Search

As healthcare professionals we want to provide patients the best possible medical care. Prehospital care, like the rest of medicine, is constantly changing, since the research performed today will shape [...]

Introduction:Emergency nurses triage nearly 137 million patients every year. Chest pain is a common complaint at the emergency department but it frequently lacks specificity, which makes the recognition of acute coronary syndrome (ACS) a very challenging and tedious task. In fact, nurses? accuracy in recognizing ACS at triage is only 45%. We tested whether machine learning models could identify chest pain patients at greatest risk for adverse cardiac events using patient features available at the time of nurse triage.Methods:We used already-existing data from EMPIRE study, which is a prospective observational cohort study that enrolled patients with prehospital chest pain transported to three tertiary care hospitals in the U.S. Independent reviewers extracted 30 patient factors available at nurse triage from hospital electronic health records: demographics, past medical history, initial vital signs, symptomatology, and automated interpretation of 12-lead ECG. The primary outcome was 30-day major adverse cardiac events (MACE) defined as: ACS, coronary revascularization, pulmonary embolus, fatal ventricular arrhythmia, cardiogenic shock, cardiac arrest or all-cause death. We tested and validated the performance of three machine leaning algorithms to predict the study outcome: logistic regression machine, classification tree, and random forest classifiers. Algorithm training was done using 10-fold cross-validation methods on 60% of the sample, and we then used the remaining 40% as a testing set. We computed the area under the ROC curve (AUC) for the best model.Results:The sample included 750 patients (age 59?17; 42% females, 40% Black). Overall, there was a total of 145 (19%) MACE events of which 90% occurred during the indexed admission. Using the 30 patient factors that were available at triage, the best classification performance on the testing test was obtained using the logistic regression machine model (AUROC = 0.72, F1 score = 0.63, accuracy = 0.79 [0.73-0.84]).Conclusions:Predictive machine learning modeling can improve nurses? accuracy in triaging high-risk chest pain patients by nearly 35%. Future work can test model performance in prospective patients with undifferentiated chest pain.

Introduction:A history of cancer among patients evaluated at the emergency department (ED) for cardiovascular complaints warrants a high degree of clinical suspicion. The HEART score is widely used for triaging chest pain at the ED. However, it is unclear whether HEART score can adequately triage chest pain events in cancer survivors. Therefore, we sought to estimate the prevalence of cancer in patients evaluated for chest pain and to compare the classification performance of the HEART score in predicting major adverse cardiac events (MACE) in patients with or without a history of cancer.Methods:This was a secondary analysis of a prospective observational cohort study of patients with chest pain and transported via ambulance to three tertiary care hospitals in the U.S. History of cancer was abstracted from the electronic health records by independent reviewers. The presence of MACE within 30 days of indexed admission was adjudicated by two independent reviewers. HEART scores were retrospectively computed from charts by a reviewer blinded to primary study outcome. Area under the ROC curve (AUC) was used to compare the classification performance of HEART score in patients with or without a history of cancer.Results:The sample included 750 patients (age 59?17; 42% females, 40% Black). A total of 69 patients (9.1%) had a history of cancer. Overall, 145 MACE events (19%) occurred during the 30-day follow-up period. There was no significant difference in the incidence of MACE between those with and without cancer history (18.1% vs. 17.4%, p=0.89). Patients with a history of cancer had significantly higher HEART score at the ED (4.6?1.8 vs. 3.9?2.0, p=0.006), and were more likely to be triaged as intermediate-to-high risk (p<0.05). HEART score performed equally well with no difference in predicting MACE in patients with or without a history of cancer (AUC=0.86 vs. 0.84, p= 0.76).Conclusions:Clinicians can use HEART score as an adequate tool to triage acute chest pain patients with regard to the presence or absence of a history of cancer.

