Predictors of Acute Atrial Fibrillation and Flutter Hospitalization across 7 U.S. Emergency Departments: A Prospective Study.

INTRODUCTION: International rates of hospitalization for atrial fibrillation and flutter (AFF) from the emergency department (ED) vary widely without clear evidence to guide the identification of high-risk patients requiring inpatient management. We sought to determine (1) variation in hospital admission and (2) modifiable factors associated with hospitalization of AFF patients within a U.S. integrated health system. METHODS: This multicenter prospective observational study of health plan members with symptomatic AFF was conducted using convenience sampling in 7 urban community EDs from 05/2011 to 08/2012. Prospective data collection included presenting symptoms, characteristics of atrial dysrhythmia, ED physician impression of hemodynamic instability, comorbid diagnoses, ED management, and ED discharge rhythm. All centers had full-time on-call cardiology consultation available. Additional variables were extracted from the electronic health record. We identified factors associated with hospitalization and included predictors in a multivariate Poisson Generalized Estimating Equations regression model to estimate adjusted relative risks while accounting for clustering by physician. RESULTS: Among 1,942 eligible AFF patients, 1,074 (55.3%) were discharged home and 868 (44.7%) were hospitalized. Hospitalization rates ranged from 37.4% to 60.4% across medical centers. After adjustment, modifiable factors associated with increased hospital admission from the ED included non-sinus rhythm at ED discharge, no attempted cardioversion, and heart rate reduction. DISCUSSION: Within an integrated health system, we found significant variation in AFF hospitalization rates and identified several modifiable factors associated with hospital admission. Standardizing treatment goals that specifically address best practices for ED rate reduction and rhythm control may reduce hospitalizations.

Graded Coronary Risk Stratification for Emergency Department Patients With Chest Pain: A Controlled Cohort Study.

Background Resource utilization among emergency department (ED) patients with possible coronary chest pain is highly variable. Methods and Results Controlled cohort study amongst 21 EDs of an integrated healthcare system examining the implementation of a graded coronary risk stratification algorithm (RISTRA-ACS [risk stratification for acute coronary syndrome]). Thirteen EDs had access to RISTRA-ACS within the electronic health record (RISTRA sites) beginning in month 24 of a 48-month study period (January 2016 to December 2019); the remaining 8 EDs served as contemporaneous controls. Study participants had a chief complaint of chest pain and serum troponin measurement in the ED. The primary outcome was index visit resource utilization (observation unit or hospital admission, or 7-day objective cardiac testing). Secondary outcomes were 30-day objective cardiac testing, 60-day major adverse cardiac events (MACE), and 60-day MACE-CR (MACE excluding coronary revascularization). Difference-in-differences analyses controlled for secular trends with stratification by estimated risk and adjustment for risk factors, ED physician and facility. A total of 154 914 encounters were included. Relative to control sites, 30-day objective cardiac testing decreased at RISTRA sites among patients with low (≤2%) estimated 60-day MACE risk (-2.5%, 95% CI -3.7 to -1.2%, P<0.001) and increased among patients with non-low (>2%) estimated risk (+2.8%, 95% CI +0.6 to +4.9%, P=0.014), without significant overall change (-1.0%, 95% CI -2.1 to 0.1%, P=0.079). There were no statistically significant differences in index visit resource utilization, 60-day MACE or 60-day MACE-CR. Conclusions Implementation of RISTRA-ACS was associated with better allocation of 30-day objective cardiac testing and no change in index visit resource utilization or 60-day MACE. Registration URL: https://www.clinicaltrials.gov; Unique identifier: NCT03286179.

Prospective Validation and Comparative Analysis of Coronary Risk Stratification Strategies Among Emergency Department Patients With Chest Pain.

Background Coronary risk stratification is recommended for emergency department patients with chest pain. Many protocols are designed as "rule-out" binary classification strategies, while others use graded-risk stratification. The comparative performance of competing approaches at varying levels of risk tolerance has not been widely reported. Methods and Results This is a prospective cohort study of adult patients with chest pain presenting between January 2018 and December 2019 to 13 medical center emergency departments within an integrated healthcare delivery system. Using an electronic clinical decision support interface, we externally validated and assessed the net benefit (at varying risk thresholds) of several coronary risk scores (History, ECG, Age, Risk Factors, and Troponin [HEART] score, HEART pathway, Emergency Department Assessment of Chest Pain Score Accelerated Diagnostic Protocol), troponin-only strategies (fourth-generation assay), unstructured physician gestalt, and a novel risk algorithm (RISTRA-ACS). The primary outcome was 60-day major adverse cardiac event defined as myocardial infarction, cardiac arrest, cardiogenic shock, coronary revascularization, or all-cause mortality. There were 13 192 patient encounters included with a 60-day major adverse cardiac event incidence of 3.7%. RISTRA-ACS and HEART pathway had the lowest negative likelihood ratios (0.06, 95% CI, 0.03-0.10 and 0.07, 95% CI, 0.04-0.11, respectively) and the greatest net benefit across a range of low-risk thresholds. RISTRA-ACS demonstrated the highest discrimination for 60-day major adverse cardiac event (area under the receiver operating characteristic curve 0.92, 95% CI, 0.91-0.94, P<0.0001). Conclusions RISTRA-ACS and HEART pathway were the optimal rule-out approaches, while RISTRA-ACS was the best-performing graded-risk approach. RISTRA-ACS offers promise as a versatile single approach to emergency department coronary risk stratification. Registration URL: https://www.clinicaltrials.gov; Unique identifier: NCT03286179.

