Originally published by our sister publication Anesthesiology News
NEW ORLEANS—Artificial intelligence and machine learning are likely to soon become essential tools in the anesthesiologist’s armamentarium, experts said in a session at the 2022 annual meeting of the American Society of Anesthesiologists.
Combined with machine-assisted modeling, AI can improve the assessment of the airway and identification of patients at risk for a difficult intubation, said Clyde Matava, MD, an assistant professor of anesthesia and associate chief of the perioperative service at the Hospital for Sick Children, in Toronto. “At present, clinicians will assess the patient based on physical factors, bedside screening tests and history, and make a call as to the risk of a difficult airway. Unfortunately, the current assessment tools that we have perform well for extremes, but not as well for patients in the middle ground. We also have good evidence that there isn’t good interrater concordance when different clinicians rate the same patient.”
Analyses of claims related to airway management have found that intubation difficulties or ventilation disorders are important causes of serious injury, brain damage or death (Best Pract Res Clin Anaesthesiol 2011;25:263-276).
Machine vision can be used to provide additional data sets and information on features that the human eye cannot differentiate, Matava said. “Using these tools, we can develop algorithms that can identify nuanced features from a patient’s face, using photographs, video images and 3D scans, to assess whether that patient will have a difficult airway. As a result of COVID, we have increased the number of patients for whom we perform a preoperative evaluation via a video connection while they are still at home, so it would be noninvasive and relatively inexpensive to capture images for intubation assessment at the same time.”
This work is still in the investigational phase, Matava noted. “The Food and Drug Administration reviews all of this technology within the lens of risk. To what extent is the algorithm ‘prescribing’ to the physician what to do? The more autonomy that is ceded to the algorithm, the more risk there is; thus it needs to be approved as a medical device and we’re not quite there yet. Nonetheless, early results show that machine learning techniques are fairly advanced and can help the anesthesiologist with quicker access to the airway.”
AI Methods for Obstetric Anesthesia
Matava also discussed another source of guidance for the anesthesiologist that may help “democratize” regional anesthesia: ultrasound images. “Regional anesthesia is very popular and shortens hospital length of stay while minimizing side effects such as vomiting, but its use is limited in part by expertise and frequency. Those who do not perform it as often may be reluctant to use it as an option,” he said. “In these situations, ultrasound can be an excellent ‘co-pilot’ to help the anesthesiologist obtain central line access, extract features and label the nerves that should be targeted. It is still up to the physician to determine if this is the right nerve, but ultrasound-enabled devices can help label the key features.”
Radiologists already have commercially available algorithms that they are deploying in CT and MRI, he noted. “Anesthesia is just a bit behind, but we will have these tools in our armamentarium. Will they take our jobs? Of course not. We still must manage the anesthesia, but the threshold to be successful will be lowered when we have these tools available at any time to help guide the clinician with challenges like regional anesthesia or airway management.”
In obstetric anesthesia, AI methods in closed-loop systems offer a promising option for automating drug infusions, said Vesela Kovacheva, MD, PhD, an attending anesthesiologist at Brigham and Women’s Hospital, an assistant professor of anesthesia at Harvard Medical School, and the director of translational and clinical research for Harvard’s division of obstetric anesthesia, both in Boston.
Closed-loop systems have existed in the research realm of anesthesiology since the 1950s, and research has demonstrated that they can improve the stability of controlled variables and reduce the workload in clinical practice, without increasing the risks to patients. A 2017 systematic review and meta-analysis found that compared with manual control, closed-loop systems remain within the desired range longer, are associated with less under- and overtreatment, and yield lower drug doses and faster recovery (Anesth Analg 2017;124[2]:446-455).
“They allow us to use less medication to achieve the same goals, with the hope that patients can be discharged earlier, feel better and experience fewer drug errors,” Kovacheva said. “To date, however, we have not been able to implement these automated closed-loop infusions in clinical practice in the U.S., due in part to regulatory issues.”
Target-controlled infusion failed to obtain regulatory approval in the United States in the 1990s and early 2000s. In 2013, the Sedasys computer-assisted personalized sedation system (CAPS) was approved by the FDA for use in administration of light sedation during gastrointestinal endoscopic procedures, but the manufacturer Ethicon withdrew it from the market three years later (Anesth Analg 2017;124[2]:686-688).
“Since then, there has been no other company willing to approach the FDA to authorize these devices and approaches,” Kovacheva said. “But several technologies now exist in the research realm that hold promise for implementation in the future for optimal hemodynamic control in obstetric anesthesia, providing excellent control of blood pressure with fewer side effects. A few studies have shown better performance, and less hypotension and nausea, when using closed-loop systems compared to manual control during cesarean delivery (Anaesthesia 2014;69:37-45; BJA 2013;110:59-65).
Closed-loop systems have several advantages over the automated open-loop sedation systems that are widely used outside this country, Kovacheva said. “With open-loop systems, we can select the desired level of sedation we want to achieve, and it is automatically calculated based on a preexisting pharmacokinetic model. While that allows for more precise drug administration and lower chance of errors, the pharmacokinetic model works best in the population in which it was developed. But in a more diverse population, it’s less accurate. Closed-loop systems, on the other hand, monitor the patient’s blood pressure and keep it within a specified range, with feedback given to the machine every cycle to improve its performance. This technology holds a lot of promise for us to provide more personalized and optimized care for our patients.”
By Gina Shaw
Kovacheva reported consulting fees from Avania CRO and a patent for control of a therapeutic delivery system.