Incorporation of Serial 12-Lead Electrocardiogram With Machine Learning to Augment the Out-of-Hospital Diagnosis of Non-ST Elevation Acute Coronary Syndrome

Abstract

Study objective: Ischemic electrocardiogram (ECG) changes are subtle and transient in patients with suspected non-ST-segment elevation (NSTE)-acute coronary syndrome. However, the out-of-hospital ECG is not routinely used during subsequent evaluation at the emergency department. Therefore, we sought to compare the diagnostic performance of out-of-hospital and ED ECG and evaluate the incremental gain of artificial intelligence-augmented ECG analysis.

Methods: This prospective observational cohort study recruited patients with out-of-hospital chest pain. We retrieved out-of-hospital-ECG obtained by paramedics in the field and the first ED ECG obtained by nurses during inhospital evaluation. Two independent and blinded reviewers interpreted ECG dyads in mixed order per practice recommendations. Using 179 morphological ECG features, we trained, cross-validated, and tested a random forest classifier to augment non ST-elevation acute coronary syndrome (NSTE-ACS) diagnosis.

Results: Our sample included 2,122 patients (age 59 [16]; 53% women; 44% Black, 13.5% confirmed acute coronary syndrome). The rate of diagnostic ST elevation and ST depression were 5.9% and 16.2% on out-of-hospital-ECG and 6.1% and 12.4% on ED ECG, with ∼40% of changes seen on out-of-hospital-ECG persisting and ∼60% resolving. Using expert interpretation of out-of-hospital-ECG alone gave poor baseline performance with area under the receiver operating characteristic (AUC), sensitivity, and negative predictive values of 0.69, 0.50, and 0.92. Using expert interpretation of serial ECG changes enhanced this performance (AUC 0.80, sensitivity 0.61, and specificity 0.93). Interestingly, augmenting the out-of-hospital-ECG alone with artificial intelligence algorithms boosted its performance (AUC 0.83, sensitivity 0.75, and specificity 0.95), yielding a net reclassification improvement of 29.5% against expert ECG interpretation.

Conclusion: In this study, 60% of diagnostic ST changes resolved prior to hospital arrival, making the ED ECG suboptimal for the inhospital evaluation of NSTE-ACS. Using serial ECG changes or incorporating artificial intelligence-augmented analyses would allow correctly reclassifying one in 4 patients with suspected NSTE-ACS.

Trial registration: ClinicalTrials.gov NCT04237688.

Figure 1:. The relationship between ischemic ECG findings and acute coronary syndrome

This figure shows how diagnostic ST changes correlated with acute coronary syndrome (ACS) on the prehospital (PH)-ECG (A), emergency department (ED)-ECG (B), and serial dynamic changes between both ECGs (C). ECG changes included diagnostic ST elevation or ST depression interpreted retrospectively by independent reviewers as per the 4^th universal definition of MI guidelines. We excluded from this analysis patients with prehospital CATH laboratory activation for suspected STE-ACS identified in the field by paramedics. Area under ROC (AUROC) curves are based on a logistic regression classifier using the ST changes seen on each ECGs or their dynamic patterns. SEN: sensitivity, SPE: specificity. AUROC, SEN, and SPE are reported as value (95% CI).

Figure 2:. Classification performance of NSTE-ACS using AI-augmented ECG analysis

This figure shows random forest classification performance using features from prehospital ECG (AI-PH-ECG) or the emergency department (AI-ED-ECG) as compared to clinical practice based on ED evaluation (CP-ED-ECG) on both training subset (left) and testing subset (right). The tables show the diagnostic accuracy measures and the net reclassification performance (NRI) index as compared to CP-ED-ECG as a reference standard (Ref).

Figure 3:. Classification performance of AI-augmented ECG analysis supplemented by serial ECG and clinical data

This figure shows the baseline classification performance of random forest model using features from prehospital ECG (AI-PH-ECG), both prehospital and ED ECGs (AI-serial-ECG), and serial ECG plus clinical data typically available during triage (AI-ECG-Clinical) on both training subset (left) and testing subset (right). This figure demonstrates that AI augmented ECG analysis reaches its classification performance plateau with PH-ECG alone, with no additional gain in performance when adding serial ECG or any other clinical data elements. In the training set, the lighter lines correspond to the results obtained for the individual folds during the 10-fold cross-validation, whereas the thicker lines correspond to the mean results for each model. The shaded areas highlight the space englobing all curves within 2 standard error around the mean curves.

Figure 4:. Correlation between the most important ECG features in the diagnosis of acute coronary syndrome

Plot A shows mean group differences in T_peak−T_end interval (left) and QRS−T angle (right) on prehospital (PH)-ECG and emergency department (ED)-ECG in those with or without acute coronary syndrome (ACS). Plot B shows the 3-D scatterplot of the three most important features in the random forest delineating a non-linear hyperplane of ACS cases characterized by prolonged global T_peak−T_end interval, ST elevation in lead III, and distorted ST-segment (elevation or depression) in lead aVL.