Effect of Clinical Decision Support on Diagnostic Imaging for Pediatric Appendicitis: A Cluster Randomized Trial.

Importance: Appendicitis is the most common pediatric surgical emergency. Efforts to improve efficiency and quality of care have increased reliance on computed tomography (CT) and ultrasonography (US) in children with suspected appendicitis. Objective: To evaluate the effectiveness of an electronic health record-linked clinical decision support intervention, AppyCDS, on diagnostic imaging, health care costs, and safety outcomes for patients with suspected appendicitis. Design, Setting, and Participants: In this parallel, cluster randomized trial, 17 community-based general emergency departments (EDs) in California, Minnesota, and Wisconsin were randomized to the AppyCDS intervention group or usual care (UC) group. Patients were aged 5 to 20 years, presenting for an ED visit with right-sided or diffuse abdominal pain lasting 5 days or less. We excluded pregnant patients, those with a prior appendectomy, those with selected comorbidities, and those with traumatic injuries. The trial was conducted from October 2016 to July 2019. Interventions: AppyCDS prompted data entry at the point of care to estimate appendicitis risk using the pediatric appendicitis risk calculator (pARC). Based on pARC estimates, AppyCDS recommended next steps in care. Main Outcomes and Measures: Primary outcomes were CT, US, or any imaging (CT or US) during the index ED visit. Safety outcomes were perforations, negative appendectomies, and missed appendicitis. Costs were a secondary outcome. Ratio of ratios (RORs) for primary and safety outcomes and differences by group in cost were used to evaluate effectiveness of the clinical decision support tool. Results: We enrolled 3161 patients at intervention EDs and 2779 patients at UC EDs. The mean age of patients was 11.9 (4.6) years and 2614 (44.0%) were boys or young men. RORs for CT (0.94; 95% CI, 0.75-1.19), US (0.98; 95% CI, 0.84-1.14), and any imaging (0.96; 95% CI, 0.86-1.07) did not differ by study group. In an exploratory analysis conducted in 1 health system, AppyCDS was associated with a reduction in any imaging (ROR, 0.82; 95% CI, 0.73- 0.93) for patients with pARC score of 15% or less and a reduction in CT (ROR, 0.58; 95% CI, 0.45-0.74) for patients with a pARC score of 16% to 50%. Perforations, negative appendectomies, and cases of missed appendicitis by study phase did not differ significantly by study group. Costs did not differ overall by study group. Conclusions and Relevance: In this study, AppyCDS was not associated with overall reductions in diagnostic imaging; exploratory analysis revealed more appropriate use of imaging in patients with a low pARC score. Trial Registration: ClinicalTrials.gov Identifier: NCT02633735.

Automated Retrospective Calculation of the EDACS and HEART Scores in a Multicenter Prospective Cohort of Emergency Department Chest Pain Patients.

OBJECTIVES: Coronary risk scores are commonly applied to emergency department patients with undifferentiated chest pain. Two prominent risk score-based protocols are the Emergency Department Assessment of Chest pain Score Accelerated Diagnostic Protocol (EDACS-ADP) and the History, ECG, Age, Risk factors, and Troponin (HEART) pathway. Since prospective documentation of these risk determinations can be challenging to obtain, quality improvement projects could benefit from automated retrospective risk score classification methodologies. METHODS: EDACS-ADP and HEART pathway data elements were prospectively collected using a Web-based electronic clinical decision support (eCDS) tool over a 24-month period (2018-2019) among patients presenting with chest pain to 13 EDs within an integrated health system. Data elements were also extracted and processed electronically (retrospectively) from the electronic health record (EHR) for the same patients. The primary outcome was agreement between the prospective/eCDS and retrospective/EHR data sets on dichotomous risk protocol classification, as assessed by kappa statistics (ĸ). RESULTS: There were 12,110 eligible eCDS uses during the study period, of which 66 and 47% were low-risk encounters by EDACS-ADP and HEART pathway, respectively. Agreement on low-risk status was acceptable for EDACS-ADP (ĸ = 0.73, 95% confidence interval [CI] = 0.72 to 0.75) and HEART pathway (ĸ = 0.69, 95% CI = 0.68 to 0.70) and for the continuous scores (interclass correlation coefficients = 0.87 and 0.84 for EDACS and HEART, respectively). CONCLUSIONS: Automated retrospective determination of low risk status by either the EDACS-ADP or the HEART pathway provides acceptable agreement compared to prospective score calculations, providing a feasible risk adjustment option for use in large data set analyses.

