Machine learning for the diagnosis of acute coronary syndrome using a 12-lead ECG: a systematic review.

OBJECTIVES: Prompt diagnosis of acute coronary syndrome (ACS) using a 12-lead electrocardiogram (ECG) is a critical task for emergency physicians. While computerized algorithms for ECG interpretation are limited in their accuracy, machine learning (ML) models have shown promise in several areas of clinical medicine. We performed a systematic review to compare the performance of ML-based ECG analysis to clinician or non-ML computerized ECG interpretation in the diagnosis of ACS for emergency department (ED) or prehospital patients. METHODS: We searched Medline, Embase, Cochrane Central, and CINAHL databases from inception to May 18, 2022. We included studies that compared ML algorithms to either clinicians or non-ML based software in their ability to diagnose ACS using only a 12-lead ECG, in adult patients experiencing chest pain or symptoms concerning for ACS in the ED or prehospital setting. We used QUADAS-2 for risk of bias assessment. Prospero registration CRD42021264765. RESULTS: Our search yielded 1062 abstracts. 10 studies met inclusion criteria. Five model types were tested, including neural networks, random forest, and gradient boosting. In five studies with complete performance data, ML models were more sensitive but less specific (sensitivity range 0.59-0.98, specificity range 0.44-0.95) than clinicians (sensitivity range 0.22-0.93, specificity range 0.63-0.98) in diagnosing ACS. In four studies that reported it, ML models had better discrimination (area under ROC curve range 0.79-0.98) than clinicians (area under ROC curve 0.67-0.78). Heterogeneity in both methodology and reporting methods precluded a meta-analysis. Several studies had high risk of bias due to patient selection, lack of external validation, and unreliable reference standards for ACS diagnosis. CONCLUSIONS: ML models have overall higher discrimination and sensitivity but lower specificity than clinicians and non-ML software in ECG interpretation for the diagnosis of ACS. ML-based ECG interpretation could potentially serve a role as a "safety net", alerting emergency care providers to a missed acute MI when it has not been diagnosed. More rigorous primary research is needed to definitively demonstrate the ability of ML to outperform clinicians at ECG interpretation.

RéSUMé: OBJECTIFS: Le diagnostic rapide du syndrome coronarien aigu (SCA) à l'aide d'un électrocardiogramme à 12 dérivations (ECG) est une tâche essentielle pour les urgentologues. Bien que la précision des algorithmes informatisés pour l'interprétation de l'ECG soit limitée, les modèles d'apprentissage automatique (ML) se sont révélés prometteurs dans plusieurs domaines de la médecine clinique. Nous avons effectué une revue systématique pour comparer la performance de l'analyse ECG basée sur le ML à l'interprétation ECG informatisée clinicienne ou non-ML dans le diagnostic du SCA pour les urgences (ED) ou les patients préhospitaliers. MéTHODES: Nous avons effectué des recherches dans les bases de données Medline, Embase, Cochrane Central et CINAHL de la création au 18 mai 2022. Nous avons inclus des études qui comparaient les algorithmes de ML à des cliniciens ou à des logiciels non basés sur ML dans leur capacité à diagnostiquer le SCA en utilisant uniquement un ECG à 12 dérivations, chez des patients adultes présentant des douleurs thoraciques ou des symptômes concernant le SCA dans le cadre de l'urgence ou préhospitalier. Nous avons utilisé QUADAS-2 pour l'évaluation du risque de biais. Prospero registration CRD42021264765. RéSULTATS: Notre recherche a donné 1062 résumés. 10 études satisfaisaient aux critères d'inclusion. Cinq types de modèles ont été testés, dont les réseaux neuronaux, la forêt aléatoire et le gradient boosting. Dans cinq études avec des données de performance complètes, les modèles de ML étaient plus sensibles mais moins spécifiques (plage de sensibilité 0,59-0,98, plage de spécificité 0,44-0,95) que les cliniciens (plage de sensibilité 0,22-0,93, plage de spécificité 0,63-0,98) dans le diagnostic du SCA. Dans quatre études qui l'ont rapporté, les modèles de ML avaient une meilleure discrimination (zone sous la courbe ROC plage 0,79-0,98) que les cliniciens (zone sous la courbe ROC 0,67-0,78). L'hétérogénéité de la méthodologie et des méthodes de déclaration a empêché une méta-analyse. Plusieurs études présentaient un risque élevé de biais en raison de la sélection des patients, du manque de validation externe et de normes de référence peu fiables pour le diagnostic du SCA. CONCLUSIONS: Les modèles de ML ont globalement une discrimination et une sensibilité plus élevées mais une spécificité plus faible que les cliniciens et les logiciels non-ML dans l'interprétation de l'ECG pour le diagnostic du SCA. L'interprétation de l'ECG basée sur le ML pourrait servir de « filet de sécurité ¼, alertant les fournisseurs de soins d'urgence d'une IM aiguë manquée lorsqu'elle n'a pas été diagnostiquée. Des recherches primaires plus rigoureuses sont nécessaires pour démontrer définitivement la capacité du ML à surpasser les cliniciens lors de l'interprétation de l'ECG.

