Enhancing the diagnosis of functionally relevant coronary artery disease with machine learning.

Functionally relevant coronary artery disease (fCAD) can result in premature death or nonfatal acute myocardial infarction. Its early detection is a fundamentally important task in medicine. Classical detection approaches suffer from limited diagnostic accuracy or expose patients to possibly harmful radiation. Here we show how machine learning (ML) can outperform cardiologists in predicting the presence of stress-induced fCAD in terms of area under the receiver operating characteristic (AUROC: 0.71 vs. 0.64, p = 4.0E-13). We present two ML approaches, the first using eight static clinical variables, whereas the second leverages electrocardiogram signals from exercise stress testing. At a target post-test probability for fCAD of <15%, ML facilitates a potential reduction of imaging procedures by 15-17% compared to the cardiologist's judgement. Predictive performance is validated on an internal temporal data split as well as externally. We also show that combining clinical judgement with conventional ML and deep learning using logistic regression results in a mean AUROC of 0.74.

Comparing the utility of clinical risk scores and integrated clinical judgement in patients with suspected acute coronary syndrome.

AIMS: The utility of clinical risk scores regarding the prediction of major adverse cardiac events (MACE) is uncertain. We aimed to directly compare the prognostic performance of five established clinical risk scores as well as an unstructured integrated clinical judgement (ICJ) of the treating emergency department (ED) physician. METHODS AND RESULTS: Thirty-day MACE including all-cause death, life-threatening arrhythmia, cardiogenic shock, acute myocardial infarction (including the index event), and unstable angina requiring urgent coronary revascularization were centrally adjudicated by two independent cardiologists in patients presenting to the ED with acute chest discomfort in an international multicentre study. We compared the prognostic performance of the HEART score, GRACE score, T-MACS, TIMI score, and EDACS, as well as the unstructured ICJ of the treating ED physician (visual analogue scale to estimate the probability of acute coronary syndrome, ranging from 0 to 100). Among 4551 eligible patients, 1110/4551 patients (24.4%) had at least one MACE within 30 days. Prognostic accuracy was high and comparable for the HEART score, GRACE score, T-MACS, and ICJ [area under the receiver operating characteristic curve (AUC) 0.85-0.87] but significantly lower and only moderate for the TIMI score (AUC 0.79, P < 0.001) and EDACS (AUC 0.74, P < 0.001), resulting in sensitivities for the rule-out of 30-day MACE of 93-96, 87 (P < 0.001), and 72% (P < 0.001), respectively. CONCLUSION: The HEART score, GRACE score, T-MACS, and unstructured ICJ of the treating physician, not the TIMI score or EDACS, performed well for the prediction of 30-day MACE and may be considered for routine clinical use. TRIAL REGISTRATION: ClinicalTrials.gov number NCT00470587.

Automated F18-FDG PET/CT image quality assessment using deep neural networks on a latest 6-ring digital detector system.

To evaluate whether a machine learning classifier can evaluate image quality of maximum intensity projection (MIP) images from F18-FDG-PET scans. A total of 400 MIP images from F18-FDG-PET with simulated decreasing acquisition time (120 s, 90 s, 60 s, 30 s and 15 s per bed-position) using block sequential regularized expectation maximization (BSREM) with a beta-value of 450 and 600 were created. A machine learning classifier was fed with 283 images rated "sufficient image quality" and 117 images rated "insufficient image quality". The classification performance of the machine learning classifier was assessed by calculating sensitivity, specificity, and area under the receiver operating characteristics curve (AUC) using reader-based classification as the target. Classification performance of the machine learning classifier was AUC 0.978 for BSREM beta 450 and 0.967 for BSREM beta 600. The algorithm showed a sensitivity of 89% and 94% and a specificity of 94% and 94% for the reconstruction BSREM 450 and 600, respectively. Automated assessment of image quality from F18-FDG-PET images using a machine learning classifier provides equivalent performance to manual assessment by experienced radiologists.

Soluble urokinase plasminogen activator receptor and functionally relevant coronary artery disease: a prospective cohort study.