Diagnostic Performance of Emergency Physician Gestalt for Predicting Acute Appendicitis in Patients Age 5 to 20 Years.

OBJECTIVES: Pediatric appendicitis remains a challenging diagnosis in the emergency department (ED). Available risk prediction algorithms may contribute to excessive ED imaging studies. Incorporation of physician gestalt assessment could help refine predictive tools and improve diagnostic imaging decisions. METHODS: This study was a subanalysis of a parent study that prospectively enrolled patients ages 5 to 20.9 years with a chief complaint of abdominal pain presenting to 11 community EDs within an integrated delivery system between October 1, 2016, and September 30, 2018. Prior to diagnostic imaging, attending emergency physicians enrolled patients with ≤5 days of right-sided or diffuse abdominal pain using a Web-based application embedded in the electronic health record. Predicted risk (gestalt) of acute appendicitis was prospectively entered using a sliding scale from 1% to 100%. As a planned secondary analysis, we assessed the performance of gestalt via c-statistics of receiver operating characteristic (ROC) curves; tested associations between gestalt performance and patient, physician, and facility characteristics; and examined clinical characteristics affecting gestalt estimates. RESULTS: Of 3,426 patients, 334 (9.8%) had confirmed appendicitis. Physician gestalt had excellent ROC curve characteristics (c-statistic = 0.83, 95% confidence interval = 0.81 to 0.85), performing particularly well in the low-risk strata (appendicitis rate = 1.1% in gestalt 1%-10% range, negative predictive value of 98.9% for appendicitis diagnosis). Physicians with ≥5 years since medical school graduation demonstrated improved gestalt performance over those with less experience (p = 0.007). All clinical characteristics tested, except pain <24 hours, were significantly associated with physician gestalt value (p < 0.05). CONCLUSION: Physician gestalt for acute appendicitis diagnosis performed well, especially in low-risk patients and when employed by experienced physicians.

Multicenter validation of a machine-learning algorithm for 48-h all-cause mortality prediction.

In order to evaluate mortality predictions based on boosted trees, this retrospective study uses electronic medical record data from three academic health centers for inpatients 18 years or older with at least one observation of each vital sign. Predictions were made 12, 24, and 48 hours before death. Models fit to training data from each institution were evaluated using hold-out test data from the same institution, and from the other institutions. Gradient-boosted trees (GBT) were compared to regularized logistic regression (LR) predictions, support vector machine (SVM) predictions, quick Sepsis-Related Organ Failure Assessment (qSOFA), and Modified Early Warning Score (MEWS) using area under the receiver operating characteristic curve (AUROC). For training and testing GBT on data from the same institution, the average AUROCs were 0.96, 0.95, and 0.94 across institutional test sets for 12-, 24-, and 48-hour predictions, respectively. When trained and tested on data from different hospitals, GBT AUROCs achieved up to 0.98, 0.96, and 0.96, for 12-, 24-, and 48-hour predictions, respectively. Average AUROC for 48-hour predictions for LR, SVM, MEWS, and qSOFA were 0.85, 0.79, 0.86 and 0.82, respectively. GBT predictions may help identify patients who would benefit from increased clinical care.

Text message alerts to emergency physicians identifying potential study candidates increase clinical trial enrollment.

Prospective enrollment of research subjects in the fast-paced emergency department (ED) is challenging. We sought to develop a software application to increase real-time clinical trial enrollment during an ED visit. The Prospective Intelligence System for Clinical Emergency Services (PISCES) scans the electronic health record during ED encounters for preselected clinical characteristics of potentially eligible study participants and notifies the treating physician via mobile phone text alerts. PISCES alerts began 3 months into a cluster randomized trial of an electronic health record-based risk stratification tool for pediatric abdominal pain in 11 Northern California EDs. We compared aggregate enrollment before (2577 eligible patients, October 2016 to December 2016) and after (12 049 eligible patients, January 2017 to January 2018) PISCES implementation. Enrollment increased from 10.8% to 21.1% following PISCES implementations (P < .001). PISCES significantly increased study enrollment and can serve as a valuable tool to assist prospective research enrollment in the ED.

Validation of the Pediatric Appendicitis Risk Calculator (pARC) in a Community Emergency Department Setting.

