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

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.

Improving reliability for measuring ST deviation in suspected acute coronary syndrome.

BACKGROUND: Standard 12­lead ECG is used for diagnosis and risk stratification in suspected acute coronary syndrome (ACS) patients. Artifacts have significant impact on the measuring quality, which consequently affect the diagnostic decision. We used a signal quality indicator (SQI) to identify the ECG segments with lower artifact levels which we hypothesized would improve ST measurements. METHODS: The Staff III 12­lead ECG database was used with the ECG segments before balloon inflation (n = 185). SQI scores per second were calculated and a 10-s ECG segment with least noise and artifacts (Clean10) was identified for each minute of recording. The first 10 s of ECG recordings (First10) for each minute were selected as a reference. The Philips DXL™ algorithm was used to measure the ST levels at J-point, +20 ms, +40 ms, +60 ms, and + 80 ms after the J-point. Standard deviations (SDs) for the ST measurements for each of the 185 ECG records were calculated for the Clean10 and for the First10 across records. The resulting SDs for the Clean10 were compared with the SDs for the First10 using the Wilcoxon signed rank test. RESULTS: The results indicated that 1) The SDs for the Clean10 are lower than that of the First10; 2) The SDs for J+20 ms and J+40 ms are lowest among the 5 different measuring points although similar improvement for the Clean10 over the First10 is observed for J+60 ms and J+80 ms as well; 3) The improvement at the J-point was not as high as other ST measurements. CONCLUSIONS: The SQI is demonstrated as an efficient tool to identify the ECG segments with lower artifacts that produce more consistent and reliable ST measurement. The measurements at J+20 ms demonstrated the highest consistency among the five studied measuring points.

Examining the ability of the Cancer and Aging Research Group tool to predict toxicity in older men receiving chemotherapy or androgen-receptor-targeted therapy for metastatic castration-resistant prostate cancer.

BACKGROUND: Because multiple treatments are available for metastatic castrate-resistant prostate cancer (mCRPC) and most patients are elderly, the prediction of toxicity risk is important. The Cancer and Aging Research Group (CARG) tool predicts chemotherapy toxicity in older adults with mixed solid tumors, but has not been validated in mCRPC. In this study, its ability to predict toxicity risk with docetaxel chemotherapy (CHEMO) was validated, and its utility was examined in predicting toxicity risk with abiraterone or enzalutamide (A/E) among older adults with mCRPC. METHODS: Men aged 65+ years were enrolled in a prospective observational study at 4 Canadian academic cancer centers. All clinically relevant grade 2 to 5 toxicities over the course of treatment were documented via structured interviews and chart review. Logistic regression was used to identify predictors of toxicity. RESULTS: Seventy-one men starting CHEMO (mean age, 73 years) and 104 men starting A/E (mean age, 76 years) were included. Clinically relevant grade 3+ toxicities occurred in 56% and 37% of CHEMO and A/E patients, respectively. The CARG tool was predictive of grade 3+ toxicities with CHEMO, which occurred in 36%, 67%, and 91% of low, moderate, and high-risk groups (P = .003). Similarly, grade 3+ toxicities occurred among A/E users in 23%, 48%, and 86% with low, moderate, and high CARG risk (P < .001). However, it was not predictive of grade 2 toxicities with either treatment. CONCLUSIONS: There is external validation of the CARG tool in predicting grade 3+ toxicity in older men with mCRPC undergoing CHEMO and demonstrated utility during A/E therapy. This may aid with treatment decision-making.

Big data reveals insights for lead importance in ECG interpretation.

