Machine Learning Methods for Predicting Patient-Level Emergency Department Workload.

BACKGROUND: Work Relative Value Units (wRVUs) are a component of many compensation models, and a proxy for the effort required to care for a patient. Accurate prediction of wRVUs generated per patient at triage could facilitate real-time load balancing between physicians and provide many practical operational and clinical benefits. OBJECTIVE: We examined whether deep-learning approaches could predict the wRVUs generated by a patient's visit using data commonly available at triage. METHODS: Adult patients presenting to an urban, academic emergency department from July 1, 2016-March 1, 2020 were included. Deidentified triage information included structured data (age, sex, vital signs, Emergency Severity Index score, language, race, standardized chief complaint) and unstructured data (free-text chief complaint) with wRVUs as outcome. Five models were examined: average wRVUs per chief complaint, linear regression, neural network and gradient-boosted tree on structured data, and neural network on unstructured textual data. Models were evaluated using mean absolute error. RESULTS: We analyzed 204,064 visits between July 1, 2016 and March 1, 2020. The median wRVUs were 3.80 (interquartile range 2.56-4.21), with significant effects of age, gender, and race. Models demonstrated lower error as complexity increased. Predictions using averages from chief complaints alone demonstrated a mean error of 2.17 predicted wRVUs per visit (95% confidence interval [CI] 2.07-2.27), the linear regression model: 1.00 wRVUs (95% CI 0.97-1.04), gradient-boosted tree: 0.85 wRVUs (95% CI 0.84-0.86), neural network with structured data: 0.86 wRVUs (95% CI 0.85-0.87), and neural network with unstructured data: 0.78 wRVUs (95% CI 0.76-0.80). CONCLUSIONS: Chief complaints are a poor predictor of the effort needed to evaluate a patient; however, deep-learning techniques show promise. These algorithms have the potential to provide many practical applications, including balancing workloads and compensation between emergency physicians, quantify crowding and mobilizing resources, and reducing bias in the triage process.

Association of resident characteristics with patterns of patient self-assignment.

OBJECTIVE: We hypothesized that resident characteristics impact patterns of patient self-assignment in the emergency department (ED). Our goal was to determine if male residents would be less likely than their female colleagues to see patients with sensitive (e.g. breast-related or gynecologic) chief complaints (CCs). We also investigated whether resident specialty was associated with preferentially choosing patients with more familiar chief complaints. METHODS: We performed a retrospective cross-sectional study at a tertiary academic medical center using data from all adult patients presenting to the ED between 2010 and 2019 with one of six CC categories (vaginal bleeding, breast-related concerns, male genitourinary [GU] concerns, gastrointestinal bleeding, epistaxis, and laceration). These CCs were chosen as they each require either an invasive medical exam or procedure, and cannot easily be evaluated with an exam in a hallway bed. We used logistic regression to assess the likelihood of being treated by a male resident compared to a female resident for each CC, adjusting for candidate variables of patient age, race, primary language, ESI score, bed location, time of day, day of week, calendar month, and resident specialty. We also similarly analyzed patterns of patient self-assignment according to resident specialty. RESULTS: Male residents were significantly less likely than female residents to treat patients with breast-related CCs (adjusted OR 0.67, 95% CI 0.54-0.83, p < 0.001) or vaginal bleeding (adjusted OR 0.73, 95% CI 0.63-0.84, p < 0.001, reference group: epistaxis). Off-service residents were more likely to assign themselves to familiar chief complaints, for example surgery residents were more likely to see patients with lacerations (adjusted OR 2.11, 95% CI 1.71-2.61, p < 0.001) and OB/GYN residents were less likely to see patients with male GU concerns (adjusted OR 0.21, 95% CI 0.05-0.85, p = 0.029), compared to emergency medicine residents. CONCLUSION: In a single facility, resident characteristics were associated with preferential patient self-assignment. Further work is necessary to determine the underlying reasons for patient avoidance, and to create work environments in which preferentially choosing patients is discouraged.

Early, awake proning in emergency department patients with COVID-19.

OBJECTIVE: Proning has been shown to improve oxygenation and mortality in certain populations of intubated patients with acute respiratory distress syndrome. Small observational analyses of COVID-19 patients suggest awake proning may lead to clinical improvement. Data on safety and efficacy is lacking. We sought to describe the effect of proning on oxygenation in nonintubated COVID-19 patients. We also evaluated feasibility, safety, and other physiological and clinical outcomes associated with this intervention. METHODS: We conducted a prospective, observational cohort study of nonintubated patients with COVID-19 who underwent proning per an Emergency Department (ED) clinical protocol. Patients with mild to moderate respiratory distress were included. We calculated change in oxygenation by comparing the oxygen saturation to fraction of inspired oxygen ratio (SpO2:FiO2) during the five minutes prior to proning and first 30 min of proning. We also captured data on respiratory rate, duration of proning, need for intubation, intensive care unit admission, survival to discharge. RESULTS: Fifty-two patients were enrolled. Thirty were excluded for not meeting protocol inclusion criteria or missing baseline oxygenation data, leaving 22 for analysis. The SpO2:FiO2 ratio increased by a median of 5 (IQR: 0-15) in the post-proning period compared to the pre-proning period (median: 298 (IQR: 263-352) vs 295 (IQR: 276-350), p = 0.01). Respiratory rate did not change significantly between time periods. No immediate adverse events occurred during proning. Five patients (23%) were intubated within 48 h of admission. CONCLUSION: Early, awake proning may be feasible in select COVID-19 patients and was associated with improved oxygenation.

