A group of researchers has created the first wearable camera system that utilizes artificial intelligence to identify possible medication delivery errors. During testing, this video system successfully recognized and distinguished the medications being administered in busy clinical environments. The AI achieved an impressive accuracy rate, with 99.6% sensitivity and 98.8% specificity for detecting vial-swap errors. This innovative system could serve as an essential safety measure, particularly in operating rooms, intensive care units, and emergency medical situations.
A group of researchers has created the first wearable camera system that utilizes artificial intelligence to identify possible medication delivery errors.
In tests published today, the video system showed exceptional capability in recognizing and identifying medications being drawn in dynamic clinical settings. The artificial intelligence reached a sensitivity of 99.6% and a specificity of 98.8% in identifying vial-swap errors.
The findings have been released on October 22 in npj Digital Medicine.
This system could become a vital safety feature, especially in operating rooms, ICUs, and emergency care environments, according to Dr. Kelly Michaelsen, a co-lead author and assistant professor of anesthesiology and pain medicine at the University of Washington School of Medicine.
“The ability to assist patients in real-time or avert a medication mistake before it occurs is immensely powerful,” she stated. “While we strive for perfect accuracy, even humans struggle with that. In a survey of over 100 anesthesia providers, most expressed a preference for a system that exceeds 95% accuracy, which we have accomplished.”
Errors in drug administration are the most commonly reported serious incidents in anesthesia and are a leading cause of significant medical errors in intensive care units. On a broader scale, it is estimated that 5% to 10% of all administered drugs are linked to errors. Adverse events related to injectable medications may impact around 1.2 million patients each year, incurring costs of approximately $5.1 billion.
Syringe and vial-swap errors frequently occur during IV injections when a clinician transfers medication from a vial to a syringe for patient delivery. About 20% of these mistakes involve substituting the wrong vial or mislabeling a syringe. Another 20% arise when a correctly labeled drug is inadvertently administered incorrectly.
To mitigate these risks, safety measures like barcode systems are implemented to rapidly verify a vial’s contents. However, in high-pressure situations, practitioners might forget this verification step, as it adds an extra task to their workflow.
The researchers aimed to design a deep-learning model that, when paired with a GoPro camera, could effectively detect the contents of vials and syringes, providing timely warnings before medication is administered to the patient.
The training of the model took several months. The team recorded 4K video of 418 medication draws by 13 anesthesia providers in operating rooms with varying setups and lighting conditions. The footage captured clinicians handling vials and syringes of selected medications, which were later cataloged and verified to educate the model in recognizing the specific contents and containers.
While the system doesn’t read the text on each vial, it identifies other visual indicators such as the size and shape of vials and syringes, the color of vial caps, and the size of label prints.
“It was particularly challenging because in the operating room, the clinician is holding a syringe and a vial, making it difficult to see either object fully. Often, some letters on the syringe and vial are obstructed by their hands, which move quickly as they focus on their tasks rather than posing for the camera,” explained Shyam Gollakota, a paper coauthor and professor at the UW’s Paul G. Allen School of Computer Science & Engineering.
Additionally, the computational model was trained to prioritize medications in the foreground and disregard vials and syringes located in the background.
“The AI is designed to detect the specific syringe that the healthcare provider is handling while ignoring other syringes lying on the table,” Gollakota mentioned.
This research illustrates that AI and deep learning hold significant potential to enhance safety and efficiency across various healthcare practices. Michaelsen noted that the exploration of this potential is only just beginning.
The study also involved contributions from researchers at Carnegie Mellon University and Makerere University in Uganda. The Toyota Research Institute was responsible for building and testing the system.
The research was funded by the Washington Research Foundation, the Foundation for Anesthesia Education and Research, and a grant from the National Institutes of Health (K08GM153069).