STUDY OBJECTIVE: The pediatric Appendicitis Risk Calculator (pARC) is a validated clinical tool for assessing a child's probability of appendicitis. Our objective was to assess the performance of the pARC in community emergency departments (EDs) and to compare its performance with that of the Pediatric Appendicitis Score (PAS). METHODS: We conducted a prospective validation study from October 1, 2016, to April 30, 2018, in 11 community EDs serving general populations. Patients aged 5 to 20.9 years and with a chief complaint of abdominal pain and less than or equal to 5 days of right-sided or diffuse abdominal pain were eligible for study enrollment. Our primary outcome was the presence or absence of appendicitis within 7 days of the index visit. We reported performance characteristics and secondary outcomes by pARC risk strata and compared the receiver operator characteristic (ROC) curves of the PAS and pARC. RESULTS: We enrolled 2,089 patients with a mean age of 12.4 years, 46% of whom were male patients. Appendicitis was confirmed in 353 patients (16.9%), of whom 55 (15.6%) had perforated appendixes. Fifty-four percent of patients had very low (<5%) or low (5% to 14%) predicted risk, 43% had intermediate risk (15% to 84%), and 4% had high risk (≥85%). In the very-low- and low-risk groups, 1.4% and 3.0% of patients had appendicitis, respectively. The area under the ROC curve was 0.89 (95% confidence interval 0.87 to 0.92) for the pARC compared with 0.80 (95% confidence interval 0.77 to 0.82) for the PAS. CONCLUSION: The pARC accurately assessed appendicitis risk for children aged 5 years and older in community EDs and the pARC outperformed the PAS.

Development of a Clinical Decision Support System for Pediatric Abdominal Pain in Emergency Department Settings Across Two Health Systems Within the HCSRN.

BACKGROUND: Appendicitis is a common surgical emergency in children, yet diagnosis can be challenging. An electronic health record (EHR) based, clinical decision support (CDS) system called Appy CDS was designed to help guide management of pediatric patients with acute abdominal pain within the Health Care Systems Research Network (HCSRN). OBJECTIVES: To describe the development and implementation of a clinical decision support tool (Appy CDS) built independently but synergistically at two large HCSRN affiliated health systems using well-established platforms, and to assess the tool's Triage component, aiming to identify pediatric patients at increased risk for appendicitis. RESULTS: Despite differences by site in design and implementation, such as the use of alerts, incorporating gestalt, and other workflow variations across sites, using simple screening questions and automated exclusions, both systems were able to identify a population with similar appendicitis rates (11.8 percent and 10.6 percent), where use of the full Appy CDS would be indicated. DISCUSSION: These 2 HCSRN sites designed Appy CDS to capture a population at risk for appendicitis and deliver CDS to that population while remaining locally relevant and adhering to organizational preferences. Despite different approaches to point-of-care CDS, the sites have identified similar cohorts with nearly identical background rates of appendicitis. NEXT STEPS: The full Appy CDS tool, providing personalized risk assessment and tailored recommendations, is undergoing evaluation as part of a pragmatic cluster randomized trial aiming to reduce reliance on advanced diagnostic imaging. The novel approaches to CDS we present could serve as the basis for future ED interventions.

Evaluation of a machine learning algorithm for up to 48-hour advance prediction of sepsis using six vital signs.

OBJECTIVE: Sepsis remains a costly and prevalent syndrome in hospitals; however, machine learning systems can increase timely sepsis detection using electronic health records. This study validates a gradient boosted ensemble machine learning tool for sepsis detection and prediction, and compares its performance to existing methods. MATERIALS AND METHODS: Retrospective data was drawn from databases at the University of California, San Francisco (UCSF) Medical Center and the Beth Israel Deaconess Medical Center (BIDMC). Adult patient encounters without sepsis on admission, and with at least one recording of each of six vital signs (SpO2, heart rate, respiratory rate, temperature, systolic and diastolic blood pressure) were included. We compared the performance of the machine learning algorithm (MLA) to that of commonly used scoring systems. Area under the receiver operating characteristic (AUROC) curve was our primary measure of accuracy. MLA performance was measured at sepsis onset, and at 24 and 48 h prior to sepsis onset. RESULTS: The MLA achieved an AUROC of 0.88, 0.84, and 0.83 for sepsis onset and 24 and 48 h prior to onset, respectively. These values were superior to those of SIRS (0.66), MEWS (0.61), SOFA (0.72), and qSOFA (0.60) at time of onset. When trained on UCSF data and tested on BIDMC data, sepsis onset AUROC was 0.89. DISCUSSION AND CONCLUSION: The MLA predicts sepsis up to 48 h in advance and identifies sepsis onset more accurately than commonly used tools, maintaining high performance for sepsis detection when trained and tested on separate datasets.

Implementation of a Clinical Decision Support System for Children With Minor Blunt Head Trauma Who Are at Nonnegligible Risk for Traumatic Brain Injuries.