Initial Diagnosis and Management of Acutely Elevated Intracranial Pressure.

Acutely elevated intracranial pressure (ICP) may have devastating effects on patient mortality and neurologic outcomes, yet its initial detection remains difficult because of the variety of manifestations that it can cause disease states it is associated with. Several treatment guidelines exist for specific disease processes such as trauma or ischemic stroke, but their recommendations may not apply to other causes. In the acute setting, management decisions must often be made before the underlying cause is known. In this review, we present an organized, evidence-based approach to the recognition and management of patients with suspected or confirmed elevated ICP in the first minutes to hours of resuscitation. We explore the utility of invasive and noninvasive methods of diagnosis, including history, physical examination, imaging, and ICP monitors. We synthesize various guidelines and expert recommendations and identify core management principles including noninvasive maneuvers, neuroprotective intubation and ventilation strategies, and pharmacologic therapies such as ketamine, lidocaine, corticosteroids, and the hyperosmolar agents mannitol and hypertonic saline. Although an in-depth discussion of the definitive management of each etiology is beyond the scope of this review, our goal is to provide an empirical approach to these time-sensitive, critical presentations in their initial stages.

Machine Learning Versus Usual Care for Diagnostic and Prognostic Prediction in the Emergency Department: A Systematic Review.

OBJECTIVE: Having shown promise in other medical fields, we sought to determine whether machine learning (ML) models perform better than usual care in diagnostic and prognostic prediction for emergency department (ED) patients. METHODS: In this systematic review, we searched MEDLINE, Embase, Central, and CINAHL from inception to October 17, 2019. We included studies comparing diagnostic and prognostic prediction of ED patients by ML models to usual care methods (triage-based scores, clinical prediction tools, clinician judgment) using predictor variables readily available to ED clinicians. We extracted commonly reported performance metrics of model discrimination and classification. We used the PROBAST tool for risk of bias assessment (PROSPERO registration: CRD42020158129). RESULTS: The search yielded 1,656 unique records, of which 23 studies involving 16,274,647 patients were included. In all seven diagnostic studies, ML models outperformed usual care in all performance metrics. In six studies assessing in-hospital mortality, the best-performing ML models had better discrimination (area under the receiver operating characteristic curve [AUROC] =0.74-0.94) than any clinical decision tool (AUROC =0.68-0.81). In four studies assessing hospitalization, ML models had better discrimination (AUROC =0.80-0.83) than triage-based scores (AUROC =0.68-0.82). Clinical heterogeneity precluded meta-analysis. Most studies had high risk of bias due to lack of external validation, low event rates, and insufficient reporting of calibration. CONCLUSIONS: Our review suggests that ML may have better prediction performance than usual care for ED patients with a variety of clinical presentations and outcomes. However, prediction model reporting guidelines should be followed to provide clinically applicable data. Interventional trials are needed to assess the impact of ML models on patient-centered outcomes.

Adherence to the Standards for Reporting of Diagnostic Accuracy (STARD) 2015 Guidelines in Acute Point-of-Care Ultrasound Research.