BACKGROUND: Soluble urokinase plasminogen activator receptor (suPAR) is an emerging biomarker associated with anatomical CAD burden and cardiovascular outcomes including myocardial infarction (MI) and death. We aimed to validate previous findings of the prognostic value of suPAR and to evaluate its diagnostic potential for functional relevant CAD (fCAD). METHODS: Consecutive patients with suspected fCAD were enrolled. Adjudication of fCAD was performed blinded to suPAR concentrations by myocardial perfusion single-photon emission tomography (MPI-SPECT) and coronary angiography. Prognostic outcome measures included all-cause death, cardiovascular death, and incident MI during 2-year follow-up. RESULTS: Among consecutive 968 patients, suPAR concentrations were higher in patients with fCAD compared to those without (3.45 vs. 3.20 ng/mL, p = 0.007), but did not provide acceptable diagnostic accuracy (area under the curve [AUC]: 0.56, 95%CI 0.52-0.60). SuPAR correlated with high-sensitivity cardiac-troponin T (Spearman's rho (ρ) 0.393, p < 0.001), NT-proBNP (ρ = 0.327, p < 0.001), age (ρ = 0.364, p < 0.001) and very weakly with coronary atherosclerosis (ρ = 0.123, p < 0.001). Prognostic discrimination of suPAR was moderate for cardiovascular death (AUC = 0.72, 95%CI 0.62-0.81) and all-cause death (AUC = 0.72, 95%CI 0.65-0.79) at 2-years. SuPAR remained a significant predictor for all-cause death in multivariable Cox regression (HR = 1.96, p = 0.001). CONCLUSIONS: SuPAR was an independent predictor of all-cause death, without diagnostic utility for fCAD. CLINICAL TRIAL REGISTRATION: NCT01838148.

Lower diagnostic accuracy of hs-cTnI in patients with prior coronary artery bypass grafting.

BACKGROUND: High-sensitivity cardiac troponin T (hs-cTnT) and the ESC 0/1h-hs-cTnT-algorithm have worse performance in the early diagnosis of myocardial infarction (MI) in patients with prior coronary artery bypass grafting (CABG). It is unknown, whether this concern applies also to hs-cTnI, the most widely used analyte worldwide. METHODS: In an international multicenter diagnostic study, two cardiologists centrally adjudicated the final diagnosis in patients presenting to the emergency department with symptoms suggestive of MI according to the Third Universal Definition of MI. The objective was to compare the diagnostic accuracy of hs-cTnI assays and their performance within the ESC hs-cTnI 0/1h-algorithms in patients with versus without prior CABG. Findings were externally validated in an U.S. multicenter diagnostic study. RESULTS: A total of 392/5'200 patients (8%) had prior coronary artery bypass grafting (CABG). Diagnostic accuracy of hs-cTnI as quantified by the area under the receiver-operating characteristics-curve (AUC) in these patients was high, but lower versus patients without prior CABG (e.g. hs-cTnI-Architect 0.91 versus 0.95; p = 0.016). Sensitivity/specificity of rule-out/in by the European Society of Cardiology (ESC) 0/1h-hs-cTnI-algorithms remained very high [e.g. hs-cTnI-Architect 100% and 93.5%], but efficacy was lower (52% versus 74%, p < 0.01). External validation (n = 2113) confirmed these findings in 192 patients with prior CABG using hs-cTnI-Atellica, with 52% versus 36% (p < 0.001) remaining in the observe zone. CONCLUSIONS: Diagnostic accuracy of hs-cTnI and efficacy of the ESC 0/1h-hs-cTnI-algorithms are lower in patients with prior CABG, but sensitivity/specificity remain very high. CLINICAL TRIAL REGISTRATION: https://clinicaltrials.gov/ct2/show/NCT00470587, number NCT00470587.

Influence of renin-angiotensin-aldosterone system inhibitors on plasma levels of angiotensin-converting enzyme 2.

