Dataset Information

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

ABSTRACT: 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.

SUBMITTER: Joseph JW

PROVIDER: S-EPMC7593422 | biostudies-literature | 2020 Oct

REPOSITORIES: biostudies-literature

ACCESS DATA

Publications

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

Joseph Joshua W JW Leventhal Evan L EL Grossestreuer Anne V AV Wong Matthew L ML Joseph Loren J LJ Nathanson Larry A LA Donnino Michael W MW Elhadad Noémie N Sanchez Leon D LD

Journal of the American College of Emergency Physicians open 20200901 5

<h4>Study objective</h4>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 ...[more]

PMID: 33145518

Similar Datasets

Project description:ObjectivesEmergency department boarding is the practice of caring for admitted patients in the emergency department after hospital admission, and boarding has been a growing problem in the United States. Boarding of the critically ill has achieved specific attention because of its association with poor clinical outcomes. Accordingly, the Society of Critical Care Medicine and the American College of Emergency Physicians convened a Task Force to understand the implications of emergency department boarding of the critically ill. The objective of this article is to review the U.S. literature on (1) the frequency of emergency department boarding among the critically ill, (2) the outcomes associated with critical care patient boarding, and (3) local strategies developed to mitigate the impact of emergency department critical care boarding on patient outcomes.Data sources and study selectionReview article.Data extraction and data synthesisEmergency department-based boarding of the critically ill patient is common, but no nationally representative frequency estimates has been reported. Boarding literature is limited by variation in the definitions used for boarding and variation in the facilities studied (boarding ranges from 2% to 88% of ICU admissions). Prolonged boarding in the emergency department has been associated with longer duration of mechanical ventilation, longer ICU and hospital length of stay, and higher mortality. Health systems have developed multiple mitigation strategies to address emergency department boarding of critically ill patients, including emergency department-based interventions, hospital-based interventions, and emergency department-based resuscitation care units.ConclusionsEmergency department boarding of critically ill patients was common and was associated with worse clinical outcomes. Health systems have generated a number of strategies to mitigate these effects. A definition for emergency department boarding is proposed. Future work should establish formal criteria for analysis and benchmarking of emergency department-based boarding overall, with subsequent efforts focused on developing and reporting innovative strategies that improve clinical outcomes of critically ill patients boarded in the emergency department.

Project description:ObjectivesIn mechanically ventilated patients, awareness with paralysis (AWP) can have devastating consequences, including post-traumatic stress disorder (PTSD), depression, and thoughts of suicide. Single-center data from the emergency department (ED) demonstrate an event rate for AWP factors higher than that reported from the operating room. However, there remains a lack of data on AWP among critically ill, mechanically ventilated patients. The objective was to assess the proportion of ED patients experiencing AWP and investigate modifiable variables associated with its occurrence.DesignAn a priori planned secondary analysis of a multicenter, prospective, before-and-after clinical trial.SettingThe ED of three academic medical centers.PatientsMechanically ventilated adult patients that received neuromuscular blockers.InterventionsNone.Measurements and main resultsAll data related to sedation and analgesia were collected. AWP was the primary outcome, assessed with the modified Brice questionnaire, and was independently adjudicated by three expert reviewers. Perceived threat, in the causal pathway for PTSD, was the secondary outcome. A total of 388 patients were studied. The proportion of patients experiencing AWP was 3.4% ( n = 13), the majority of whom received rocuronium ( n = 12/13; 92.3%). Among patients who received rocuronium, 5.5% ( n = 12/230) experienced AWP, compared with 0.6% ( n = 1/158) among patients who did not receive rocuronium in the ED (odds ratio, 8.64; 95% CI, 1.11-67.15). Patients experiencing AWP had a higher mean ( sd ) threat perception scale score, compared with patients without AWP (15.6 [5.8] vs 7.7 [6.0]; p < 0.01).ConclusionsAWP was present in a concerning proportion of mechanically ventilated ED patients, was associated with rocuronium exposure in the ED, and led to increased levels of perceived threat, placing patients at greater risk for PTSD. Studies that aim to further quantify AWP in this vulnerable population and eliminate its occurrence are urgently needed.

Project description:ObjectiveTo measure effects of ED crowding on lung-protective ventilation (LPV) utilization in critically ill ED patients.MethodsThis is a retrospective cohort study of adult mechanically ventilated ED patients admitted to the medical intensive care unit (MICU), over a 3.5-year period at a single academic tertiary care hospital. Clinical data, including reason for intubation, severity of illness (MPM0-III), acute respiratory distress syndrome (ARDS) risk score (EDLIPS), and ventilator settings were extracted via electronic query of electronic health record and standardized chart abstraction. Crowding metrics were obtained at 5-min intervals and averaged over the ED stay, stratified by acuity and disposition. Multivariate logistic regression was used to predict likelihood of LPV prior to ED departure.ResultsMechanical ventilation was used in 446 patients for a median ED duration of 3.7 h (interquartile ratio, IQR, 2.3, 5.6). Mean MPM0-III score was 32.5 ± 22.7, with high risk for ARDS (EDLIPS ≥5) seen in 373 (82%) patients. Initial and final ED ventilator settings differed in 134 (30.0%) patients, of which only 47 (35.1%) involved tidal volume changes. Higher percentages of active ED patients (workup in-progress) and those requiring eventual admission were associated with lower odds of LPV utilization by ED departure (OR 0.97, 95%CI 0.94-1.00; OR 0.97, 95%CI 0.94-1.00, respectively). In periods of high volume, ventilator adjustments to settings other than the tidal volume were associated with higher odds of LPV utilization. Reason for intubation, MPM0-III, and EDLIPS were not associated with LPV utilization, with no interactions detected in times of crowding.ConclusionsED patients remain on suboptimal tidal volume settings with infrequent ventilator adjustments during the ED stay. Hospitals should focus on both systemic factors and bedside physician and/or respiratory therapist interventions to increase LPV utilization in times of ED boarding and crowding for all patients.

