A team of researchers has presented a novel wearable camera system able to identify possible mistakes with regard to medication delivery together with the assistance of AI.
More to the point, in a trial whose findings were reported on Oct. 22 in npj Digital Medicine, the video system accurately singled out which drugs were being administered, in real wards, and with impressively low error rates.Error detection included vial-swap errors; AI yielded 99.6% sensitivity and 98.8% specificity on this aspect.
The system could become a crucial preventer in operating rooms, ICU, and emergency medicine doctor’s practice, according to co-lead author Dr. Kelly Michaelsen, an assistant professor of anesthesiology and pain medicine at the University of Washington School of Medicine.
You may expect a one hundred percent performance from an employee but even humans cannot do that. The web-based survey administered to more than 100 anaesthesia providers revealed that they wanted the system to be ‘‘more than 95% accurate,’ Each of these goals was met.”
Medication administration mistakes are the commonest ‘sentinel’ events in anaesthesia and the leading cause of significant adverse occurrences in the critical care setting. On the broader scale, between 5% and 10% of drugs administered are administered with errors.
Avoidable injection related complications are attributed to result in 1.2 million patient incidents per year at a cost of $5.1 billion. Patient admixture errors are most common in situations that require drawing a medication from the vial into a syringe for an intravenous injection.
About a fifth of the cases that contain an error fall under the substitution class where an incorrect vial is selected or where a syringe is labelled incorrectly. Another 20% comes up when the drug is correctly labelled but is given in wrong dosing.
Loose safety measures that would prevent such incidents include, a bar code scanning method that ensures validation of the contents of a vial in the shortest time possible.
However, it means that practitioners might sometimes neglect this check during high stressful situations as it is an added check on practitioners.
The participating researchers’ idea was a deep-learning model with the GoPro camera that is complex enough to identify the content of the cylindrical vial and syringes and provide an alarm before the medicine reaches the patient.
Even getting a model trained took several months given the many classes involved. The investigators collected 418 drug draw videos in total from 13 anesthesiology providers at 11 operating rooms with various setups and lighting.
It also compiled in the video made included clinicians handling vials and syringes of certain medicines. These video snippets were later logged and the content of the syringes and vials noted to teach the model as to what to look for in the syringes or vials contained.
This work proves that AI and deep learning can contribute extremely important to safety and efficiency upgrades in numerous health care practices. Michaelsen confirmed that scientists have not thought of it.
Reference: Chan J, Nsumba S, Wortsman M, Dave A, Schmidt L, Shyamnath Gollakota, et al. Detecting clinical medication errors with AI enabled wearable cameras.