AIMS: Concern has been raised that treatment with angiotensin-converting enzyme inhibitors and angiotensin receptor blockers may increase the expression of angiotensin-converting enzyme 2 (ACE2), which acts as the entry receptor for SARS-CoV-2, and lead to an increased risk of death from SARS-CoV-2. We aimed to address this concern by evaluating the in vivo relationship of treatment with ACE inhibitors and angiotensin receptor blockers (ARB) with circulating plasma concentrations of ACE2 in a large cohort of patients with established cardiovascular disease (n = 1864) or cardiovascular risk factors (n = 2144) but without a history of heart failure. METHODS AND RESULTS: Angiotensin-converting enzyme 2 was measured in 4008 patients (median age 68, 33% women, 31% on ACE-inhibitors, 31% on ARB) using the SOMAscan proteomic platform (SomaLogic Inc, Colorado, USA). Plasma concentration of ACE2 was comparable in 1250 patients on ACE inhibitors (mean 5.99) versus patients without ACE inhibitors (mean 5.98, P = 0.54). Similarly, plasma concentration of ACE2 was comparable in 1260 patients on ARB (mean 5.99) versus patients without ARB (mean 5.98, P = 0.50). Plasma concentration of ACE2 was comparable in 2474 patients on either ACE inhibitors or ARB (mean 5.99) versus patients without ACE inhibitors or ARB (mean 5.98, P = 0.31). Multivariable quantile regression model analysis confirmed the lack of association between treatment with ACE inhibitors or ARB and ACE2 concentrations. Body mass index showed the only positive association with ACE2 plasma concentration (effect 0.015, 95% confidence interval 0.002 to 0.028, P = 0.024). CONCLUSIONS: In a large cohort of patients with established cardiovascular disease or cardiovascular risk factors but without heart failure, ACE inhibitors and ARB were not associated with higher plasma concentrations of ACE2.

Cardiovascular Biomarkers in the Early Discrimination of Type 2 Myocardial Infarction.

Importance: Rapid and accurate noninvasive discrimination of type 2 myocardial infarction (T2MI), which is because of a supply-demand mismatch, from type 1 myocardial infarction (T1MI), which arises via plaque rupture, is essential, because treatment differs substantially. Unfortunately, this is a major unmet clinical need, because even high-sensitivity cardiac troponin (hs-cTn) measurement provides only modest accuracy. Objective: To test the hypothesis that novel cardiovascular biomarkers quantifying different pathophysiological pathways involved in T2MI and/or T1MI may aid physicians in the rapid discrimination of T2MI vs T1MI. Design, Setting, and Participants: This international, multicenter prospective diagnostic study was conducted in 12 emergency departments in 5 countries (Switzerland, Spain, Italy, Poland, and the Czech Republic) with patients presenting with acute chest discomfort to the emergency departments. The study quantified the discrimination of hs-cTn T, hs-cTn I, and 17 novel cardiovascular biomarkers measured in subsets of consecutively enrolled patients against a reference standard (final diagnosis), centrally adjudicated by 2 independent cardiologists according to the fourth universal definition of MI, using all information, including cardiac imaging and serial measurements of hs-cTnT or hs-cTnI. Results: Among 5887 patients, 1106 (18.8%) had an adjudicated final diagnosis of MI; of these, 860 patients (77.8%) had T1MI, and 246 patients (22.2%) had T2MI. Patients with T2MI vs those with T1MI had lower concentrations of biomarkers quantifying cardiomyocyte injury hs-cTnT (median [interquartile range (IQR)], 30 (17-55) ng/L vs 58 (28-150) ng/L), hs-cTnI (median [IQR], 23 [10-83] ng/L vs 115 [28-576] ng/L; P < .001), and cardiac myosin-binding protein C (at presentation: median [IQR], 76 [38-189] ng/L vs 257 [75-876] ng/L; P < .001) but higher concentrations of biomarkers quantifying endothelial dysfunction, microvascular dysfunction, and/or hemodynamic stress (median [IQR] values: C-terminal proendothelin 1, 97 [75-134] pmol/L vs 68 [55-91] pmol/L; midregional proadrenomedullin, 0.97 [0.67-1.51] pmol/L vs 0.72 [0.53-0.99] pmol/L; midregional pro-A-type natriuretic peptide, 378 [207-491] pmol/L vs 152 [90-247] pmol/L; and growth differentiation factor 15, 2.26 [1.44-4.35] vs 1.56 [1.02-2.19] ng/L; all P < .001). Discrimination for these biomarkers, as quantified by the area under the receiver operating characteristics curve, was modest (hs-cTnT, 0.67 [95% CI, 0.64-0.71]; hs-cTn I, 0.71 [95% CI, 0.67-0.74]; cardiac myosin-binding protein C, 0.67 [95% CI, 0.61-0.73]; C-terminal proendothelin 1, 0.73 [95% CI, 0.63-0.83]; midregional proadrenomedullin, 0.66 [95% CI, 0.60-0.73]; midregional pro-A-type natriuretic peptide, 0.77 [95% CI, 0.68-0.87]; and growth differentiation factor 15, 0.68 [95% CI, 0.58-0.79]). Conclusions and Relevance: In this study, biomarkers quantifying myocardial injury, endothelial dysfunction, microvascular dysfunction, and/or hemodynamic stress provided modest discrimination in early, noninvasive diagnosis of T2MI.