STUDY OBJECTIVE: To determine the effect of providing risk estimates of clinically important traumatic brain injuries and management recommendations on emergency department (ED) outcomes for children with isolated intermediate Pediatric Emergency Care Applied Research Network clinically important traumatic brain injury risk factors. METHODS: This was a secondary analysis of a nonrandomized clinical trial with concurrent controls, conducted at 5 pediatric and 8 general EDs between November 2011 and June 2014, enrolling patients younger than 18 years who had minor blunt head trauma. After a baseline period, intervention sites received electronic clinical decision support providing patient-level clinically important traumatic brain injury risk estimates and management recommendations. The following primary outcomes in patients with one intermediate Pediatric Emergency Care Applied Research Network risk factor were compared before and after clinical decision support: proportion of ED computed tomography (CT) scans, adjusted for age, time trend, and site; and prevalence of clinically important traumatic brain injuries. RESULTS: The risk of clinically important traumatic brain injuries was known for 3,859 children with isolated findings (1,711 at intervention sites before clinical decision support, 1,702 at intervention sites after clinical decision support, and 446 at control sites). In this group, pooled CT proportion decreased from 24.2% to 21.6% after clinical decision support (odds ratio 0.86; 95% confidence interval 0.73 to 1.01). Decreases in CT use were noted across intervention EDs, but not in controls. The pooled adjusted odds ratio for CT use after clinical decision support was 0.73 (95% confidence interval 0.60 to 0.88). Among the entire cohort, clinically important traumatic brain injury was diagnosed at the index ED visit for 37 of 37 (100%) patients before clinical decision support and 32 of 33 patients (97.0%) after clinical decision support. CONCLUSION: Providing specific risks of clinically important traumatic brain injury through electronic clinical decision support was associated with a modest and safe decrease in ED CT use for children at nonnegligible risk of clinically important traumatic brain injuries.

Increasing Safe Outpatient Management of Emergency Department Patients With Pulmonary Embolism: A Controlled Pragmatic Trial.

Background: Many low-risk patients with acute pulmonary embolism (PE) in the emergency department (ED) are eligible for outpatient care but are hospitalized nonetheless. One impediment to home discharge is the difficulty of identifying which patients can safely forgo hospitalization. Objective: To evaluate the effect of an integrated electronic clinical decision support system (CDSS) to facilitate risk stratification and decision making at the site of care for patients with acute PE. Design: Controlled pragmatic trial. (ClinicalTrials.gov: NCT03601676). Setting: All 21 community EDs of an integrated health care delivery system (Kaiser Permanente Northern California). Patients: Adult ED patients with acute PE. Intervention: Ten intervention sites selected by convenience received a multidimensional technology and education intervention at month 9 of a 16-month study period (January 2014 to April 2015); the remaining 11 sites served as concurrent controls. Measurements: The primary outcome was discharge to home from either the ED or a short-term (<24-hour) outpatient observation unit based in the ED. Adverse outcomes included return visits for PE-related symptoms within 5 days and recurrent venous thromboembolism, major hemorrhage, and all-cause mortality within 30 days. A difference-in-differences approach was used to compare pre-post changes at intervention versus control sites, with adjustment for demographic and clinical characteristics. Results: Among 881 eligible patients diagnosed with PE at intervention sites and 822 at control sites, adjusted home discharge increased at intervention sites (17.4% pre- to 28.0% postintervention) without a concurrent increase at control sites (15.1% pre- and 14.5% postintervention). The difference-in-differences comparison was 11.3 percentage points (95% CI, 3.0 to 19.5 percentage points; P = 0.007). No increases were seen in 5-day return visits related to PE or in 30-day major adverse outcomes associated with CDSS implementation. Limitation: Lack of random allocation. Conclusion: Implementation and structured promotion of a CDSS to aid physicians in site-of-care decision making for ED patients with acute PE safely increased outpatient management. Primary Funding Source: Garfield Memorial National Research Fund and The Permanente Medical Group Delivery Science and Physician Researcher Programs.

Prediction of Acute Kidney Injury With a Machine Learning Algorithm Using Electronic Health Record Data.

