Cardiologists manage a large volume of referrals for complex patients with limited time. ChatGPT (cGPT), an artificial intelligence (AI) language model, has garnered significant interest as an off-label medical search engine. This study compared cGPT with cardiologist referral advice.

Timely assessment of cardiac patients in the ambulatory setting is crucial to implement evidence-based therapies and to prevent hospital admissions. We implemented a rapid cardiac screening (RCS) clinic to achieve this. We established a single point of contact RCS clinic using biomarkers, advanced ECG (A-ECG), scout echocardiography (sEcho) and brief consultation. This involved three Phases—Building: 1) an evidence base for artificial intelligence (AI) technology, 2) a pilot screening of patients on an Echo waiting list, 3) a clinic using improvement methodology. Time in motion, patient feedback and outcomes were compared to 2016 historic controls (n=13,107). Phases 1 to 2 involved n=559 patients. A-ECG predicted structural heart disease, (AUC=0.85; 95% CI, 0.78–0.91; p < 0.0001), Long QT Syndrome (AUC=0.87; 95% CI, 0.75 to 0.97; p<0.0001), left ventricular systolic dysfunction, (AUC: 0.92; 95% CI, 0.86–0.96; p<0.0001). sEcho-AI correlated with manual measures of left ventricular ejection fraction (LVEF) (r=0.71; p < 0.005). We screened 45 patients in 5 serial clinics, mean 74 (±15) mins, 30 (62%) were discharged vs 31% in controls (95% CI, 15% to 46%; p<0.0001), 6 (13.3%) proceeded to full transthoracic echocardiography (TTE) vs 68% in controls, (95% CI, 42% to 63%; p<0.0001). Patient responses were favourable. Rapid screening for cardiac pathology during a single point of contact is feasible but needs upscaling to demonstrate its utility.

Identifying familial hypercholesterolaemia (FH) using data mining, genetic testing and stratifying risk has value for targeting novel lipid lowering agents. We undertook a multistep approach to do this using electronic health records. An algorithm to identify FH was applied to a database of 57,643 cardiac patients, comparing statin use and mean age of death in high and low probability groups. A further 6,267 acute coronary syndrome (ACS) patients were manually audited for missed FH diagnoses. 100 clinic patients with suspected FH were sequenced. 27 and 50 single nucleotide polymorphism (SNP) polygenic risk scores (PRS) were validated in 78 patients, who underwent CTCA and Duke scoring and then used to risk stratify FH patients. n=5 were enrolled in siRNA ANGPTL3 clinical trials. Predicted/estimated FH was 0.9%/0.4% in the automated and 0.7%/2% in the manually adjudicated populations. In high versus low FH probability, statin use was 95% versus 68%, p<0.0001 and mean age of death was 68 versus 79 years, p<0.0001. Of 100 clinic patients 21% had a pathogenic variant: 15 LDLR, 3 ApoB and 1 PCSK9 variant of uncertain significance (VUS). Duke score predicted CAD AUROC 0.66; 95% CI, 0.5 to 0.8; p =0.04 compared to PRS AUROC 0.58; 95% CI, 0.4 to 0.7; p=0.3. FH algorithms need further validation with genotyping to match epidemiological statistics. Despite adequate use of statins, probable FH resulted in an estimated 11-year loss of life. PRS can be used for risk stratification in FH patients.