Does the number of pages received per hour affect resident productivity?

Background: Workflow interruptions are common for emergency physicians and are shown to have downstream consequences such as patient dissatisfaction, delay in clinical response, and increase in medical error. However, the impact of passive interruptions on physician productivity is unclear and has not been well studied. We sought to evaluate if the number of pages received per hour significantly affects the number of patients seen per hour. Methods: Retrospective data was collected on resident physician (RP) emergency department shifts from July 1st, 2021 to June 30th, 2022 at an academic medical center with an annual census of 55,000 patients. A total of 2865 RP shifts were collected among the 26 postgraduate year (PGY) 1 and PGY2 residents. For each RP shift, we identified the number of pages received per hour and the number of new patients seen per hour. Pages consist of any transmitted message that was sent to the RP's personal pager, which includes both automatic (eg, bed assignments, abnormal lab values) and personalized pages from other healthcare practicioners (eg, nursing, consultants). Data were analyzed using Poisson regression controlling for clustering at the physician level to determine if the number of patients seen per hour is significantly affected by the number of pages (divided into quartiles) received. Results: We found the number of pages received per hour did not decrease the number of patients seen per hour. Contrary to our hypothesis, there was a strong positive relationship between the number of pages received per hour and the number of patients seen by RPs in that hour and subsequent hours. During the middle of a shift (hours 3, 4, and 5), RPs receiving pages in the third and fourth quartile (top 50% of pages) saw significantly more patients during that same hour and the next hour (p <0.001). Conclusion: The number of pages received by RPs per hour did not decrease the number of patients seen per hour. When RPs receive a higher number of pages, there is a positive association with the number of patients they see in that hour and subsequent hours. Further studies will be needed to determine whether the content of pages affects resident productivity.

Impact of a Mandatory Prescription Drug Monitoring Program Check on Emergency Department Opioid Prescribing Rates.

BACKGROUND: Prescription drug monitoring programs (PDMPs) exist in 49 states to guide opioid prescribing. In 40 states, clinicians must check the PDMP prior to prescribing an opioid. Data on mandated PDMP checks show mixed results on opioid prescribing. OBJECTIVES: This study sought to examine the impact of the Massachusetts mandatory PDMP check on opioid prescribing for discharges from an urban tertiary emergency department (ED). METHODS: This was a retrospective cohort study of discharges from one ED from 7/1/2010-10/15/2018. The primary outcome was the monthly percentage of patients discharged from the ED with an opioid prescription. The intervention was Massachusetts mandating a PDMP check for all opioid prescriptions. Prescribing was compared pre- and post-mandate. Interrupted time series (ITS) analysis accounted for known declining trends in opioid prescribing. RESULTS: Of 273,512 ED discharges, 35,050 (12.8%) received opioid prescriptions. Mean monthly opioid prescribing decreased post-intervention from 15.1% (SD ± 3.5%) to 5.1% (SD ± 0.9%; p < 0.001). ITS showed equal pre and post-intervention slopes (-0.002, p = 0.819). A small immediate decrease occurred in prescribing around the mandated check: a 3-month level effect decrease of 0.018 (p = 0.039), 6-month level effect 0.019 (p = 0.023), and a 12-month level effect of 0.020 (p = 0.019). The 24-month level effect was not decreased. CONCLUSION: Prior to the mandated PDMP check, ED opioid prescribing was declining. The mandate did not change the rate of decline but was associated with a non-sustained drop in opioid prescribing immediately following enactment.

Conformational changes in the carpus during finger trap distraction.

