A recent study led by researchers at the University of Washington School of Medicine has shown that a novel wearable camera system leveraging artificial intelligence (AI) can detect medication errors during drug administration. The system, designed to be used in environments such as operating rooms and intensive care units, demonstrated a high level of accuracy, achieving 99.6% sensitivity and 98.8% specificity in identifying vial-swap errors.
The findings, published in npj Digital Medicine, shows that such a system could become an important safeguard in busy clinical settings to improve patient safety.
Previous studies have shown that roughly one in every twenty patients experiences preventable harm in clinical settings. Drug-related errors are a leading cause of these errors with as many as 12% of these errors leading to serious harm to the patient or death. In the U.S., studies estimate that medical errors contribute to between 140,000 and 440,000 deaths annually, with 80% occurring in hospitals and 41% operating rooms.
“The thought of being able to help patients in real time or to prevent a medication error before it happens is very powerful,” said Kelly Michaelsen, MD, PhD, co-lead author and assistant professor of anesthesiology and pain medicine at the University of Washington School of Medicine. “One can hope for a 100% performance but even humans cannot achieve that.”
Michaelsen also noted that a survey of more than 100 anesthesia providers revealed the majority expressed a desire for systems to exceed 95% accuracy—a target that this new technology has achieved.
Medication administration errors, particularly in anesthesia, are a common concern, with syringe and vial-swap errors representing significant risks. These errors often arise during intravenous injections when clinicians transfer medications from vials to syringes. Approximately 20% of errors result from selecting the wrong vial or a syringe is mislabeled, while another 20% of errors occur when the drug is labeled correctly but administered incorrectly.
Despite existing safety measures designed to prevent errors, such as barcode scanning systems designed to confirm a vial’s contents, the stress of some clinical settings can still result in drug administration errors. The solution developed at the University of Washington to address this care gap was to develop a deep-learning model integrated with a GoPro camera. This model can identify the contents of cylindrical vials and syringes and issue warnings before medications are administered to patients.
Training of the AI model took the team months. To accomplish this, the investigators collected 4K video footage of 418 drug draws performed by 13 anesthesiologists in a variety of operating room situations. This dataset allowed the model to learn to recognize visual cues, such as the size and shape of vials and syringes, as well as vial cap colors.
“It was particularly challenging because the person in the OR is holding a syringe and a vial, and you don’t see either of those objects completely. Some letters (on the syringe and vial) are covered by the hands. And the hands are moving fast. They are doing the job. They aren’t posing for the camera,” said study co-author Shyam Gollakota, PhD, a professor at the University of Washington’s Paul G. Allen School of Computer Science & Engineering.
In addition, the algorithm needed to be trained to only analyze medications in the foreground while ignoring other images that might be in the background such as vials and syringes. “AI is doing all that: detecting the specific syringe that the healthcare provider is picking up, and not detecting a syringe that is lying on the table,” Gollakota added.
The new technology is in its early stages and will be further developed by the University of Washington team, but the potential for an important new safety measure is apparent. These automated, AI-enabled medication checks in busy clinical settings could be employed as a second set of “eyes” to verify the accuracy of the work done by anesthesiologists and provide warnings of potential errors before they occur. The researchers also noted that anesthesiologists are already required to wear protective eyewear in the OR environment to protect them from fluids, so incorporating a camera into this protective equipment help foster adoption of the technology.