We applied machine learning (ML) to routine full blood count (FBC), urea and electrolytes (U&Es) to build predictive models for heart failure which could be used at population scale. Routine blood Results from 8,031 patients with heart failure were used in ML training and testing datasets (Split 80:20), with equal number of controls. N-terminal pro-hormone B-type natriuretic peptide (NTproBNP) was used for diagnostic comparison. Raw FBC (rawFBC) metadata was used in a dataset of 698 patients, 314 of whom had heart failure to train and test ML models (Split 70:30) from rawFBC, rawFBC + U&Es and routine bloods. The rawFBC model was used to predict heart failure in a third validation dataset of 69,492 FBCs (2.3% heart failure prevalence). Heart failure was predicted from rawFBC + U&Es (AUROC 0.91) and rawFBC (AUROC 0.88) compared to routine bloods (AUROC 0.81; 95% CI, 0.05 to 0.17, p<0.001) and NTproBNP (AUROC 0.83; 95% CI, 0.15 to 0.3, p<0.001). rawFBC was able to predict NTproBNP ≥ 34pmol/L (AUROC 0.7; 95% CI, 0.64 to 0.76, p<0.001) and heart failure with sensitivity 75%, specificity 76% (Figure 1), positive predictive value (PPV) 7%, negative predictive value (NPV) 99.2%. (AUROC 0.83; 95% CI, 0.83 to 0.84, p<0.001) in validation. Predictive features included markers of erythropoiesis (red cell distribution width, haemoglobin, haematocrit, etc). Heart failure can be predicted from routine bloods with accuracy equivalent to NTproBNP. Predictive features included markers of erythropoiesis, with therapeutic monitoring implications.

Stratifying cardiovascular risk in patients with familial hypercholesterolaemia (FH) has value for targeting novel lipid lowering agents. We describe here the development of a 27 SNP polygenic risk score (PRS), used alongside a registry for FH in New Zealand. A 27 SNP panel was created on a MassArray platform with automated interpretive software which was then validated in two cohorts of mixed ethnicity (n=796), with and without acute coronary syndrome (ACS). 68 patients with suspected FH were sequenced for LDLR and ApoB mutations using high resolution melt curve (HRM) analysis multiplex probe ligation amplification (MLPA). 27PRS was clinically applied in 37 patients with and without gene positive FH. Mean GRS27 in those without a history of ACS was significantly lower than GRS27 in an ACS cohort, mean 1.33 vs 1.42, 95% CI 0.04 to 0.14, p = 0.0002. Inter-ethnic differences in SNP frequencies were identified e.g. (Figure 1), and population distributions and quintiles of risk mapped. 13 (20%) of suspected FH patients had a pathogenic variants, 10 LDLR, 3 ApoB and 1 LDLR variant of uncertain significance (VUS). 1 patient had both a pathogenic LDLR mutation and high PRS. From the registry 5 FH patients were recalled for a Phase I siRNA trial targeting ANGPTL3. A CAD 27PRS distinguishes risk in patients with and without ACS. Interethnic variances in PRS deserve further study before widespread adoption. PRS can be delivered clinically to stratify risk in FH and registries facilitate the trial of novel therapies.

Abstract Background Elevated LDL-C and triglyceride rich lipoproteins (TRLs) are independent risk factors for cardiovascular disease (CVD). Genetic deficiency of angiopoietin-like protein 3 (ANGPTL3) is associated with reduced circulating levels of LDL-C, triglycerides (TGs), VLDL-C, HDL-C and reduced CVD risk, with no described adverse phenotype. ARO-ANG3 is a RNA interference drug designed to silence expression of ANGPTL3. Single doses of ARO-ANG3 have been shown to reduce ANGPTL3, TGs, VLDL-C and LDL-C in healthy volunteers (HVs, AHA 2019). We report the effects of multiple doses of ARO-ANG3 in HVs with a focus on the duration of action. Methods ARO-ANG3 was administered subcutaneously to HVs on days 1 and 29 at doses of 100, 200 or 300 mg (n=4 per group). Measured parameters included ANGPTL3, LDL-C, TGs, VLDL-C and HDL-C. Follow up is ongoing. Results All HVs have received both doses and follow-up is currently through week 16 (12 weeks after second dose). Mean nadir for ANGPTL3 levels occurred 2 weeks after the second dose (−83–93%) with minimal change for 200 and 300 mg but 16% recovery for 100 mg at week 16. Mean TGs and VLDL-C reached nadir earlier (3 wks, −61–65%) without apparent dose response and minimal change for any dose at wk 16. LDL-C nadir occurred 4–6 wks after the second dose (−45–54%), again with minimal evidence for dose response or change through wk 16. HDL-C was reduced 14–37% at wk 16. ARO-ANG3 was well tolerated without serious or severe adverse events or dropouts related to drug. The most common adverse events have been headache and upper respiratory infections. Conclusions Genetic deficiency of ANGPTL3 is a cause of familial combined hypolipemia and is associated with a decreased risk of CVD. Using RNAi to selectively suppress ANGPTL3 production reproduces these genetic effects with a duration of at least 12 weeks following a second dose and with good tolerability over 16 wks. ANGPTL3 inhibition results in lowering of LDL-C and TRLs which may confer protection against CVD in patients with atherogenic mixed dyslipidemia. Funding Acknowledgement Type of funding source: Private company. Main funding source(s): Arrowhead Pharmaceuticals