BACKGROUND: A major problem in treating acute kidney injury (AKI) is that clinical criteria for recognition are markers of established kidney damage or impaired function; treatment before such damage manifests is desirable. Clinicians could intervene during what may be a crucial stage for preventing permanent kidney injury if patients with incipient AKI and those at high risk of developing AKI could be identified. OBJECTIVE: In this study, we evaluate a machine learning algorithm for early detection and prediction of AKI. DESIGN: We used a machine learning technique, boosted ensembles of decision trees, to train an AKI prediction tool on retrospective data taken from more than 300 000 inpatient encounters. SETTING: Data were collected from inpatient wards at Stanford Medical Center and intensive care unit patients at Beth Israel Deaconess Medical Center. PATIENTS: Patients older than the age of 18 whose hospital stays lasted between 5 and 1000 hours and who had at least one documented measurement of heart rate, respiratory rate, temperature, serum creatinine (SCr), and Glasgow Coma Scale (GCS). MEASUREMENTS: We tested the algorithm's ability to detect AKI at onset and to predict AKI 12, 24, 48, and 72 hours before onset. METHODS: We tested AKI detection and prediction using the National Health Service (NHS) England AKI Algorithm as a gold standard. We additionally tested the algorithm's ability to detect AKI as defined by the Kidney Disease: Improving Global Outcomes (KDIGO) guidelines. We compared the algorithm's 3-fold cross-validation performance to the Sequential Organ Failure Assessment (SOFA) score for AKI identification in terms of area under the receiver operating characteristic (AUROC). RESULTS: The algorithm demonstrated high AUROC for detecting and predicting NHS-defined AKI at all tested time points. The algorithm achieves AUROC of 0.872 (95% confidence interval [CI], 0.867-0.878) for AKI detection at time of onset. For prediction 12 hours before onset, the algorithm achieves an AUROC of 0.800 (95% CI, 0.792-0.809). For 24-hour predictions, the algorithm achieves AUROC of 0.795 (95% CI, 0.785-0.804). For 48-hour and 72-hour predictions, the algorithm achieves AUROC values of 0.761 (95% CI, 0.753-0.768) and 0.728 (95% CI, 0.719-0.737), respectively. LIMITATIONS: Because of the retrospective nature of this study, we cannot draw any conclusions about the impact the algorithm's predictions will have on patient outcomes in a clinical setting. CONCLUSIONS: The results of these experiments suggest that a machine learning-based AKI prediction tool may offer important prognostic capabilities for determining which patients are likely to suffer AKI, potentially allowing clinicians to intervene before kidney damage manifests.

CONTEXTE: Une des principales difficultés liées au traitement de l'insuffisance rénale aiguë (IRA) est le fait que les critères cliniques diagnostiques sont des marqueurs d'une lésion ou d'une dysfonction rénale déjà établie. Il est souhaitable d'intervenir avant une telle issue. En dépistant les patients à risque d'IRA ou atteints d'IRA débutante, les cliniciens seraient en mesure d'intervenir précocement et ainsi prévenir les lésions rénales permanentes. OBJECTIF DE L'ÉTUDE: L'étude visait à évaluer un algorithme d'apprentissage automatique destiné à la prédiction des cas d'IRA et à sa détection précoce. TYPE D'ÉTUDE: Nous avons employé une technique d'apprentissage automatique, soit des ensembles d'arbres décisionnels amplifiés, pour entrainer un outil de prédiction de l'IRA à partir de données rétrospectives provenant de plus de 300 000 consultations auprès de patients hospitalisés. CADRE DE L'ÉTUDE: Les données ont été colligées à partir des dossiers des unités d'hospitalisation du centre médical de l'université Stanford et de l'unité des soins intensifs du centre médical Beth Israel Deaconess. PARTICIPANTS: Ont été inclus dans l'étude tous les patients adultes dont l'hospitalisation avait duré de 5 à 1 000 heures et pour lesquels on disposait d'au moins une mesure parmi les suivantes : pouls, rythme respiratoire, température corporelle, taux de créatinine sérique (SCr) et score de Glasgow. MESURES: Nous avons testé l'efficacité de l'algorithme à détecter l'IRA dès son apparition, et à la prédire 12, 24, 48 et 72 heures avant qu'elle ne se manifeste. MÉTHODOLOGIE: L'algorithme du NHS England a servi de référence pour tester l'efficacité de notre algorithme de prédiction et de détection de l'IRA. Nous avons également testé l'efficacité de notre algorithme à détecter l'IRA telle que définie par les Recommandations de Bonnes Pratiques Cliniques du KDIGO (Kidney Disease: Improving Global Outcomes). Nous avons utilisé la surface sous la courbe ROC (Receiver Operating Characteristic) pour comparer le score SOFA à l'efficacité de validation croisée tripartite de notre algorithme. RÉSULTATS: L'algorithme a démontré une SSROC (surface sous la courbe ROC) élevée pour la détection et la prédiction de l'IRA (telle que définie par le NHS) pour tous les moments testés. En détection de la maladie à son apparition, l'algorithme a obtenu une SSROC de 0,872 (IC 95 % : 0,867-0,878). En prédiction, l'algorithme a obtenu une SSROC de 0,800 (IC 95 % : 0,792-0,809) à 12 heures, de 0,795 à 24 heures (IC 95 % : 0,785-0,804), de 0,761 (IC 95 % : 0,753-0,768) à 48 heures et de 0,728 (IC 95 % : 0,719-0,737) à 72 heures avant l'apparition des premiers symptômes. LIMITES DE L'ÉTUDE: La nature rétrospective de l'étude ne nous permet pas de tirer de conclusions sur les conséquences qu'auront les prédictions de l'algorithme sur les résultats cliniques des patients. CONCLUSION: Les résultats de nos essais laissent supposer qu'un outil de prédiction de l'IRA fondé sur l'apprentissage automatique pourrait offrir d'importantes fonctions pronostiques pour détecter les patients susceptibles de développer une IRA en vue d'une intervention précoce.