PURPOSE: Wrist distraction is a common treatment maneuver used clinically for the reduction of distal radial fractures and midcarpal dislocations. Wrist distraction is also required during wrist arthroscopy to access the radiocarpal joint and has been used as a test for scapholunate ligament injury. However, the effect of a distraction load on the normal wrist has not been well studied. The purpose of this study was to measure the three-dimensional conformational changes of the carpal bones in the normal wrist as a result of a static distractive load. METHODS: Using computed tomography, the dominant wrists of 14 healthy volunteers were scanned at rest and during application of 98 N of distraction. Load was applied using finger traps, and volunteers were encouraged to relax their forearm muscles and to allow distraction of the wrist. The motions of the bones in the wrist were tracked between the unloaded and loaded trial using markerless bone registration. The average displacement vector of each bone relative to the radius was calculated, as were the interbone distances for 20 bone-bone interactions. Joint separation was estimated at the radiocarpal, midcarpal, and carpometacarpal joints in the direction of loading using the radius, lunate, capitate, and third metacarpal. RESULTS: With loading, the distance between the radius and third metacarpal increased an average of 3.3 mm +/- 3.1 in the direction of loading. This separation was primarily in the axial direction at the radiocarpal (1.0 mm +/- 1.0) and midcarpal (2.0 mm +/- 1.7) joints. There were minimal changes in the transverse direction within the distal row, although the proximal row narrowed by 0.98 mm +/- 0.7. Distraction between the radius and scaphoid (2.5 mm +/- 2.2) was 2.4 times greater than that between the radius and lunate (1.0 mm +/- 1.0). CONCLUSIONS: Carpal distraction has a significant (p < .01) effect on the conformation of the carpus, especially at the radiocarpal and midcarpal joints. In the normal wrist, external traction causes twice as much distraction at the lunocapitate joint than at the radiolunate joint.

Carpal and forearm kinematics during a simulated hammering task.

PURPOSE: Hammering is a functional task in which the wrist generally follows a path of motion from a position of combined radial deviation and extension to combined ulnar deviation and flexion, colloquially referred to as a dart thrower's motion. The purpose of this study was to measure wrist and forearm motion and scaphoid and lunate kinematics during a simulated hammering task. We hypothesized that the wrist follows an oblique path from radial extension to ulnar flexion and that there would be minimal radiocarpal motion during the hammering task. METHODS: Thirteen healthy volunteers consented to have their wrist and distal forearm imaged with computed tomography at 5 positions while performing a simulated hammering task. The kinematics of the carpus and distal radioulnar joint were calculated using established markerless bone registration methods. The path of wrist motion was described relative to the sagittal plane. Forearm rotation and radioscaphoid and radiolunate motion were computed as a function of wrist position. RESULTS: All volunteers performed the simulated hammering task using a path of wrist motion from radial extension to ulnar flexion that was oriented an average of 41 degrees +/- 3 degrees from the sagittal plane. These paths did not pass through the anatomic neutral wrist position; rather, they passed through a neutral hammering position, which was offset by 36 degrees +/- 8 degrees in extension. Rotations of the scaphoid and lunate were not minimal but averaged 40% and 41%, respectively, of total wrist motion. The range of forearm pronation-supination during the task averaged 12 degrees +/- 8 degrees . CONCLUSIONS: The simulated hammering task was performed using a wrist motion that followed a coupled path of motion, from extension and radial deviation to flexion and ulnar deviation. Scaphoid and lunate rotations were greatly reduced, but not minimized, compared with rotations during pure wrist flexion/extension. This is likely because an extended wrist position was maintained throughout the entire task studied.

Information Management in the Emergency Department.

Information management in the emergency department (ED) is a challenge for all providers. The volume of information required to care for each patient and to keep the ED functioning is immense. It must be managed through varying means of communication and in connection with ED information systems. Management of information in the ED is imperfect; different modes and methods of identification, interpretation, action, and communication can be beneficial or harmful to providers, patients, and departmental flow. This article reviews the state of information management in the ED and proposes recommendations to improve the management of information in the future.

Deep-learning approaches to identify critically Ill patients at emergency department triage using limited information.

STUDY OBJECTIVE: Triage quickly identifies critically ill patients, facilitating timely interventions. Many emergency departments (EDs) use emergency severity index (ESI) or abnormal vital sign triggers to guide triage. However, both use fixed thresholds, and false activations are costly. Prior approaches using machinelearning have relied on information that is often unavailable during the triage process. We examined whether deep-learning approaches could identify critically ill patients only using data immediately available at triage. METHODS: We conducted a retrospective, cross-sectional study at an urban tertiary care center, from January 1, 2012-January 1, 2020. De-identified triage information included structured (age, sex, initial vital signs) and textual (chief complaint) data, with critical illness (mortality or ICU admission within 24 hours) as the outcome. Four progressively complex deep-learning models were trained and applied to triage information from all patients. We compared the accuracy of the models against ESI as the standard diagnostic test, using area under the receiver-operator curve (AUC). RESULTS: A total of 445,925 patients were included, with 60,901 (13.7%) critically ill. Vital sign thresholds identified critically ill patients with AUC 0.521 (95% confidence interval [CI] = 0.519-0.522), and ESI <3 demonstrated AUC 0.672 (95% CI = 0.671-0.674), logistic regression classified patients with AUC 0.803 (95% CI = 0.802-0.804), 2-layer neural network with structured data with AUC 0.811 (95% CI = 0.807-0.815), gradient tree boosting with AUC 0.820 (95% CI = 0.818-0.821), and the neural network model with textual data with AUC 0.851 (95% CI = 0.849-0.852). All successive increases in AUC were statistically significant. CONCLUSION: Deep-learning techniques represent a promising method of augmenting triage, even with limited information. Further research is needed to determine if improved predictions yield clinical and operational benefits.