Shelly-Cohen, Michal

Abstract Aims The current gold standard comprehensive assessment of coronary microvascular dysfunction (CMD) is through a limited-access invasive catheterization lab procedure. We aimed to develop a point-of-care tool to assist clinical guidance in patients presenting with chest pain and/or an abnormal cardiac functional stress test and with non-obstructive coronary artery disease (NOCAD). Methods and results This study included 1893 NOCAD patients (<50% angiographic stenosis) who underwent CMD evaluation as well as an electrocardiogram (ECG) up to 1-year prior. Endothelial-independent CMD was defined by coronary flow reserve (CFR) ≤2.5 in response to intracoronary adenosine. Endothelial-dependent CMD was defined by a maximal percent increase in coronary blood flow (%ΔCBF) ≤50% in response to intracoronary acetylcholine infusion. We trained algorithms to distinguish between the following outcomes: CFR ≤2.5, %ΔCBF ≤50, and the combination of both. Two classes of algorithms were trained, one depending on ECG waveforms as input, and another using tabular clinical data. Mean age was 51 ± 12 years and 66% were females (n = 1257). Area under the curve values ranged from 0.49 to 0.67 for all the outcomes. The best performance in our analysis was for the outcome CFR ≤2.5 with clinical variables. Area under the curve and accuracy were 0.67% and 60%. When decreasing the threshold of positivity, sensitivity and negative predictive value increased to 92% and 90%, respectively, while specificity and positive predictive value decreased to 25% and 29%, respectively. Conclusion An artificial intelligence-enabled algorithm may be able to assist clinical guidance by ruling out CMD in patients presenting with chest pain and/or an abnormal functional stress test. This algorithm needs to be prospectively validated in different cohorts.

Abstract Aims Spectrum bias can arise when a diagnostic test is derived from study populations with different disease spectra than the target population, resulting in poor generalizability. We used a real-world artificial intelligence (AI)-derived algorithm to detect severe aortic stenosis (AS) to experimentally assess the effect of spectrum bias on test performance. Methods and results All adult patients at the Mayo Clinic between 1 January 1989 and 30 September 2019 with transthoracic echocardiograms within 180 days after electrocardiogram (ECG) were identified. Two models were developed from two distinct patient cohorts: a whole-spectrum cohort comparing severe AS to any non-severe AS and an extreme-spectrum cohort comparing severe AS to no AS at all. Model performance was assessed. Overall, 258 607 patients had valid ECG and echocardiograms pairs. The area under the receiver operator curve was 0.87 and 0.91 for the whole-spectrum and extreme-spectrum models, respectively. Sensitivity and specificity for the whole-spectrum model was 80% and 81%, respectively, while for the extreme-spectrum model it was 84% and 84%, respectively. When applying the AI-ECG derived from the extreme-spectrum cohort to patients in the whole-spectrum cohort, the sensitivity, specificity, and area under the curve dropped to 83%, 73%, and 0.86, respectively. Conclusion While the algorithm performed robustly in identifying severe AS, this study shows that limiting datasets to clearly positive or negative labels leads to overestimation of test performance when testing an AI algorithm in the setting of classifying severe AS using ECG data. While the effect of the bias may be modest in this example, clinicians should be aware of the existence of such a bias in AI-derived algorithms.

Abstract Background We have demonstrated that artificial intelligence interpretation of ECGs (AI-ECG) can estimate an individual's physiologic age and that the gap between AI-ECG and chronologic age (Age-Gap) is associated with increased mortality. We hypothesized that Age-Gap would predict long-term atherosclerotic cardiovascular disease (ASCVD) and that Age-Gap would refine the ACC/AHA Pooled Cohort Equations' (PCE) predictive abilities. Methods Using the Rochester Epidemiology Project (REP) we evaluated a community-based cohort of consecutive patients seeking primary care between 1998–2000 and followed through March 2016. Inclusion criteria were age 40–79 and complete data to calculate PCE. We excluded those with known ASCVD, AF, HF or an event within 30 days of baseline.A neural network, trained, validated, and tested in an independent cohort of ∼ 500,000 independent patients, using 10-second digital samples of raw, 12 lead ECGs. PCE was categorized as low<5%, intermediate 5–9.9%, high 10–19.9%, and very high≥20%. The primary endpoint was ASCVD and included fatal and non-fatal myocardial infarction and ischemic stroke; the secondary endpoint also included coronary revascularization [Percutaneous Coronary Intervention (PCI) or Coronary Artery Bypass Graft (CABG)], TIA and Cardiovascular mortality. Events were validated in duplicate. Follow-up was truncated at 10 years for PCE analysis. The association between Age-Gap with ASCVD and expanded ASCVD was assessed with cox proportional hazard models that adjusted for chronological age, sex and risk factors. Models were stratified by PCE risk categories to evaluate the effect of PCE predicted risk. Results We included 24,793 patients (54% women, 95% Caucasian) with mean follow up of 12.6±5.1 years. 2,366 (9.5%) developed ASCVD events and 3,401 (13.7%) the expanded ASCVD. Mean chronologic age was 53.6±11.6 years and the AI-ECG age was 54.5±10.9 years, R2=0.7865, p<0.0001. The mean Age-Gap was 0.87±7.38 years. After adjusting for age and sex, those considered older by ECG, compared to their chronologic age had a higher risk for ASCVD when compared to those with <−2 SD age gap (considered younger by ECG). (Figure 1A), with similar results when using the expanded definition of ASCVD (data not shown). Furthermore, Age-Gap enhanced predicted capabilities of the PCE among those with low 10-year predicted risk (<5%): Age and sex adjusted HR 4.73, 95% CI 1.42–15.74, p-value=0.01 and among those with high predicted risk (>20%) age and sex adjusted HR 6.90, 95% CI 1.98–24.08, p-value=0.0006, when comparing those older to younger by ECG respectively (Figure 1B). Conclusion The difference between physiologic AI-ECG age and chronologic age is associated with long-term ASCVD, and enhances current risk calculators (PCE) ability to identify high and low risk individuals. This may help identify individuals who should or should not be treated with newer, expensive risk-reducing therapies. Funding Acknowledgement Type of funding source: Foundation. Main funding source(s): Mayo Clinic