Variation in Intraoperative Opioid Administration by Patient, Clinician, and Hospital Contribution.

Importance: The opioid crisis has led to scrutiny of opioid exposures before and after surgical procedures. However, the extent of intraoperative opioid variation and the sources and contributing factors associated with it are unclear. Objective: To analyze attributable variance of intraoperative opioid administration for patient-, clinician-, and hospital-level factors across surgical and analgesic categories. Design, Setting, and Participants: This cohort study was conducted using electronic health record data collected from a national quality collaborative database. The cohort consisted of 1 011 268 surgical procedures at 46 hospitals across the US involving 2911 anesthesiologists, 2291 surgeons, and 8 surgical and 4 analgesic categories. Patients without ambulatory opioid prescriptions or use history undergoing an elective surgical procedure between January 1, 2014, and September 11, 2020, were included. Data were analyzed from January 2022 to July 2023. Main Outcomes and Measures: The rate of intraoperative opioid administration as a continuous measure of oral morphine equivalents (OMEs) normalized to patient weight and case duration was assessed. Attributable variance was estimated in a hierarchical structure using patient, clinician, and hospital levels and adjusted intraclass correlations (ICCs). Results: Among 1 011 268 surgical procedures (mean [SD] age of patients, 55.9 [16.2] years; 604 057 surgical procedures among females [59.7%]), the mean (SD) rate of intraoperative opioid administration was 0.3 [0.2] OME/kg/h. Together, clinician and hospital levels contributed to 20% or more of variability in intraoperative opioid administration across all analgesic and surgical categories (adjusting for surgical or analgesic category, ICCs ranged from 0.57-0.79 for the patient, 0.04-0.22 for the anesthesiologist, and 0.09-0.26 for the hospital, with the lowest ICC combination 0.21 for anesthesiologist and hosptial [0.12 for the anesthesiologist and 0.09 for the hospital for opioid only]). Comparing the 95th and fifth percentiles of opioid administration, variation was 3.3-fold among anesthesiologists (surgical category range, 2.7-fold to 7.7-fold), 4.3-fold among surgeons (surgical category range, 3.4-fold to 8.0-fold), and 2.2-fold among hospitals (surgical category range, 2.2-fold to 4.3-fold). When adjusted for patient and surgical characteristics, mean (square error mean) administration was highest for cardiac surgical procedures (0.54 [0.56-0.52 OME/kg/h]) and lowest for orthopedic knee surgical procedures (0.19 [0.17-0.21 OME/kg/h]). Peripheral and neuraxial analgesic techniques were associated with reduced administration in orthopedic hip (51.6% [95% CI, 51.4%-51.8%] and 60.7% [95% CI, 60.5%-60.9%] reductions, respectively) and knee (48.3% [95% CI, 48.0%-48.5%] and 60.9% [95% CI, 60.7%-61.1%] reductions, respectively) surgical procedures, but reduction was less substantial in other surgical categories (mean [SD] reduction, 13.3% [8.8%] for peripheral and 17.6% [9.9%] for neuraxial techniques). Conclusions and Relevance: In this cohort study, clinician-, hospital-, and patient-level factors had important contributions to substantial variation of opioid administrations during surgical procedures. These findings suggest the need for a broadened focus across multiple factors when developing and implementing opioid-reducing strategies in collaborative quality-improvement programs.

Classification of Current Procedural Terminology Codes from Electronic Health Record Data Using Machine Learning.