BACKGROUND: Not every lead contributes equally in the interpretation of an ECG. There are some abnormalities in which the lead importance is not clear either from cardiac electrophysiology or experience. Therefore, it is beneficial to develop an algorithm to quantify the lead importance in the reading of ECGs, namely to determine how much to weigh the evidence from each individual lead when interpreting ECG. METHODS: One representative beat per ECG lead was constructed for each ECG in a database. An algorithm was developed to find the top K (K = 1, 5, 10, 20, 50, 100) ECGs in the database that had the most similar morphology to the query ECG, independently for each lead. For each lead, the query ECG was interpreted based on the weighted average voting on the most similar ECGs by applying a variety of thresholds. For each category of abnormality, we found the threshold that maximized the median F1 score of sensitivity and positive predictive value among all ECG leads. Finally, the F1 score of each lead at this chosen threshold was defined as the importance value for that lead. RESULTS: Eighteen morphology-based categories of abnormality were investigated for two databases. For most, the lead importance confirmed what expert ECG readers already know. However, it also revealed new insights. For example, lead aVR appeared in the top 6 most important leads in 11 and 12 categories of abnormality in two databases respectively, and ranked first among 12 leads if summarizing all categories. CONCLUSIONS: Lead importance information may be useful in selecting only the most important leads to screen for a specific abnormality, for example using wearable patches.

ECG reading differences demonstrated on two databases.

Many studies that rely on manual ECG interpretation as a reference use multiple ECG expert interpreters and a method to resolve differences between interpreters, reflecting the fact that experts sometimes use different criteria. The aim of this study was to show the effect of manual ECG interpretation style on training automated ECG interpretation. METHODS: The effect of ECG interpretation style or differing ECG criteria on algorithm training was shown in this study by careful analysis of the changes in algorithm performance when the algorithm was trained on one database and tested on a different database. Morphology related ECG interpretation was summarized in eleven abnormalities such as left bundle branch block (LBBB) and old anterior myocardial infarction (MI). Each of the two databases used in the study had a reference interpretation mapped to those eleven abnormalities. F1 algorithm performance scores across abnormalities were compared for four cases. First, the algorithm was trained and tested on randomly split database A and then trained on the training set of database A and tested on randomly chosen test set of database B. The previous two test cases were repeated for opposite databases, train and test on database B and then train on database B and test on the test set of database A. RESULTS: F1 scores across abnormalities were generally higher when training and testing on the same database. F1 scores were high for bundle branch blocks (BBB) no matter the training and testing database combination. Old anterior MI F1 score dropped for one cross-database comparison and not the other suggesting a difference in manual interpretation. CONCLUSION: For some abnormalities, human experts appear to have used different criteria for ECG interpretation, as evident by the difference between cross-database and within-database performance. Bundle branch blocks appear to be interpreted in a consistent manner.

Diagnostic performance of a new ECG algorithm for reducing false positive cases in patients suspected acute coronary syndrome.

BACKGROUND: Early and correct diagnosis of ST-segment elevation myocardial infarction (STEMI) is crucial for providing timely reperfusion therapy. Patients with ischemic symptoms presenting with ST-segment elevation on the electrocardiogram (ECG) are preferably transported directly to a catheterization laboratory (Cath-lab) for primary percutaneous coronary intervention (PPCI). However, the ECG often contains confounding factors making the STEMI diagnosis challenging leading to false positive Cath-lab activation. The objective of this study was to test the performance of a standard automated algorithm against an additional high specificity setting developed for reducing the false positive STEMI calls. METHODS: We included consecutive patients with an available digital prehospital ECG triaged directly to Cath-lab for acute coronary angiography between 2009 and 2012. An adjudicated discharge diagnosis of STEMI or no myocardial infarction (no-MI) was assigned for each patient. The new automatic algorithm contains a feature to reduce false positive STEMI interpretation. The STEMI performance with the standard setting (STD) and the high specificity setting (HiSpec) was tested against the adjudicated discharge diagnosis in a retrospective manner. RESULTS: In total, 2256 patients with an available digital prehospital ECG (mean age 63 ± 13 years, male gender 71%) were included in the analysis. The discharge diagnosis of STEMI was assigned in 1885 (84%) patients. The STD identified 165 true negative and 1457 true positive (206 false positive and 428 false negative) cases (77.3%, 44.5%, 87.6% and 17.3% for sensitivity, specificity, PPV and NPV, respectively). The HiSpec identified 191 true negative and 1316 true positive (180 false positive and 569 false negative) cases (69.8%, 51.5%, 88.0% and 25.1% for sensitivity, specificity, PPV and NPV, respectively). From STD to HiSpec, false positive cases were reduced by 26 (12,6%), but false negative results were increased by 33%. CONCLUSIONS: Implementing an automated ECG algorithm with a high specificity setting was able to reduce the number of false positive STEMI cases. However, the predictive values for both positive and negative STEMI identification were moderate in this highly selected STEMI population. Finally, due the reduced sensitivity/increased false negatives, a negative AMI statement should not be solely based on the automated ECG statement.