Development of an electrocardiogram-based risk calculator for a cardiac cause of syncope.

OBJECTIVE: To develop an ECG-based tool for rapid risk assessment of a cardiac cause of syncope in patients ≥40 years. METHODS: In a prospective international multicentre study, 2007 patients ≥40 years presenting with syncope were recruited in the emergency department (ED) of participating centres ranging from large university hospitals to smaller rural hospitals in eight countries from May 2010 to July 2017. 12-Lead ECG recordings were obtained at ED presentation following the syncopal event. The primary diagnostic outcome, a cardiac cause of syncope, was centrally adjudicated by two independent cardiologists using all available clinical information including 12-month follow-up. ECG predictors for a cardiac cause of syncope were identified using penalised backward selection and a continuous-scale likelihood was calculated based on regression analysis coefficients. Findings were validated in an independent US multicentre cohort including 2269 patients. RESULTS: In the derivation cohort, a cardiac cause of syncope was adjudicated in 267 patients (16%). Seven ECG criteria were identified as predictors for this outcome: heart rate and QTc-interval (continuous predictors), rhythm, atrioventricular block, ST-segment depression, bundle branch block and ventricular extrasystole/non-sustained ventricular tachycardia (categorical predictors). Diagnostic accuracy of these combined predictors for a cardiac cause of syncope was high (area under the curve 0.80, 95% CI 0.77 to 0.83). Overall, 138 patients (8%) were rapidly triaged towards rule-out and 181 patients (11%) towards rule-in of a cardiac cause of syncope. External validation showed similar performance. CONCLUSION: In patients ≥40 years with a syncopal event, a combination of seven ECG criteria enabled rapid assessment of the likelihood that syncope was due to a cardiac cause. TRIAL REGISTRATION NUMBER: NCT01548352 (BASEL IX), NCT01802398 (SRS study).

Patterns of organizing pneumonia and microinfarcts as surrogate for endothelial disruption and microangiopathic thromboembolic events in patients with coronavirus disease 2019.

BACKGROUND: To evaluate chest-computed-tomography (CT) scans in coronavirus-disease-2019 (COVID-19) patients for signs of organizing pneumonia (OP) and microinfarction as surrogate for microscopic thromboembolic events. METHODS: Real-time polymerase-chain-reaction (RT-PCR)-confirmed COVID-19 patients undergoing chest-CT (non-enhanced, enhanced, pulmonary-angiography [CT-PA]) from March-April 2020 were retrospectively included (COVID-19-cohort). As control-groups served 175 patients from 2020 (cohort-2020) and 157 patients from 2019 (cohort-2019) undergoing CT-PA for pulmonary embolism (PE) during the respective time frame at our institution. Two independent readers assessed for presence and location of PE in all three cohorts. In COVID-19 patients additionally parenchymal changes typical of COVID-19 pneumonia, infarct pneumonia and OP were assessed. Inter-reader agreement and prevalence of PE in different cohorts were calculated. RESULTS: From 68 COVID-19 patients (42 female [61.8%], median age 59 years [range 32-89]) undergoing chest-CT 38 obtained CT-PA. Inter-reader-agreement was good (k = 0.781). On CT-PA, 13.2% of COVID-19 patients presented with PE whereas in the control-groups prevalence of PE was 9.1% and 8.9%, respectively (p = 0.452). Up to 50% of COVID-19 patients showed changes typical for OP. 21.1% of COVID-19 patients suspected with PE showed subpleural wedge-shaped consolidation resembling infarct pneumonia, while only 13.2% showed visible filling defects of the pulmonary artery branches on CT-PA. CONCLUSION: Despite the reported hypercoagulability in critically ill patients with COVID-19, we did not encounter higher prevalence of PE in our patient cohort compared to the control cohorts. However, patients with suspected PE showed a higher prevalence of lung changes, resembling patterns of infarct pneumonia or OP and CT-signs of pulmonary-artery hypertension.

Early Diagnosis of Myocardial Infarction With Point-of-Care High-Sensitivity Cardiac Troponin I.