The early adoption of Electronic Health Records (EHRs) and ability to monitor longitudinal outcomes is a major strength of the New Zealand healthcare system. Individual-level patient databases spanning an array of hospital investigations are a resource for identifying novel disease phenotypes, cohort studies, treatment-outcome relationships and predictive analytics using machine learning. 75,617 patients undergoing echocardiography were identified (2009 and 2018) from an Excelera database and linked with ECG metadata, ICD10 coding, laboratory data, Pyxis prescribing, Paceart Optima data, and genetic data. Machine learning (Labs, advanced ECG (A-ECG), echocardiography) was performed to predict heart failure diagnosis and readmission, LV systolic dysfunction (LVSD), mortality and ICD use. Multiomics was applied to discrete cohorts identifying novel biomarkers and disease patterns. Linking multiple rich datasets was plausible and resulted in broad but sparse data. Labs predicted heart failure diagnosis (AUC 0.94, sensitivity 74%, specificity 94%, Phi 0.58) and readmissions (AUC 0.87, sensitivity of 62%, specificity 98%, Phi 0.18). A-ECG predicted LVSD (AUC 0.92, 95% CI 0.86 to 0.96, P<0.0001), mortality and ICD use. Deep learning (DL) applied to echo images derived metrics similar to a human reader, at x30 the speed. DL discriminated HFpEF from healthy (AUC0.99), and k-means clustering identifed 3 phenogroups within HFpEF that had varying outcomes. Metabolomics identified novel biomarkers of heart failure e.g. breath acetone. Existing clinical data is a rich source of information for machine learning and has significant potential for impacting on clinical care. More resources should be applied to this area.

Each year, approximately 5000 New Zealanders are admitted to hospital with a first-time acute coronary syndrome (ACS). The aim of the Multi-Ethnic New Zealand Study of Acute Coronary Syndromes (MENZACS) is to examine the relationship of clinical, genomic and cardiometabolic markers in relation to presentation and outcomes post-ACS. MENZACS is a prospective, longitudinal cohort study embedded within the All New Zealand Acute Coronary Syndrome Quality Improvement (ANZACS-QI) programme in 6 hospitals. Patients with first-time ACS were enrolled and study-specific research data collected alongside the ANZACS-QI registry. Research blood samples were stored for future genetic/biomarker assays. We report here the baseline characteristics of the MENZACS cohort and compare to the ANZACS-QI registry (first-time ACS). Between 2015-2019, 2026 patients were enrolled, mean age 61yrs, 21% were female. Ethnicity and CV risk factor distribution was similar to ANZACS-QI: 13% Māori, 5% Pacific, 5% Indian and 74% NZ European. CV risk factors included 56% ex-/current smoker, 42% hypertension, 18% diabetes. 60% of the patients were aged < 65 years. ACS subtype for MENZACS vs ANZACS-QI: STEMI 40% vs 31%, NSTEMI 52% vs 59%. 99% had coronary angiography and 90% had revascularisation; there were high rates of secondary prevention medications (table).Tabled 1Discharge MedicationAspirin1927 (98%)Dual antiplatelet therapy1611 (82%)Statin1922 (97%)Beta-blocker1652 (84%)ACEi/ARB1550 (79%) Open table in a new tab MENZACS represents a cohort with optimal contemporary management and will be a significant epidemiological bioresource for the study of environmental and genetic factors contributing to ACS in New Zealand's multi-ethnic environment. The study will utilise clinical, nutritional, lifestyle, genomic and biomarker analyses to explore aetiological factors and develop risk prediction models for outcome.