Performance and limitations of automated ECG interpretation statements in patients with suspected acute coronary syndrome.

BACKGROUND: The 12­lead ECG plays an important role in triaging patients with symptomatic coronary artery disease, making automated ECG interpretation statements of "Acute MI" or "Acute Ischemia" crucial, especially during prehospital transport when access to physician interpretation of the ECG is limited. However, it remains unknown how automated interpretation statements correspond to adjudicated clinical outcomes during hospitalization. We sought to evaluate the diagnostic performance of prehospital automated interpretation statements to four well-defined clinical outcomes of interest: confirmed ST- segment elevation myocardial infarction (STEMI); presence of actionable coronary culprit lesions, myocardial necrosis, or any acute coronary syndrome (ACS). METHODS: An observational cohort study that enrolled consecutive patients with non-traumatic chest pain transported via ambulance. Prehospital ECGs were obtained with the Philips MRX monitor from the medical command center and re-processed using manufacturer-specific diagnostic algorithms to denote the likelihood of >>>Acute MI<<< or >>>Acute Ischemia<<<. Two independent reviewers retrospectively adjudicated the study outcomes and disagreements were resolved by a third reviewer. RESULTS: Our study included 2400 patients (age 59 ± 16, 47% females, 41% Black), with 190 (8%) patients with documented automated diagnostic statements of acute MI or acute ischemia. The sensitivity/specificity of the automated algorithm for detecting confirmed STEMI (n = 143, 6%); presence of actionable coronary culprit lesions (n = 258, 11%), myocardial necrosis (n = 291, 12%), or any ACS (n = 378, 16%) were 62.9%/95.6%; 37.2%/95.6%; 38.5%/96.4%; and 30.7%/96.3%, respectively. CONCLUSION: Although being very specific, automated interpretation statements of acute MI/acute ischemia on prehospital ECGs are not satisfactorily sensitive to exclude symptomatic coronary disease. Patients without these automated interpretation statements should be considered further for significant underlying coronary disease based on the clinical context. TRIAL REGISTRATION: ClinicalTrials.gov # NCT04237688.

Novel ECG features and machine learning to optimize culprit lesion detection in patients with suspected acute coronary syndrome.

BACKGROUND: Novel temporal-spatial features of the 12­lead ECG can conceptually optimize culprit lesions' detection beyond that of classical ST amplitude measurements. We sought to develop a data-driven approach for ECG feature selection to build a clinically relevant algorithm for real-time detection of culprit lesion. METHODS: This was a prospective observational cohort study of chest pain patients transported by emergency medical services to three tertiary care hospitals in the US. We obtained raw 10-s, 12­lead ECGs (500 s/s, HeartStart MRx, Philips Healthcare) during prehospital transport and followed patients 30 days after the encounter to adjudicate clinical outcomes. A total of 557 global and lead-specific features of P-QRS-T waveform were harvested from the representative average beats. We used Recursive Feature Elimination and LASSO to identify 35/557, 29/557, and 51/557 most recurrent and important features for LAD, LCX, and RCA culprits, respectively. Using the union of these features, we built a random forest classifier with 10-fold cross-validation to predict the presence or absence of culprit lesions. We compared this model to the performance of a rule-based commercial proprietary software (Philips DXL ECG Algorithm). RESULTS: Our sample included 2400 patients (age 59 ± 16, 47% female, 41% Black, 10.7% culprit lesions). The area under the ROC curves of our random forest classifier was 0.85 ± 0.03 with sensitivity, specificity, and negative predictive value of 71.1%, 84.7%, and 96.1%. This outperformed the accuracy of the automated interpretation software of 37.2%, 95.6%, and 92.7%, respectively, and corresponded to a net reclassification improvement index of 23.6%. Metrics of ST80; Tpeak-Tend; spatial angle between QRS and T vectors; PCA ratio of STT waveform; T axis; and QRS waveform characteristics played a significant role in this incremental gain in performance. CONCLUSIONS: Novel computational features of the 12­lead ECG can be used to build clinically relevant machine learning-based classifiers to detect culprit lesions, which has important clinical implications.