Importance: Incomplete reporting of diagnostic accuracy research impairs assessment of risk of bias and limits generalizability. Point-of-care ultrasound has become an important diagnostic tool for acute care physicians, but studies assessing its use are of varying methodological quality. Objective: To assess adherence to the Standards for Reporting of Diagnostic Accuracy (STARD) 2015 guidelines in the literature on acute care point-of-care ultrasound. Evidence Review: MEDLINE was searched to identify diagnostic accuracy studies assessing point-of-care ultrasound published in critical care, emergency medicine, or anesthesia journals from 2016 to 2019. Studies were evaluated for adherence to the STARD 2015 guidelines, with the following variables analyzed: journal, country, STARD citation, STARD-adopting journal, impact factor, patient population, use of supplemental material, and body region. Data analysis was performed in November 2019. Findings: Seventy-four studies were included in this systematic review for assessment. Overall adherence to STARD was moderate, with 66% (mean [SD], 19.7 [2.9] of 30 items) of STARD items reported. Items pertaining to imaging specifications, patient population, and readers of the index test were frequently reported (>66% of studies). Items pertaining to blinding of readers to clinical data and to the index or reference standard, analysis of heterogeneity, indeterminate and missing data, and time intervals between index and reference test were either moderately (33%-66%) or infrequently (<33%) reported. Studies in STARD-adopting journals (mean [SD], 20.5 [2.9] items in adopting journals vs 18.6 [2.3] items in nonadopting journals; P = .002) and studies citing STARD (mean [SD], 21.3 [0.9] items in citing studies vs 19.5 [2.9] items in nonciting studies; P = .01) reported more items. Variation by country and journal of publication were identified. No differences in STARD adherence were identified by body region imaged (mean [SD], abdominal, 20.0 [2.5] items; head and neck, 17.8 [1.6] items; musculoskeletal, 19.2 [3.1] items; thoracic, 20.2 [2.8] items; and other or procedural, 19.8 [2.7] items; P = .29), study design (mean [SD], prospective, 19.7 [2.9] items; retrospective, 19.7 [1.8] items; P > .99), patient population (mean [SD], pediatric, 20.0 [3.1] items; adult, 20.2 [2.7] items; mixed, 17.9 [1.9] items; P = .09), use of supplementary materials (mean [SD], yes, 19.2 [3.0] items; no, 19.7 [2.8] items; P = .91), or journal impact factor (mean [SD], higher impact factor, 20.3 [3.1] items; lower impact factor, 19.1 [2.4] items; P = .08). Conclusions and Relevance: Overall, the literature on acute care point-of-care ultrasound showed moderate adherence to the STARD 2015 guidelines, with more complete reporting found in studies citing STARD and those published in STARD-adopting journals. This study has established a current baseline for reporting; however, future studies are required to understand barriers to complete reporting and to develop strategies to mitigate them.

Clinical Examination for Acute Aortic Dissection: A Systematic Review and Meta-analysis.

OBJECTIVES: Acute aortic dissection is a life-threatening condition due to a tear in the aortic wall. It is difficult to diagnose and if missed carries a significant mortality. METHODS: We conducted a librarian-assisted systematic review of PubMed, MEDLINE, Embase, and the Cochrane database from 1968 to July 2016. Titles and abstracts were reviewed and data were extracted by two independent reviewers (agreement measured by kappa). Studies were combined if low clinical and statistical heterogeneity (I2  < 30%). Study quality was assessed using the QUADAS-2 tool. Bivariate random effects meta analyses using Revman 5 and SAS 9.3 were performed. RESULTS: We identified 792 records: 60 were selected for full text review, nine studies with 2,400 participants were included (QUADAS-2 low risk of bias, κ = 0.89 [for full-text review]). Prevalence of aortic dissection ranged from 21.9% to 76.1% (mean ± SD = 39.1% ± 17.1%). The clinical findings increasing probability of aortic dissection were 1) neurologic deficit (n = 3, specificity = 95%, positive likelihood ratio [LR+] = 4.4, 95% confidence interval [CI] = 3.3-5.7, I2  = 0%) and 2) hypotension (n = 4, specificity = 95%, LR+ = 2.9 95% CI = 1.8-4.6, I2  = 42%), and decreasing probability were the absence of a widened mediastinum (n = 4, sensitivity = 76%-95%, negative likelihood ratio [LR-] = 0.14-0.60, I2  = 93%) and an American Heart Association aortic dissection detection (AHA ADD) risk score < 1 (n = 1, sensitivity = 91%, LR- = 0.22, 95% CI = 0.15-0.33). CONCLUSIONS: Suspicion for acute aortic dissection should be raised with hypotension, pulse, or neurologic deficit. Conversely, a low AHA ADD score decreases suspicion. Clinical gestalt informed by high- and low-risk features together with an absence of an alternative diagnosis should drive investigation for acute aortic dissection.