Project description:OBJECTIVES:ICU admission delays can negatively affect patient outcomes, but emergency department volume and boarding times may also affect these decisions and associated patient outcomes. We sought to investigate the effect of emergency department and ICU capacity strain on ICU admission decisions and to examine the effect of emergency department boarding time of critically ill patients on in-hospital mortality. DESIGN:A retrospective cohort study. SETTING:Single academic tertiary care hospital. PATIENTS:Adult critically ill emergency department patients for whom a consult for medical ICU admission was requested, over a 21-month period. INTERVENTIONS:None. MEASUREMENTS AND MAIN RESULTS:Patient data, including severity of illness (Mortality Probability Model III on Admission), outcomes of mortality and persistent organ dysfunction, and hourly census reports for the emergency department, for all ICUs and all adult wards were compiled. A total of 854 emergency department requests for ICU admission were logged, with 455 (53.3%) as "accept" and 399 (46.7%) as "deny" cases, with median emergency department boarding times 4.2 hours (interquartile range, 2.8-6.3 hr) and 11.7 hours (3.2-20.3 hr) and similar rates of persistent organ dysfunction and/or death 41.5% and 44.6%, respectively. Those accepted were younger (mean ± SD, 61 ± 17 vs 65 ± 18 yr) and more severely ill (median Mortality Probability Model III on Admission score, 15.3% [7.0-29.5%] vs 13.4% [6.3-25.2%]) than those denied admission. In the multivariable model, a full medical ICU was the only hospital-level factor significantly associated with a lower probability of ICU acceptance (odds ratio, 0.55 [95% CI, 0.37-0.81]). Using propensity score analysis to account for imbalances in baseline characteristics between those accepted or denied for ICU admission, longer emergency department boarding time after consult was associated with higher odds of mortality and persistent organ dysfunction (odds ratio, 1.77 [1.07-2.95]/log10 hour increase). CONCLUSIONS:ICU admission decisions for critically ill emergency department patients are affected by medical ICU bed availability, though higher emergency department volume and other ICU occupancy did not play a role. Prolonged emergency department boarding times were associated with worse patient outcomes, suggesting a need for improved throughput and targeted care for patients awaiting ICU admission.

Project description:Objective: Delivery of prompt and adequate care for critically ill and injured children presenting to the pediatric emergency department (PED) is paramount for optimal outcomes. Knowledge of the local epidemiology, patient profile, and presentation modes are key for organizational planning, staff education strategy, and optimal care in a PED. Our aim was to analyze the profile of critically ill and injured children admitted to a tertiary, non-academic Swiss PED, to investigate potential risk factors associated with admission to the pediatric intensive care unit (PICU), and the outcomes mortality and PICU admission. Methods: Prospective cohort study of critically ill and injured children presenting to the PED over a two-year period (2018-2019). Inclusion criteria were Australasian triage scale category (ATS) 1, trauma team activation (TTA), medical emergency response (MER) activation, additional critical care consult, and transfer to an outside hospital. Results: Of 42,579 visits during the two-year period, 347 presentations matched the inclusion criteria (0.81%). Leading presentations were central nervous system (CNS) disorders (26.2%), trauma (25.1%), and respiratory emergencies (24.2%). 288 out of 347 cases (83%) arrived during the day or evening with an even distribution over the days of the week. 128 out of 347 (37%) arrived unexpectedly as walk-ins. 233 (67.15%) were ATS category 1. 51% of the cohort was admitted to PICU. Australasian triage scale category 1 was significantly more common in this group (p = 0.0001). Infants with respiratory disease had an increased risk of PICU transfer compared to other age groups (OR 4.18 [95%CI 2.46, 7.09] p = 0.0001), and this age group presented mainly as walk-in (p = 0.0001). Pediatric intensive care unit admissions had a longer hospital stay (4 [2, 8] days vs. 2 [1, 4] days, p = 0.0001) compared to other patients. 0.045% of all PED patients had to be transferred out. Three deaths (0.86%) occurred in the PED, 10 patients died in the PICU (2.9%). Conclusions: High acuity presentations in the PED were rare, more likely to be young with CNS disorders, trauma and respiratory diseases. A significant proportion were unexpected walk-in presentations, mainly during day and evening shifts. Low exposure to high-acuity patients highlights the importance of deliberate learning and simulation for all professionals in the PED.

Dataset Information

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

Publications

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

Similar Datasets

OmicsDI is part of the ELIXIR infrastructure

Tweets