BACKGROUND: Accurate anesthesiology procedure code data are essential to quality improvement, research, and reimbursement tasks within anesthesiology practices. Advanced data science techniques, including machine learning and natural language processing, offer opportunities to develop classification tools for Current Procedural Terminology codes across anesthesia procedures. METHODS: Models were created using a Train/Test dataset including 1,164,343 procedures from 16 academic and private hospitals. Five supervised machine learning models were created to classify anesthesiology Current Procedural Terminology codes, with accuracy defined as first choice classification matching the institutional-assigned code existing in the perioperative database. The two best performing models were further refined and tested on a Holdout dataset from a single institution distinct from Train/Test. A tunable confidence parameter was created to identify cases for which models were highly accurate, with the goal of at least 95% accuracy, above the reported 2018 Centers for Medicare and Medicaid Services (Baltimore, Maryland) fee-for-service accuracy. Actual submitted claim data from billing specialists were used as a reference standard. RESULTS: Support vector machine and neural network label-embedding attentive models were the best performing models, respectively, demonstrating overall accuracies of 87.9% and 84.2% (single best code), and 96.8% and 94.0% (within top three). Classification accuracy was 96.4% in 47.0% of cases using support vector machine and 94.4% in 62.2% of cases using label-embedding attentive model within the Train/Test dataset. In the Holdout dataset, respective classification accuracies were 93.1% in 58.0% of cases and 95.0% among 62.0%. The most important feature in model training was procedure text. CONCLUSIONS: Through application of machine learning and natural language processing techniques, highly accurate real-time models were created for anesthesiology Current Procedural Terminology code classification. The increased processing speed and a priori targeted accuracy of this classification approach may provide performance optimization and cost reduction for quality improvement, research, and reimbursement tasks reliant on anesthesiology procedure codes.

Alerting thresholds for the prevention of intraoperative awareness with explicit recall: a secondary analysis of the Michigan Awareness Control Study.

BACKGROUND: Intraoperative awareness with explicit recall is a potentially devastating complication of surgery that has been attributed to low anaesthetic concentrations in the vast majority of cases. Past studies have proposed the determination of an adequate dose for general anaesthetics that could be used to alert providers of potentially insufficient anaesthesia. However, there have been no systematic analyses of appropriate thresholds to develop population-based alerting algorithms for preventing intraoperative awareness. OBJECTIVE: To identify a threshold for intraoperative alerting that could be applied for the prevention of awareness with explicit recall. DESIGN: Secondary analysis of a randomised controlled trial (Michigan Awareness Control Study). SETTING: Three hospitals at a tertiary care centre in the USA. PATIENTS: Unselected patients presenting for surgery under general anaesthesia. INTERVENTIONS: Alerts based on end-tidal anaesthetic concentration or bispectral index values. MAIN OUTCOME MEASURES: Using case and outcomes data from the primary study, end-tidal anaesthetic concentration and bispectral index values were analysed using Youden's index and c-statistics derived from a receiver operating characteristic curve to determine a specific alerting threshold for the prevention of awareness. RESULTS: No single population-based threshold that maximises sensitivity and specificity could be identified for the prevention of intraoperative awareness, using either anaesthetic concentration or bispectral index values. The c-statistic for anaesthetic concentration was 0.431 ±â€Š0.046, and 0.491 ±â€Š0.056 for bispectral index values. CONCLUSION: We could not derive a single population-based alerting threshold for the prevention of intraoperative awareness using either anaesthetic concentration or bispectral index values. These data indicate a need to move towards individualised alerting strategies in the prevention of intraoperative awareness. TRIAL REGISTRATION: Primary trial registration (Michigan Awareness Control Study) ClinicalTrials.gov identifier: NCT00689091.

Prevention of intraoperative awareness with explicit recall in an unselected surgical population: a randomized comparative effectiveness trial.

BACKGROUND: Intraoperative awareness with explicit recall occurs in approximately 0.15% of all surgical cases. Efficacy trials based on the Bispectral Index® (BIS) monitor (Covidien, Boulder, CO) and anesthetic concentrations have focused on high-risk patients, but there are no effectiveness data applicable to an unselected surgical population. METHODS: We conducted a randomized controlled trial of unselected surgical patients at three hospitals of a tertiary academic medical center. Surgical cases were randomized to alerting algorithms based on either BIS values or anesthetic concentrations. The primary outcome was the incidence of definite intraoperative awareness; prespecified secondary outcomes included postanesthetic recovery variables. RESULTS: The study was terminated because of futility. At interim analysis the incidence of definite awareness was 0.12% (11/9,376) (95% CI: 0.07-0.21%) in the anesthetic concentration group and 0.08% (8/9,460) (95% CI: 0.04-0.16%) in the BIS group (P = 0.48). There was no significant difference between the two groups in terms of meeting criteria for recovery room discharge or incidence of nausea and vomiting. By post hoc secondary analysis, the BIS protocol was associated with a 4.7-fold reduction in definite or possible awareness events compared with a cohort receiving no intervention (P = 0.001; 95% CI: 1.7-13.1). CONCLUSION: This negative trial could not detect a difference in the incidence of definite awareness or recovery variables between monitoring protocols based on either BIS values or anesthetic concentration. By post hoc analysis, a protocol based on BIS monitoring reduced the incidence of definite or possible intraoperative awareness compared with routine care.