Incidence of Appendicitis over Time: A Comparative Analysis of an Administrative Healthcare Database and a Pathology-Proven Appendicitis Registry.

IMPORTANCE: At the turn of the 21st century, studies evaluating the change in incidence of appendicitis over time have reported inconsistent findings. OBJECTIVES: We compared the differences in the incidence of appendicitis derived from a pathology registry versus an administrative database in order to validate coding in administrative databases and establish temporal trends in the incidence of appendicitis. DESIGN: We conducted a population-based comparative cohort study to identify all individuals with appendicitis from 2000 to2008. SETTING & PARTICIPANTS: Two population-based data sources were used to identify cases of appendicitis: 1) a pathology registry (n = 8,822); and 2) a hospital discharge abstract database (n = 10,453). INTERVENTION & MAIN OUTCOME: The administrative database was compared to the pathology registry for the following a priori analyses: 1) to calculate the positive predictive value (PPV) of administrative codes; 2) to compare the annual incidence of appendicitis; and 3) to assess differences in temporal trends. Temporal trends were assessed using a generalized linear model that assumed a Poisson distribution and reported as an annual percent change (APC) with 95% confidence intervals (CI). Analyses were stratified by perforated and non-perforated appendicitis. RESULTS: The administrative database (PPV = 83.0%) overestimated the incidence of appendicitis (100.3 per 100,000) when compared to the pathology registry (84.2 per 100,000). Codes for perforated appendicitis were not reliable (PPV = 52.4%) leading to overestimation in the incidence of perforated appendicitis in the administrative database (34.8 per 100,000) as compared to the pathology registry (19.4 per 100,000). The incidence of appendicitis significantly increased over time in both the administrative database (APC = 2.1%; 95% CI: 1.3, 2.8) and pathology registry (APC = 4.1; 95% CI: 3.1, 5.0). CONCLUSION & RELEVANCE: The administrative database overestimated the incidence of appendicitis, particularly among perforated appendicitis. Therefore, studies utilizing administrative data to analyze perforated appendicitis should be interpreted cautiously.

Incidence of primary sclerosing cholangitis: a systematic review and meta-analysis.

UNLABELLED: Incidence studies of primary sclerosing cholangitis (PSC) are important for describing the disease's burden and for shedding light on the disease's etiology. The purposes of this study were to conduct a systematic review of the incidence studies of PSC with a meta-analysis and to investigate possible geographic variations and temporal trends in the incidence of the disease. A systematic literature search of MEDLINE (1950-2010) and Embase (1980-2010) was conducted to identify studies investigating the incidence of PSC. The incidence of PSC was summarized with an incidence rate (IR) and 95% confidence intervals. The test of heterogeneity was performed with the Q statistic. Secondary variables extracted from the articles included the following: the method of case ascertainment, the country, the time period, the age, the male/female incidence rate ratio (IRR), and the incidence of PSC subtypes (small-duct or large-duct PSC and inflammatory bowel disease). Stratified and sensitivity analyses were performed to explore heterogeneity between studies and to assess effects of study quality. Time trends were used to explore differences in the incidence across time. The search retrieved 1669 potentially eligible citations; 8 studies met the inclusion criteria. According to a random-effects model, the pooled IR was 0.77 (0.45-1.09) per 100,000 person-years. However, significant heterogeneity was observed between studies (P < 0.001). Sensitivity analyses excluding non-population-based studies increased the overall IR to 1.00 (0.82-1.17) and eliminated the heterogeneity between studies (P = 0.08). The IRR for males versus females was 1.70 (1.34-2.07), and the median age was 41 years (35-47 years). All studies investigating time trends reported an overall increase in the incidence of PSC. CONCLUSION: The incidence of PSC is similar in North American and European countries and continues to increase over time. Incidence data from developing countries are lacking, and this limits our understanding of the global incidence of PSC.