Diagnosis of congenital LQTS can be challenging if the QT interval is normal on conventional 12-lead ECG. Advanced-ECG (A-ECG) analysis uses pattern recognition of both conventional and advanced-ECG parameters, measures from derived vectorcardiography, and waveform complexity from singular value decomposition, to generate probabilistic scores for disease. We aim to validate an A-ECG score for LQTS in a case-control study. Conventional ECG digital files were retrieved from hospital database for LQTS cases (n=21) and age- and sex-matched controls (n=29). A-ECG analysis was performed using conventional and advanced-ECG parameters using logistic regression and linear discriminant analysis applying previously validated LQTS scores. Blinded A-ECG diagnoses were compared with a "crowd" average of three cardiologists. A-ECG had a sensitivity of 81% and specificity of 90% (AUC [95%CI] 0.87 [0.75–0.97]) compared to a single cardiologist (0.74 [0.6-0.86], p = 0.001) or crowd (0.80 [0.67–0.9], p<0.001) and QTc (0.76 [0.61–0.87], p<0.001) in discriminating LQTS from controls (Table 1). 67% of both carriers (n=15) and probands (n=6) had QTc<470ms. A-ECG diagnoses were provided for 3 of 4 asymptomatic gene carriers, labelled false negatives, providing a sensitivity of 95% for detecting cardiac pathology. Average LQTS probability in true versus false positives was 90% and 52% respectively, (95%CI 0.56 to -0.20, p<0.001). False positives had significant alternative cardiac pathology. 80% of A-ECG positive diagnoses of LQTS subtypes (1, 2 and 3), concurred with genetic results in 94% of cases. A-ECG is not only highly sensitive and specific for LQTS but also accurate in identifying genetic subtypes.

A full transthoracic echocardiogram (TTE) study usually takes 40 - 60 mins to perform and report. Our aim was to validate an artificial intelligence (AI) which automatically calculates measurements with manual standard clinical metrics. 41 patients with heart failure (HF) and 19 controls were enrolled retrospectively. A shortened 5-minute TTE exam was performed. Studies were exported from the hospital database in a DICOM format and fed to an AI pipeline to classify, segment and analyse each image. A convolutional neural network (CNN) was used to label each view into one of 23 classes. Views of interest (Apical 2-, 4-Chamber and Parasternal Short/Long Axis) were individually segmented using a segmentation CNN. View classification was trained on 4,000 labelled studies, segmentation models were trained for each view with 72 manually segmented images for PSAX, 128 for PLAX, 168 for A4C and 198 for A2C. The area-length formula was used to calculate left-ventricular volumes (LVEDV/LVESV), ejection fraction (AI-LVEF). Indexed LA volume (LAVOLI) LV mass (LVMI) were also compared. LVEDV, LVESV, LVEF and LVMI were averaged over multiple videos. Mean manual LVEF (M-LVEF) in HF patients was 39±10% vs 57±5% in controls. Compute time using was between 4 to 7 mins for classification, segmentation and analysis using a single Graphics Processing Unit (GPU). 11 (18%) non-physiological AI-ESV and associated AI-LVEF were excluded vs 2 (3%) M-LVEF (×2 7 95% CI 3 to 27%, p=0.008). AI generated measurements correlated well with manual measures LVEDV r=0.77, LVESV r=0.8, LVEF r=0.71, LAVOLI r=0.71, LVMI r=0.6, p<0.005. Mean absolute error of M-LVEF vs AI-LVEF was 7.4±6.6%. AI-LVEF, M-LVEF and other HF biomarkers had a similar discrimination for HF (AUC M-LVEF 0.93 vs AI-LVEF 0.88, 95% CI-0.03 to 0.15, p=0.19). AI vs Manual, Correlation Matrix and ROC AI with minimal human input is approaching the accuracy required for clinical utility. AI has the ability to distinguish LV systolic dysfunction, and chamber volumes which could be applied to handheld ultrasound in real-time. Health Research Council of New Zealand, Auckland Bioengineering Institute

