and so is every update we release. Each one has to be safe, in every hospital. That means having documented procedures, following them consistently and proving that to an auditor several times a year. We keep refining those procedures so we can roll out safe updates quickly, whether that’s a recalibration for a new hospital or an improvement for all of them. For a hospital, this comes down to one thing: a product that stays compliant without the hospital having to manage it.
Getting it to the bedside
A safe, compliant product is only valuable if it’s actually used. And a prediction only matters if it reaches the clinician or nurse at the right moment. That’s why PERISCOPE® is built into the EHR (such as HiX), the system clinicians already work in every day. They don’t have to open a separate application.
It does ask for a small change in routine: using one new piece of information when making clinical decisions. It also requires hospital IT, application administrators and the EHR vendor to work together. We have done this before, together with EHR vendors, and we bring that experience to every new hospital. Their team doesn’t have to figure it out from scratch.
Why it’s worth it
After the data work, the safety procedures and the integration, the day comes when a surgical ward opens the first dashboards in their EHR. For each patient, the dashboard shows the risk of a postoperative infection. Nurses and doctors can now see earlier which patients need extra attention, based on data that was there all along.
That is when everything below the surface makes sense.
To them, it looks simple: one new piece of information, in a system they already know. That is exactly the point. The certificate, the model and the dashboard are the tip of the iceberg. Everything underneath is our work, so the hospital only ever has to see the top.
Written by:
CTO & Co-founder
