**4. Intraoperative surgical decision-making**

Although AL/ML-based algorithms and approaches can greatly improve patient outcomes during preoperative use, perhaps the most promising and powerful use of these programs is their ability to improve intraoperative care. Algorithms trained on patient vital signs, various biometric and non-biometric characteristics, electrocardiography (EKG), and other data points could be utilized to help facilitate real-time reduction of various intraoperative risks, including those of hypertension, hypoxemia, massive hemorrhage, and other complications [42–44]. Loftus et al. write that this comprehensive analysis of patient parameters using AI is especially important for more complex disease states, such as frailty [45]. Though frailty is a multifactorial disease state affected by physical, cognitive, and social variables, frailty is currently diagnosed by a few physical, often subjective criteria. For instance, the Fried frailty phenotype assesses patients based on their recent physical activity, subjective feelings of exhaustion, walking speed, handgrip strength, and unintentional weight loss. Diagnosing frailty can therefore be inconsistent, even though frailty is known to increase morbidity, mortality, and risk of other comorbidities that also increase surgical risk. Through expert-led ML training on large sets, algorithms could be developed to better classify complex disease systems such as frailty or sepsis and improve intraoperative risk assessment [45]. These outputs could further allow for augmented decision-making, or the advanced application of highly sophisticated models that are trained on multiple iterations of the same surgical procedure type. This, in turn, could provide decision-making assistance for surgical teams performing same-type operation based on the patient's vital signs, procedural characteristics, the progression of the surgery, and various other potential characteristics [46]. For instance, if a machine learning model identifies that a certain constellation of parameters was associated with worse outcomes, it could potentially suggest that the surgical team addresses a specific aspect of patient care to improve the projected outcome, or perhaps to reduce various complication risks [4, 47–50]. Komorowski et al. showed the possibility of this type of AI through an algorithm that was able to suggest optimal treatment and dosing options for sepsis patients leading to lower patient mortality than human clinicians alone [51].

Surgery often places high demands on surgeons' cognition, creating an opportunity for ML/AI algorithms to reduce cognitive load and further identify ways to improve surgical outcomes [50, 52, 53].