Background: The conventional use of high-sensitivity troponins (hs-troponins) is for diagnosing myocardial infarction however they also have a role in chronic disease management. This pilot study assessed the relationship of hs-troponins with echocardiographic markers of left ventricular hypertrophy (LVH) and structural heart disease. Methods: Patients undergoing CT coronary angiogram for low-intermediate risk chest pain, and healthy volunteers were recruited. Hs-troponins Singulex I, Abbott I and Roche T and NT-proBNP were evaluated in relation to clinical characteristics, interventricular thickness (IVSd), left atrial enlargement (LAE) and composite structural heart disease (SHD) on echocardiography. Results: 78 subjects who underwent echocardiography were included in this study. Male gender and hypertension were associated with higher hs-troponins (p < 0.001). Both Singulex and Abbott hs-troponins had high discriminatory accuracy for all SHD measures (Table 1). Optimal cutpoints for Singulex, Abbott, Roche assays and NT-proBNP for SHD were >1.5 ng/L, >1.7 ng/L, >6.5 ng/L and >6.3 pmol/L respectively. Correlation was strong between the two hs-troponin I assays, Spearman coefficient r = 0.85, but weaker between I and T assays r = 0.46/0.54 (Abbott/Singulex with Roche).Table 1C-statistics (95% confidence interval) of hs-troponin assays at detecting structural heart disease on echocardiography.BiomarkerLVHLAEComposite SHDSingulex hs-troponin I0.86 (0.71-0.92)0.76 (0.64-0.86)0.88 (0.76-0.95)Abbott hs-troponin I0.84 (0.70-0.92)0.80 (0.69-0.89)0.90 (0.79-0.96)Roche hs-troponin T0.78 (0.65-0.88)0.60 (0.48-0.72)0.68 (0.54-0.80)NT-proBNP0.64 (0.44-0.72)0.74 (0.62-0.84)0.69 (0.56-0.81) Open table in a new tab Conclusion: These results advocate the potential role of hs-troponins as screening tools for structural heart disease with theranostic implications.

Introduction: Despite the use of guideline-based selection criteria between 10 to 20% of patients will be non-responders to CRT. We evaluated the predictors of response using pre-implant clinical, echocardiographic, 12L advanced ECG (A-ECG) and the application of machine learning. Methods: 61 consecutive patients referred for CRT were identified. Digital ECG files were processed using a novel algorithm, echocardiographic metadata, clinical factors and outcomes were sourced from electronic clinical records. Statistical analysis using Medcalc and Microsoft Azure Machine Learning (ML) was applied. Results: 15 (26%) patients were considered non-responders, based on clinical and echocardiographic criteria. Non-responders had lower stroke volume, LV mass, as well as larger LVESD and LVOT area. A-ECG univariate predictors of response included QTc (AUC 0.74, 95%CI 0.6 - 0.8, p = 0.0007), QRS area and vectorcardiographic markers of RV dysfunction. A neural network of 4 ECG features yielded an AUC 0.77 compared with an AUC of 0.56 using a control feature set of traditional CRT selection criteria (cardiomyopathy type, QRSd, LVEF, and AF presence). Conclusion: Numerous clinical features are associated with CRT nonresponse. Of these QTc had the highest discriminatory ability, though this was enhanced with multivariate ML.