Development and Validation of a Novel Pediatric Appendicitis Risk Calculator (pARC).

OBJECTIVES: We sought to develop and validate a clinical calculator that can be used to quantify risk for appendicitis on a continuous scale for patients with acute abdominal pain. METHODS: The pediatric appendicitis risk calculator (pARC) was developed and validated through secondary analyses of 3 distinct cohorts. The derivation sample included visits to 9 pediatric emergency departments between March 2009 and April 2010. The validation sample included visits to a single pediatric emergency department from 2003 to 2004 and 2013 to 2015. Variables evaluated were as follows: age, sex, temperature, nausea and/or vomiting, pain duration, pain location, pain with walking, pain migration, guarding, white blood cell count, and absolute neutrophil count. We used stepwise regression to develop and select the best model. Test performance of the pARC was compared with the Pediatric Appendicitis Score (PAS). RESULTS: The derivation sample included 2423 children, 40% of whom had appendicitis. The validation sample included 1426 children, 35% of whom had appendicitis. The final pARC model included the following variables: sex, age, duration of pain, guarding, pain migration, maximal tenderness in the right-lower quadrant, and absolute neutrophil count. In the validation sample, the pARC exhibited near perfect calibration and a high degree of discrimination (area under the curve: 0.85; 95% confidence interval: 0.83 to 0.87) and outperformed the PAS (area under the curve: 0.77; 95% confidence interval: 0.75 to 0.80). By using the pARC, almost half of patients in the validation cohort could be accurately classified as at <15% risk or ≥85% risk for appendicitis, whereas only 23% would be identified as having a comparable PAS of <3 or >8. CONCLUSIONS: In our validation cohort of patients with acute abdominal pain, the pARC accurately quantified risk for appendicitis.

Thromboprophylaxis for Patients with High-risk Atrial Fibrillation and Flutter Discharged from the Emergency Department.

INTRODUCTION: Many patients with atrial fibrillation or atrial flutter (AF/FL) who are high risk for ischemic stroke are not receiving evidence-based thromboprophylaxis. We examined anticoagulant prescribing within 30 days of receiving dysrhythmia care for non-valvular AF/FL in the emergency department (ED). METHODS: This prospective study included non-anticoagulated adults at high risk for ischemic stroke (ATRIA score ≥7) who received emergency AF/FL care and were discharged home from seven community EDs between May 2011 and August 2012. We characterized oral anticoagulant prescribing patterns and identified predictors of receiving anticoagulants within 30 days of the index ED visit. We also describe documented reasons for withholding anticoagulation. RESULTS: Of 312 eligible patients, 128 (41.0%) were prescribed anticoagulation at ED discharge or within 30 days. Independent predictors of anticoagulation included age (adjusted odds ratio [aOR] 0.89 per year, 95% confidence interval [CI] 0.82-0.96); ED cardiology consultation (aOR 1.89, 95% CI [1.10-3.23]); and failure of sinus restoration by time of ED discharge (aOR 2.65, 95% CI [1.35-5.21]). Reasons for withholding anticoagulation at ED discharge were documented in 139 of 227 cases (61.2%), the most common of which were deferring the shared decision-making process to the patient's outpatient provider, perceived bleeding risk, patient refusal, and restoration of sinus rhythm. CONCLUSION: Approximately 40% of non-anticoagulated AF/FL patients at high risk for stroke who presented for emergency dysrhythmia care were prescribed anticoagulation within 30 days. Physicians were less likely to anticoagulate older patients and those with ED sinus restoration. Opportunities exist to improve rates of thromboprophylaxis in this high-risk population.

Performance of Coronary Risk Scores Among Patients With Chest Pain in the Emergency Department.

BACKGROUND: Both the modified History, Electrocardiogram, Age, Risk factors and Troponin (HEART) score and the Emergency Department Assessment of Chest pain Score (EDACS) can identify patients with possible acute coronary syndrome (ACS) at low risk (<1%) for major adverse cardiac events (MACE). OBJECTIVES: The authors sought to assess the comparative accuracy of the EDACS (original and simplified) and modified HEART risk scores when using cardiac troponin I (cTnI) cutoffs below the 99th percentile, and obtain precise MACE risk estimates. METHODS: The authors conducted a retrospective study of adult emergency department (ED) patients evaluated for possible ACS in an integrated health care system between 2013 and 2015. Negative predictive values for MACE (composite of myocardial infarction, cardiogenic shock, cardiac arrest, and all-cause mortality) were determined at 60 days. Reclassification analyses were used to assess the comparative accuracy of risk scores and lower cTnI cutoffs. RESULTS: A total of 118,822 patients with possible ACS were included. The 3 risk scores' accuracies were optimized using the lower limit of cTnI quantitation (<0.02 ng/ml) to define low risk for 60-day MACE, with reclassification yields ranging between 3.4% and 3.9%, while maintaining similar negative predictive values (range 99.49% to 99.55%; p = 0.27). The original EDACS identified the largest proportion of patients as low risk (60.6%; p < 0.0001). CONCLUSIONS: Among ED patients with possible ACS, the modified HEART score, original EDACS, and simplified EDACS all predicted a low risk of 60-day MACE with improved accuracy using a cTnI cutoff below the 99th percentile. The original EDACS identified the most low-risk patients, and thus may be the preferred risk score.