BACKGROUND: Until now, high-sensitivity cardiac troponin (hs-cTn) assays were mainly developed for large central laboratory platforms. OBJECTIVES: This study aimed to assess the clinical performance of a point-of-care (POC)-hs-cTnI assay in patients with suspected myocardial infarction (MI). METHODS: This study enrolled patients presenting to the emergency department with symptoms suggestive of MI. Two cardiologists centrally adjudicated the final diagnosis using all clinical data including cardiac imaging. The primary objective was to directly compare diagnostic accuracy of POC-hs-cTnI-TriageTrue versus best-validated central laboratory assays. Secondary objectives included the derivation and validation of a POC-hs-cTnI-TriageTrue-specific 0/1-h algorithm. RESULTS: MI was the adjudicated final diagnosis in 178 of 1,261 patients (14%). The area under the curve (AUC) for POC-hs-cTnI-TriageTrue at presentation was 0.95 (95% confidence interval [CI]: 0.93 to 0.96) and was at least comparable to hs-cTnT-Elecsys (AUC: 0.94; 95% CI: 0.93 to 0.96; p = 0.213) and hs-cTnI-Architect (AUC: 0.92; 95% CI: 0.90 to 0.93; p < 0.001). A single cutoff concentration <3 ng/l at presentation identified 45% of patients at low risk with a negative predictive value (NPV) of 100% (95% CI: 99.4% to 100%). A single cutoff concentration >60 ng/l identified patients at high risk with a positive predictive value (PPV) of 76.8% (95% CI: 68.9% to 83.6%). The 0/1-h algorithm ruled out 55% of patients (NPV: 100%; 95% CI: 98.8% to 100%), and ruled in 18% of patients (PPV: 76.8%; 95% CI: 67.2% to 84.7%). Ruled-out patients had cumulative event rates of 0% at 30 days and 1.6% at 2 years. This study confirmed these findings in a secondary analysis including hs-cTnI-Architect for central adjudication. CONCLUSIONS: The POC-hs-cTnI-TriageTrue assay provides high diagnostic accuracy in patients with suspected MI with a clinical performance that is at least comparable to that of best-validated central laboratory assays. (Advantageous Predictors of Acute Coronary Syndromes Evaluation Study [APACE]; NCT00470587).

Daytime variation of perioperative myocardial injury in non-cardiac surgery and effect on outcome.

OBJECTIVE: Recently, daytime variation in perioperative myocardial injury (PMI) has been observed in patients undergoing cardiac surgery. We aim at investigating whether daytime variation also occurs in patients undergoing non-cardiac surgery. METHODS: In a prospective diagnostic study, we evaluated the presence of daytime variation in PMI in patients at increased cardiovascular risk undergoing non-cardiac surgery, as well as its possible impact on the incidence of acute myocardial infarction (AMI), and death during 1-year follow-up in a propensity score-matched cohort. PMI was defined as an absolute increase in high-sensitivity cardiac troponin T (hs-cTnT) concentration of ≥14 ng/L from preoperative to postoperative measurements. RESULTS: Of 1641 patients, propensity score matching defined 630 with similar baseline characteristics, half undergoing non-cardiac surgery in the morning (starting from 8:00 to 11:00) and half in the afternoon (starting from 14:00 to 17:00). There was no difference in PMI incidence between both groups (morning: 50, 15.8% (95% CI 12.3 to 20.3); afternoon: 52, 16.4% (95% CI 12.7 to 20.9), p=0.94), nor if analysing hs-cTnT release as a quantitative variable (median morning group: 3 ng/L (95% CI 1 to 7 ng/L); median afternoon group: 2 ng/L (95% CI 0 to 7 ng/L; p=0.16). During 1-year follow-up, the incidence of AMI was 1.2% (95% CI 0.4% to 3.2%) among morning surgeries versus 4.1% (95% CI 2.3% to 6.9%) among the afternoon surgeries (corrected HR for afternoon surgery 3.44, bootstrapped 95% CI 1.33 to 10.49, p log-rank=0.03), whereas no difference in mortality emerged (p=0.70). CONCLUSIONS: Although there is no daytime variation in PMI in patients undergoing non-cardiac surgery, the incidence of AMI during follow-up is increased in afternoon surgeries and requires further study. CLINICAL TRIAL REGISTRATION: NCT02573532;Results.