Background: Genome-wide association studies and population studies have identified single nucleotide polymorphisms (SNPs) associated with increased risk of development and progression of coronary artery disease, and with response to therapeutic intervention. An emerging way of managing the complexity of large numbers of SNPs that modestly predict outcome individually, is to combine them into a genetic risk score (GRS). However, before GRS tools developed in European and North American populations can be utilised in New Zealand, it is important to examine the SNP distribution involved in GRS in our own primary and secondary prevention populations. Methods: We examined the frequency of 27 SNPs from a previously validated GRS in a cohort of patients presenting with myocardial infarction (MI). The observed frequency of each SNP was compared between MI patients identifying as Maori or European. Results: In 683 MI patients, 64 (9.8%) identified as Maori, 572 (83.8%) as European and 47 (6.7%) as other. We observed significantly higher frequency of risk alleles in 11 of the 27 SNPs in Maori compared to European patients, with frequencies 9-28% higher, and in 6 of the 27 SNPs the risk allele was significantly less frequency (9-20%) in Maori patients. Overall, the unweighted GRS was higher in Maori than European patients (28.5 ± 3.0 versus 26.7 ± 3.4, p < 0.0001). Conclusion: We observed significant differences in the distribution of risk alleles in Maori MI patients. Understanding how these differences impact on health outcomes in Maori requires careful evaluation before internationally developed GRS tools can be applied.

Background: Current coronary artery disease (CAD) risk models use age as a significant component of predicting patient risk. Despite this, there is a group of relatively young patients who present with myocardial infarction (MI). Incorporation of a genetic based risk score may help improve risk stratification within this age group. This study examined a 27-genetic variant risk panel in young MI patients. Method: Patients diagnosed with an MI undergoing an invasive approach were prospectively enrolled. iPLEX Mass-array Sequenom genotyping was used to generate a risk score based on 27 known genetic polymorphisms in “young MI” patients ≤50 years. Results: Of 1199 enrolled MI patients, 154 (13%) were classified as “young MI” patients. The average composite genetic risk score in this group was 1.68 (1.63-1.72), which is considerably higher than previously reported risk scores in older MI populations (mean scores 1.28). The young MI group contained a higher proportion of Maori patients than the older MI group (21% versus 10%, p<0.0001), but the genetic risk score was not elevated in Maori versus non-Maori (1.70 vs 1.67, p=0.68). We did see a significant increase in genetic risk in patients with family history of premature CAD compared to those without (1.76 vs 1.60, p<0.0001). Conclusion: We report a 13% rate of young MI patients who have high genetic risk scores. Genetic risk was not associated with ethnicity, but was linked to family history of CAD. How to incorporate genetic risk scores into current risk models, particularly in the young, requires further evaluation.

Background: The impact of long-term microgravity on cardiovascular function may become a critical limitation to human space exploration. Myocardial strain analysis and mathematical cardiac modelling could be used as a predictive tool for cardiac dysfunction in space. We aimed to develop an open-access strain-based model, using speckle tracking, onto which specific astronaut data could be imposed. Methods: 10 patients with cardiomyopathy underwent echocardiography with a GE Vivid 7 and Philips HDI5000. EchoPAC global longitudinal strain (GLS) analysis was applied to the GE data, while two universal strain software programs, Velocity Vector Imaging v3.0 (VVI) (Siemens Corp) and EchoInsight v1.0 (Epsilon Imaging, USA), were applied to raster versions of both sets of images. A novel universal strain modelling software was developed to derive strain (usmGLS) directly from DICOM data and was applied to preflight, inflight (HDI5000) and postflight echocardiograms, on astronauts traveling to the International Space Station (ISS). Results: GLS analysed with the Vivid 7 and EchoPAC correlated with universal software systems: VVI r = 0.92, p <0.0001 (mean difference -1+/- 4) and EchoInsight r = 0.84, p = 0.001 (mean difference -3 +/- 9). GLS for the same patients on the HDI5000 was also comparable r = 0.69, p = 0.002 (mean difference -3 +/- 7). The open-source usmGLS had comparable correlation with VVI r = 0.72, p <0.05 when applied to astronaut data. usmGLS reduced in microgravity (mean 2.8%, p<0.05) and decreased with days in flight (Fig. 1. r = 0.16, p = 0.5). Conclusion: Raster-based, speckle tracking strain analysis from the ISS echo model provides comparable strain values to contemporary standards. A universal strain software platform has been developed which can be applied to any input, including Vscan avi files. Source code is freely available at https://github.com/ABI-Software-Laboratory/ICMA.