Multicentre validation of a sepsis prediction algorithm using only vital sign data in the emergency department, general ward and ICU.

OBJECTIVES: We validate a machine learning-based sepsis-prediction algorithm (InSight) for the detection and prediction of three sepsis-related gold standards, using only six vital signs. We evaluate robustness to missing data, customisation to site-specific data using transfer learning and generalisability to new settings. DESIGN: A machine-learning algorithm with gradient tree boosting. Features for prediction were created from combinations of six vital sign measurements and their changes over time. SETTING: A mixed-ward retrospective dataset from the University of California, San Francisco (UCSF) Medical Center (San Francisco, California, USA) as the primary source, an intensive care unit dataset from the Beth Israel Deaconess Medical Center (Boston, Massachusetts, USA) as a transfer-learning source and four additional institutions' datasets to evaluate generalisability. PARTICIPANTS: 684 443 total encounters, with 90 353 encounters from June 2011 to March 2016 at UCSF. INTERVENTIONS: None. PRIMARY AND SECONDARY OUTCOME MEASURES: Area under the receiver operating characteristic (AUROC) curve for detection and prediction of sepsis, severe sepsis and septic shock. RESULTS: For detection of sepsis and severe sepsis, InSight achieves an AUROC curve of 0.92 (95% CI 0.90 to 0.93) and 0.87 (95% CI 0.86 to 0.88), respectively. Four hours before onset, InSight predicts septic shock with an AUROC of 0.96 (95% CI 0.94 to 0.98) and severe sepsis with an AUROC of 0.85 (95% CI 0.79 to 0.91). CONCLUSIONS: InSight outperforms existing sepsis scoring systems in identifying and predicting sepsis, severe sepsis and septic shock. This is the first sepsis screening system to exceed an AUROC of 0.90 using only vital sign inputs. InSight is robust to missing data, can be customised to novel hospital data using a small fraction of site data and retains strong discrimination across all institutions.

Using Transfer Learning for Improved Mortality Prediction in a Data-Scarce Hospital Setting.

Algorithm-based clinical decision support (CDS) systems associate patient-derived health data with outcomes of interest, such as in-hospital mortality. However, the quality of such associations often depends on the availability of site-specific training data. Without sufficient quantities of data, the underlying statistical apparatus cannot differentiate useful patterns from noise and, as a result, may underperform. This initial training data burden limits the widespread, out-of-the-box, use of machine learning-based risk scoring systems. In this study, we implement a statistical transfer learning technique, which uses a large "source" data set to drastically reduce the amount of data needed to perform well on a "target" site for which training data are scarce. We test this transfer technique with AutoTriage, a mortality prediction algorithm, on patient charts from the Beth Israel Deaconess Medical Center (the source) and a population of 48 249 adult inpatients from University of California San Francisco Medical Center (the target institution). We find that the amount of training data required to surpass 0.80 area under the receiver operating characteristic (AUROC) on the target set decreases from more than 4000 patients to fewer than 220. This performance is superior to the Modified Early Warning Score (AUROC: 0.76) and corresponds to a decrease in clinical data collection time from approximately 6 months to less than 10 days. Our results highlight the usefulness of transfer learning in the specialization of CDS systems to new hospital sites, without requiring expensive and time-consuming data collection efforts.

Cost and mortality impact of an algorithm-driven sepsis prediction system.

AIMS: To compute the financial and mortality impact of InSight, an algorithm-driven biomarker, which forecasts the onset of sepsis with minimal use of electronic health record data. METHODS: This study compares InSight with existing sepsis screening tools and computes the differential life and cost savings associated with its use in the inpatient setting. To do so, mortality reduction is obtained from an increase in the number of sepsis cases correctly identified by InSight. Early sepsis detection by InSight is also associated with a reduction in length-of-stay, from which cost savings are directly computed. RESULTS: InSight identifies more true positive cases of severe sepsis, with fewer false alarms, than comparable methods. For an individual ICU with 50 beds, for example, it is determined that InSight annually saves 75 additional lives and reduces sepsis-related costs by $560,000. LIMITATIONS: InSight performance results are derived from analysis of a single-center cohort. Mortality reduction results rely on a simplified use case, which fixes prediction times at 0, 1, and 2 h before sepsis onset, likely leading to under-estimates of lives saved. The corresponding cost reduction numbers are based on national averages for daily patient length-of-stay cost. CONCLUSIONS: InSight has the potential to reduce sepsis-related deaths and to lead to substantial cost savings for healthcare facilities.