Identifying the most important ECG predictors of reduced ejection fraction in patients with suspected acute coronary syndrome.

BACKGROUND: Non-invasive screening tools of cardiac function can play a significant role in the initial triage of patients with suspected acute coronary syndrome. Numerous ECG features have been previously linked with cardiac contractility in the general population. We sought to identify ECG features that are most predictive for real-time screening of reduced left ventricular ejection fraction (LVEF) in the acute care setting. METHODS: We performed a secondary analysis of a prospective, observational cohort study of patients evaluated for suspected acute coronary syndrome. We included consecutive patients in whom an echocardiogram was performed during indexed encounter. We evaluated 554 automated 12-lead ECG features in multivariate linear regression for predicting LVEF. We then used regression trees to identify the most important predictive ECG features. RESULTS: Our final sample included 297 patients (aged 63 ± 15, 45% females). The mean LVEF was 57% ± 13 (IQR 50%-65%). In multivariate analysis, depolarization dispersion in the horizontal plane; global repolarization dispersion; and abnormal temporal indices in inferolateral leads were all independent predictors of LVEF (R2 = 0.452, F = 6.679, p < 0.001). Horizontal QRS axis deviation and prolonged ventricular activation time in left ventricular apex were the most important determinants of reduced LVEF, while global QRS duration was of less importance. CONCLUSIONS: Poor R wave progression in precordial leads with dominant QS pattern in V3 is the most predictive feature of reduced LVEF in suspected ACS. This feature constitutes a simple visual marker to aid clinicians in identifying those with impaired cardiac function.

Discovery of Potent Inhibitors of 11ß-Hydroxysteroid Dehydrogenase Type 1 Using a Novel Growth-Based Protocol of in Silico Screening and Optimization in CONTOUR.

With the continuous progress in ultralarge virtual libraries which are readily accessible, it is of great interest to explore this large chemical space for hit identification and lead optimization using reliable structure-based approaches. In this work, a novel growth-based screening protocol has been designed and implemented in the structure-based design platform CONTOUR. The protocol was used to screen the ZINC database in silico and optimize hits to discover 11ß-HSD1 inhibitors. In contrast to molecular docking, the virtual screening process makes significant improvements in computational efficiency without losing chemical equities through partitioning 1.8 million ZINC compounds into fragments, docking fragments to form key hydrogen bonds with anchor residues, reorganizing molecules into molecular fragment trees using matched fragments and common substructures, and then regrowing molecules with the help of developed intelligent growth features inside the protein binding site to find hits. The growth-base screening approach is validated by the high hit rate. A total of 50 compounds have been selected for testing; of these, 15 hits having diverse scaffolds are found to inhibit 11ß-HSD1 with IC50 values of less than 1 µM in a biochemical enzyme assay. The best hit which exhibits an enzyme IC50 of 33 nM is further developed to a novel series of bicyclic 11ß-HSD1 inhibitors with the best inhibition of enzyme IC50 of 3.1 nM. The final lead candidate exhibits IC50 values of 7.2 and 21 nM in enzyme and adipocyte assays, respectively, displayed greater than 1000-fold of selectivity over 11ß-HSD2 and two other related hydroxysteroid dehydrogenases, and can serve as good starting points for further optimization to develop clinical candidates.