What can keywords be used for, concretely? The medical director gives us the example of the patient's weight: “We enter the weight each time, in pediatrics, in dietetics, etc. With a file oriented towards data recovery we can obtain a patient's weight curve and see the transformations.“Artificial intelligence is capable of making connections within a medical report.
How to block access to your medical records on the Walloon health network?
“Take a doctor who makes a report saying 'the patient presents because he has had an erythema', which means a skin reaction, probably allergic. The AI reads ten lines later 'a peanut allergy was confirmed by a blood test… All these links are automatic.” This also makes it possible to eliminate false allergies, “like that to penicillin, which is rarely objectified, because the system makes it possible to treat a suspected allergy differently from an allergy confirmed by a blood test and prick tests.”
Hospital data encoded by human hand, soon to be finished
Currently, at the CHC of Liège, around fifteen coders read the omnipro files and code the pathologies with the ICD10 code (a WHO classification: International classification of diseases). This is necessary for the “case mix”, the analysis of the activity of health establishments. “Hospitex funding is linked to the severity of the damage and the presence of comorbidities, because they increase the duration of the hospital stay,” explains Dr Philippe Olivier. “If we under-code, we lose part of the funding, because the duration of the stay is considered unjustified.”
Human coders therefore read the file and identify the presence of diabetes, hepatitis B, phlebitis or gallbladder surgery.
“For a long time, we were two years late when we talked about the minimum clinical summary (MHR). So the encoded data was already out of date, it no longer had any use“
-Today, to optimize this information in free text, Dr Olivier is working with the company Effixis, a company specializing in what we call LLMs (large language models, which are the basis of GPTs, pre-learned generative transformers). . “For a long time, we were two years late,” notes Dr Olivier.
The almost immediacy that artificial intelligence allows opens up the field of possibilities, particularly for colon cancer screening. But above all, AI goes further than humans. “The purpose of the coder is to justify the stay financially to the Ministry of Health. She enters a certain number of necessary codes into the system, relative to the length of stay, and then stops. The AI adds codes that have no impact on length of stay, but may be interesting. For example, on colonoscopy, we saw that artificial intelligence encoded ten codes on average, while humans only chose three.”
Radiology triage is already implemented
CHC Montlégia is not starting from a blank sheet, AI is already in use at the moment: “We have several software programs that analyze images and offer diagnostics. It is up to the radiologists to validate or refute this initial diagnosis. explains Philippe Olivier.
“In imaging, artificial intelligence will streamline queues in emergency rooms, by creating two piles of files: fractures and the absence of fracture. This will speed up patient care. The current slowness comes because it is necessary to look at all the x-rays, including the normal ones, to take care of the abnormal ones.”
This sorting, during rush periods or at night, is always “rechecked” by the man afterwards.