A novel electronic algorithm for detecting potentially insufficient anesthesia: implications for the prevention of intraoperative awareness.

OBJECTIVE: A recent clinical trial compared a minimum alveolar concentration (MAC)-based protocol to an electroencephalography (EEG)-based protocol for the prevention of intraoperative awareness. One limitation of this study design is that MAC-based protocols are not sensitive to the use of intravenous agents, while EEG-based protocols are. Our objective was to develop a MAC alert that incorporates intravenous agents. METHODS: We developed an electronic algorithm and alerting system that calculates a total age-adjusted MAC value based on inhalational agents, but also incorporates intravenous agents. We retrospectively applied the algorithm to adult general anesthesia cases over a 1 year period to assess the frequency of alert triggers, using thresholds of <0.8, <0.7, <0.6, <0.5 and <0.4 age-adjusted MAC. We also electronically analyzed 12 cases of intraoperative awareness that occurred over a 4-year period for the frequency of alert triggers using the same thresholds. Finally, we calculated positive and negative likelihood ratios based on these analyses. RESULTS: Over a 1-year period we identified 15,091 cases without self-reported awareness that were valid for analysis. At all age-adjusted MAC thresholds, the incidence of triggered alerts was higher in the awareness cases. The threshold of<0.8 age-adjusted MAC was associated with the highest negative likelihood ratio; the<0.5 age-adjusted MAC was associated with the highest positive likelihood ratio. CONCLUSIONS: Our novel electronic alerting system incorporates both age-adjusted MAC and intravenous anesthesia, and triggers with a higher frequency in cases of awareness. These data suggest the potential for our system to alert clinicians to insufficient anesthesia.

A retrospective study of intraoperative awareness with methodological implications.

BACKGROUND: Awareness during general anesthesia is a problem receiving increased attention from physicians and patients. Large multicentered studies have established an accepted incidence of awareness during general anesthesia as approximately 1-2 per 1000 cases or 0.15%. More recent retrospective data, however, suggest that the actual incidence may be as low as 0.0068%. METHODS: To assess the incidence of awareness at our institution, we conducted a review of adult patients undergoing surgical procedures over a 3-year period. Information on awareness came from entries of "Intraoperative Awareness" captured during our standard evaluations on postoperative day one in our perioperative information system. Patients were not questioned specifically about awareness. RESULTS: We reviewed 116,478 charts; 65,061 patients received general anesthesia and 51,417 received other types of anesthesia. Of the patients receiving general anesthesia, 44,006 had complete postoperative documentation. The reported incidence of undesired intraoperative awareness in this population was 10/44,006 (1/4401 or 0.023%). Of the patients who received other anesthetic modalities, 22,885 had complete postoperative documentation. Undesired intraoperative awareness was reported in 7/22,885 patients who did not receive general anesthesia (1/3269 or 0.03%). The reported incidence of intraoperative awareness was not statistically different between the two groups (P = 0.54). Relative risk of intraoperative awareness during a general anesthetic compared with a nongeneral anesthetic was 0.74, with 95% confidence interval [0.28, 2.0]. CONCLUSION: Using a retrospective methodology, reports of intraoperative awareness are not statistically different in patients who received general anesthesia compared with those who did not. These results suggest that, despite success with other rare perioperative events, the resolution of retrospective database analyses may be too low to study intraoperative awareness.