Development and Validation of a Machine Learning Model to Identify Patients Before Surgery at High Risk for Postoperative Adverse Events.

Importance: Identifying patients at high risk of adverse outcomes prior to surgery may allow for interventions associated with improved postoperative outcomes; however, few tools exist for automated prediction. Objective: To evaluate the accuracy of an automated machine-learning model in the identification of patients at high risk of adverse outcomes from surgery using only data in the electronic health record. Design, Setting, and Participants: This prognostic study was conducted among 1 477 561 patients undergoing surgery at 20 community and tertiary care hospitals in the University of Pittsburgh Medical Center (UPMC) health network. The study included 3 phases: (1) building and validating a model on a retrospective population, (2) testing model accuracy on a retrospective population, and (3) validating the model prospectively in clinical care. A gradient-boosted decision tree machine learning method was used for developing a preoperative surgical risk prediction tool. The Shapley additive explanations method was used for model interpretability and further validation. Accuracy was compared between the UPMC model and National Surgical Quality Improvement Program (NSQIP) surgical risk calculator for predicting mortality. Data were analyzed from September through December 2021. Exposure: Undergoing any type of surgical procedure. Main Outcomes and Measures: Postoperative mortality and major adverse cardiac and cerebrovascular events (MACCEs) at 30 days were evaluated. Results: Among 1 477 561 patients included in model development (806 148 females [54.5%; mean [SD] age, 56.8 [17.9] years), 1 016 966 patient encounters were used for training and 254 242 separate encounters were used for testing the model. After deployment in clinical use, another 206 353 patients were prospectively evaluated; an additional 902 patients were selected for comparing the accuracy of the UPMC model and NSQIP tool for predicting mortality. The area under the receiver operating characteristic curve (AUROC) for mortality was 0.972 (95% CI, 0.971-0.973) for the training set and 0.946 (95% CI, 0.943-0.948) for the test set. The AUROC for MACCE and mortality was 0.923 (95% CI, 0.922-0.924) on the training and 0.899 (95% CI, 0.896-0.902) on the test set. In prospective evaluation, the AUROC for mortality was 0.956 (95% CI, 0.953-0.959), sensitivity was 2148 of 2517 patients (85.3%), specificity was 186 286 of 203 836 patients (91.4%), and negative predictive value was 186 286 of 186 655 patients (99.8%). The model outperformed the NSQIP tool as measured by AUROC (0.945 [95% CI, 0.914-0.977] vs 0.897 [95% CI, 0.854-0.941], for a difference of 0.048), specificity (0.87 [95% CI, 0.83-0.89] vs 0.68 [95% CI, 0.65-0.69]), and accuracy (0.85 [95% CI, 0.82-0.87] vs 0.69 [95% CI, 0.66, 0.72]). Conclusions and Relevance: This study found that an automated machine learning model was accurate in identifying patients undergoing surgery who were at high risk of adverse outcomes using only preoperative variables within the electronic health record, with superior performance compared with the NSQIP calculator. These findings suggest that using this model to identify patients at increased risk of adverse outcomes prior to surgery may allow for individualized perioperative care, which may be associated with improved outcomes.

Association between conventional or blue-light-filtering intraocular lenses and survival in bilateral cataract surgery patients.

Circadian rhythms regulate adaptive alterations in mammalian physiology and are maximally entrained by the short wavelength blue spectrum; cataracts block the transmission of light, particularly blue light. Cataract surgery is performed with two types of intraocular lenses (IOL): (1) conventional IOL that transmit the entire visible spectrum and (2) blue-light-filtering (BF) IOL that block the short wavelength blue spectrum. We hypothesized that the transmission properties of IOL are associated with long-term survival. This retrospective cohort study of a 15-hospital healthcare system identified 9,108 participants who underwent bilateral cataract surgery; 3,087 were implanted with conventional IOL and 6,021 received BF-IOL. Multivariable Cox proportional hazards models that included several a priori determined subgroup and sensitivity analyses yielded estimates supporting that conventional IOL compared with BF-IOL may be associated with significantly reduced risk of long-term death. Confirming these differences and identifying any potential causal mechanisms await the conduct of appropriately controlled prospective translational trials.

Incentive payments to academic anesthesiologists for late afternoon work did not influence turnover times.

BACKGROUND: Anesthesiologists are often paid extra for hours worked in the late afternoon and evening. Although anesthesiologists have little influence on their operating room (OR) assignments and workloads late in the afternoon, they can influence turnover times. METHODS: OR turnover times on workdays were reviewed for n = 30 mo before there was incremental pay, for n = 15 mo with incremental pay for work past 3:30 pm, and for n = 8 mo with pay for work past 4:00 pm. The end point was the percentage of turnovers that were prolonged, defined as longer than 1 h. Turnovers straddling 3:30 pm (n = 3945), 4:00 pm (n = 3602), and 5:00 pm (n = 2834) were studied, as were those straddling 2:00 pm (n = 4407) as a control. In addition, qualitative (survey) assessment of n = 30 anesthesiologists was performed the last month to learn about their opinions on working late on weekdays. RESULTS: Most respondents considered an OR to run late if it finished after a specific time of day (87%, P < 0.001), unrelated to the room's type of procedures (90%, P < 0.001) or to the payment for working after 4:00 pm (100%, P < 0.001). There was no significant effect of implementation or changes to the incentive program on the incidences of prolonged turnover times at each of the studied times in the afternoon (all P > 0.14). CONCLUSION: Our results suggest that hospital administrators, deans, and other executives need not be especially concerned about disincentives produced by methods of internal compensation of anesthesiologists on highly visible OR turnover times late in afternoons.

Both bias and lack of knowledge influence organizational focus on first case of the day starts.

BACKGROUND: The economic costs of reducing first case delays are often high, because efforts need to be applied to multiple operating rooms (ORs) simultaneously. Nevertheless, delays in starting first cases of the day are a common topic in OR committee meetings. METHODS: We added three scientific questions to a 24 question online, anonymous survey performed before the implementation of a new OR information system. The 57 respondents cared sufficiently about OR management at the United States teaching hospital to complete all questions. RESULTS: The survey revealed reasons why personnel may focus on the small reductions in nonoperative time achievable by reducing tardiness in first cases of the day. (A) Respondents lacked knowledge about principles in reducing over-utilized OR time to increase OR efficiency, based on their answering the relevant question correctly at a rate no different from guessing at random. Those results differed from prior findings of responses at a rate worse than random, resulting from a bias on the day of surgery of making decisions that increase clinical work per unit time. (B) Most respondents falsely believed that a 10 min delay at the start of the day causes subsequent cases to start at least 10 min late (P < 0.0001 versus random chance). (C) Most respondents did not know that cases often take less time than scheduled (P = 0.008 versus chance). No one who demonstrated knowledge (C) about cases sometimes taking less time than scheduled applied that information to their response to (B) regarding cases starting late (P = 0.0002). CONCLUSIONS: Knowledge of OR efficiency was low among the respondents working in ORs. Nevertheless, the apparent absence of bias shows that education may influence behavior. In contrast, presence of bias on matters of tardiness of start times shows that education may be of no benefit. As the latter results match findings of previous studies of scheduling decisions, interventions to reduce patient and surgeon waiting from start times may depend principally on the application of automation to guide decision-making.