Background: Exhaled human breath contains a plethora of volatile organic compounds (VOCs) and is a promising source of novel biomarkers. This complex network of VOCs and the human volatome, derived from skin, urine, faeces and other sources have significant potential for cardiovascular theranostic applications. Methods: Candidate biomarkers were identified from a literature search and included VOCs known to be associated with glycolysis, gut microbial metabolism, inflammation and oxidative stress. Acetone, ethanol, ethyl benzene, methyl nitrate were identified as predictors of plasma glucose. In addition to acetone, 8-isoprostane, pentane, nitric oxide were considered biomarkers for heart failure. Trimethylamine and carbon monoxide are biomarkers for both heart failure and coronary artery disease. Acute coronary syndrome biomarkers include various methylated hydrocarbons and isoprene. Objectives: Using a 3D printer we have developing a novel positional sensing platform, containing a filter, dessicant, temperature, humidity and pressure sensors to capture and analyse exhaled breath, using a novel sensor array (Fig. 1). The array comprises 32 sensors fabricated using functionalised carbon nanotubes. The device will also measure multiple human vital signs, such as heart rate, respiration rate and core body temperature as well as posture. Advanced multivariate analytics will be used to identify VOC networks and patterns of disease using wireless sensing and a centralised cloud-server. Pilot studies will evaluate this sensor apparatus in patients with diabetes, heart failure, acute coronary syndromes and normal subjects. Conclusion: A multifunctional sensory ‘Tricorder’-type platform will be developed with supportive computational flow and personalised modelling to validate the clinical applications of a novel carbon nanotube-based sensor array.

Background: ETT has a sensitivity (68%) and specificity (77%) for angiographic coronary artery disease (CAD), leading to missed diagnoses as well as to false positive results. Advanced ECG (A-ECG) is a low-cost, software method for improving the diagnostic characteristics of traditional resting 12L-ECG. Methods: We retrospectively performed A-ECG analysis on the electronically stored, 10-sec resting 12L-ECG files of 58 patients with intermediate to high clinical suspicion of CAD who had positive ETT tests that prompted coronary angiography. A-ECG calls of ‘Disease’ versus ‘No Disease’ were made in a blinded and automated fashion on the stored data using a validated 6-parameter A-ECG score that combines results from certain conventional ECG parameters with those from T-wave complexity and derived 3-dimensional ECG. Results: Of the 58 patient 12L-ECGs, one was excluded due to noise and six due to other proven cardiac abnormalities found on echocardiography (e.g., LVH, systolic dysfunction). Resting A-ECG returned 42 ‘Disease’ and 9 ‘No Disease’ (e.g. Figure 1) calls with a sensitivity of 97.2% and specificity of 53% for the angiography results. By design in this retrospective study with known referral-bias, the positive ETT results were 71% sensitive and 0% specific for the same angiography results in the same patients. Conclusion: In this group of patients who underwent angiography prompted by positive ETT, the addition of A-ECG test to the ETT would have substantially increased overall specificity without statistically significantly compromising sensitivity. AECG would theoretically reduce unnecessary invasive coronary angiograms by 53%, with a cost savings of $528/A-ECG.

Background: Clinical use of 2D speckle-tracking echocardiography (2D STE) has been limited by vendor-specific software applied to proprietary scan line data stored in a polar (P) format. Extending 2D STE to raster-based DICOM (D) images would greatly expand its application. We sought to compare strain (S) analysis of the same images stored in P and D formats.