Introduction:Oxygen has been historically used for the initial management of acute myocardial infarction (MI). A recent meta-analysis of randomized clinical trials showed that oxygen use at high-flow rate (6-8 L/min) is not associated with adverse cardiac events in patients with acute MI. However, these clinical trials primarily enrolled patients with ST-elevation MI. Evidence regarding the safety of oxygen use in patients with non-ST elevation MI, especially at the typical low-flow rate (<6 L/min), remains lacking.Methods:This was a secondary analysis of an observational prospective cohort study that enrolled consecutive patients with suspected acute MI transported by ambulance to three tertiary care hospitals in the US. Clinical data were obtained from charts. The presence of acute MI was adjudicated by two independent reviewers according to the 4thUniversal Definition guidelines. Oxygen administration during prehospital transport or during the emergency department stay was used as the independent variable. The primary outcome was the incidence of major adverse cardiac events (MACE) within 30 days of indexed admission. The secondary outcome was infarct size measured by peak troponin level during the acute phase of MI.Results:Among the 2,065 patients enrolled in the parent study, a total of 154 patients (7.5%) had confirmed acute MI (age 64?16, 41% females, 29% Blacks). Although only 7% of patients had low oxygen saturation <94%; nearly two thirds (64%) received oxygen during their initial management (median flow-rate 3 L/min [IQR 2-6]). During 30-day follow up, 48 patients (31%) experienced MACE. The incidence of MACE was not different between those with or without oxygen therapy in either ST elevation (28% vs 35%, p=0.54) or non-ST elevation (38% vs 31%, p=0.37) MI. In multivariate linear regression, oxygen use was an independent predictor of infarct size in patients with non-ST elevation MI (?=0.93, p=0.002); but not in those with ST elevation MI (?=0.14, p=0.238).Conclusions:In this prehospital cohort, low flow oxygen at 2-6 L/min was not associated with excess risk of 30-day MACE in acute ST-elevation MI, but there is an association with secondary myocardial injury in those with non-ST elevation MI that requires further investigation.

Introduction:QTc interval prolongation is a well-established marker of adverse cardiac events in a wide range of clinical and population-based populations. This interval, however, has been found to be non-specific. Recent work has demonstrated that QTc subintervals (QRS, J-Tpeak, and Tpeak-Tend) can resemble more specific cardiac ion channel pathways. We sought to evaluate and compare the clinical significance of QTc subintervals in patients with suspected acute coronary syndrome (ACS).Methods:This was a secondary analysis of a prospective, observational, cohort study that enrolled consecutive chest pain patients transported by ambulance to three hospitals in the U.S. We retrieved the prehospital 12-lead ECGs obtained by paramedics and completed standard ECG signal preprocessing using manufacturer-specific software. QTc subintervals were computed using a previously-validated algorithm (Philips Healthcare). Relevant clinical data were obtained from charts. The presence of 30-day major adverse cardiac events (MACE) was adjudicated by two independent reviewers. MACE consisted of ACS, cardiac arrest, all-cause death, acute heart failure, cardiogenic shock, pulmonary embolism, or re-infarction.Results:Our sample included 750 patients (age 59?17; 42% females, 40% Black). A total of 145 patients (19.3%) experienced MACE during the 30-day follow up. Overall, there was no difference between those with or without MACE in term of global QTc interval (382?75 vs. 374?65, p=0.204). When examining the specific subintervals, those with MACE had wider QRS duration (104?26 vs. 97?21, p=0.002) and Tpeak-Tendinterval (102?28 vs. 92?28, p<0.001), but smaller J-Tpeakinterval (247?37 vs. 262?33, p<0.001). Using previously published cutoff value of 100 ms for Tpeak-Tendinterval, the odds of 30-day MACE was nearly three folds (OR=2.9 [95% CI 2.0-4.1]).Conclusions:In this cohort of prehospital chest pain, Tpeak-Tendinterval seems to contain the most prognostic information. A similar QTc interval between groups but shortened J-Tpeakinterval in the high-risk group suggest that the resulting global repolarization dispersion is mediated through a cardiac ion multichannel pathway rather than a selective hREG potassium channel block.