Time series clustering of left atrial strain curves for risk stratification in the general population

Objective

Currently, only the peak reservoir left atrial (LA) strain has been in use while a huge amount of useful information during different phases of cardiac cycle has been ignored. In this study, therefore, we tested the hypothesis that an unsupervised time series analysis utilizing the whole LA deformation curves will identify distinct clusters associated to risk factors and improve cardiovascular (CV) risk prediction over the traditionally used covariables in the general population.

Design and method

We prospectively studied 929 community-dwelling individuals (mean age, 51.6 years; 52.9% women), in whom we acquired clinical and echocardiographic data including LA strain traces at baseline and collected cardiac events on average 7.2 years later. We employed two unsupervised time series techniques such as Self-Organizing map and Variational Autoencoder (VAE) with Convolutional Neural Network as building block to cluster spatiotemporal patterns of LA strain curves. Clinical characteristics and cardiac outcome were used to evaluate the validity of the clusters (k).

Results

According to the quantization error value for every k, the optimal number of clusters was 5 for applied methods. Figure 1 (left panel) illustrates the centroids of the clusters using the proposed VAE network. The first three clusters had differences in sex distribution and heart rate, but had similar low CV risk profiles. On the other hand, cluster 5 had the worst CV risk factors combination, and higher prevalence of left ventricular remodelling and diastolic dysfunction (ie, lowest e' velocity and highest E/e') compared to other clusters. We also observed an increase in the risk for incidence of adverse events between cluster 5 and other clusters (Figure 1, right panel). After adjustment for traditional risk factors, cluster 5 had the highest risk of cardiac events as compared to cluster 1, 2 and 3 (HR: 1.36; 95% CI: 1.10–1.69); P=0.0046).

Conclusion

Unsupervised Machine/Deep Learning algorithms employed on time series LA strain curves identifies clinically meaningful clusters of LA deformation, and provides incremental prognostic information over traditional risk stratification.

Funding Acknowledgement

Type of funding sources: Public Institution(s). Main funding source(s): KU Leuven

Echocardiographic diversity associated with exercise capacity in heart failure precursor stage B: the Project Baseline Health Study

Background

The current paradigm to stage heart failure (HF) defines precursor stage B (or “pre-HF”) as having any subclinical change in cardiac structure or function. Yet, HF risk factors and type of cardiac abnormality may induce clinically relevant heterogeneity in HF stage B. Therefore, we assessed clinical and echocardiographic heterogeneity within stage B HF and its association with exercise capacity in a large community-based sample.

Methods

2071 participants to the Baseline Health Study (50.4±17.0 years, 56.2% women, 37.8% hypertensive) underwent echocardiography and physical performance testing including 6-minute walk (6MWT) and treadmill exercise test. We defined echocardiographic profiles of left and right heart remodeling and dysfunction using sex-specific internal reference values. We assessed HF stages (0-A-B-C-D) following HF societies recommendations. stage B participants were stratified according to presence/absence of HF risk factors and to the most severe echocardiographic abnormality present (reduced ejection fraction (EF), left ventricular (LV) hypertrophy/diastolic dysfunction or other abnormalities). We reported associations between physical performance metrics and HF (sub)stages.

Results

Stage B HF was present in 516 participants (24.9%). Within stage B HF, we observed a large diversity in echocardiographic profiles. Yet, stage B participants without HF risk factors (n=96, 18.6% of stage B) predominantly presented echo abnormalities other than LV diastolic dysfunction, hypertrophy and reduced EF, while their physical performance profile resembled that of people with normal echocardiography without HF risk factors. In contrast, stage B participants with HF risk factors (n=420) were characterized by LV diastolic dysfunction, hypertrophy or reduced EF, three phenotypes associated with lower 6MWT distance and lower exercise capacity. Concomitant presence of HF risk factors and LV dysfunction/hypertrophy was associated with worst physical performance.

Conclusions

We observed a wide clinical and echocardiographic diversity affecting physical performance in HF precursor stage B when defined by the current staging paradigm. Concomitant presence of HF risk factors and LV dysfunction/hypertrophy may mark individuals at highest risk for progression towards overt HF.

Funding Acknowledgement

Type of funding sources: Other. Main funding source(s): This research was made supported by an institutional research grant from Verily Inc. (CA, USA) and by the Research Foundation Flanders (Belgium).

Personalized remotely guided preventive exercise therapy for a healthy heart: protocol and design of the PRIORITY study

Funding Acknowledgements

Type of funding sources: Public grant(s) – National budget only. Main funding source(s): Research Foundation – Flanders (FWO)

Introduction

Approximately half of the heart failure population has heart failure with preserved ejection fraction (HFpEF), a chronic disease starting with cardiovascular risk factors such as hypertension, diabetes and obesity (stage A) which can further emerge in a structural heart disease without (stage B) or with (stage C) signs or symptoms. Structured exercise therapy is recommended as a Class IA intervention in national and international guidelines and, as such, should be offered to all patients. Yet, in current practice, exercise therapy is often only offered within a secondary prevention program. At present, there exists no comprehensive preventive care program that includes structured exercise for patients in the early stages of heart failure, when cardiovascular risk factors are present, but cardiac remodeling and dysfunction might still be reversible or even preventable.

Purpose

PRIORITY aims to investigate the use of remotely guided exercise therapy as a preventive clinical and cost-effective treatment in the HFpEF continuum. This includes both prevention of progression of asymptomatic diastolic dysfunction towards symptomatic HFpEF (= primary prevention) and delaying progression of symptomatic HFpEF (= secondary prevention).

Methods

A randomized controlled multicenter trial will be conducted in 450 patients (men and women, aged 35-80 years) with heart failure (n = 180 stage A, 180 stage B, 90 stage C). Participants are being recruited from 3 different hospitals and the general population during a 16-month period which started in September 2021. Patients will be randomized (1:1) to usual care or to the PRIORITY exercise intervention (i.e. a combination of supervised with remotely guided home-based training sessions). Training prescription is based on the EXPERT tool and includes person-tailored endurance and dynamic strength training. During one year, participants will receive 18 supervised exercise sessions supplemented with a structured progressive home-based exercise program. Outcomes will be assessed at baseline, 4 months, one and two-years. Primary outcome is the proportion of patients with a clinically relevant improvement in peak oxygen uptake at one-year. Secondary outcomes include vascular health, muscle metabolism, change in electrocardiographic parameters and physical fitness parameters (muscle strength, body composition). Further, big data of physical activity collected during the trial will be used to develop models using machine-learning algorithms which can predict physical activity uptake and changes in fitness to facilitate the creation of more personalized interventions and better tailored exercise prescription.

Conclusion

We anticipate that the PRIORITY study will contribute to better prevention of heart failure thanks to an early easily accessible person-tailored exercise intervention.

Proteomic profiling for detection of early-stage heart failure in the community

Background and purpose

Biomarkers may provide insight into the molecular mechanisms underlying cardiac remodelling and dysfunction. Using a targeted proteomic approach, we aimed to identify circulating biomarkers associated with early-stage heart failure and extract a proteome-based risk classifier for this condition.

Methods

575 community-based participants (mean age, 57 years; 51.7% women) underwent echocardiography and proteomic profiling (CVD II panel, Olink Proteomics). We applied partial least squares-discriminant analysis (PLS-DA) and a machine learning algorithm (extreme gradient boosting, XGBoost) to identify key proteins associated with echocardiographic abnormalities. We used Gaussian Mixture modelling for unbiased clustering to construct phenogroups based on influential proteins in PLS-DA and XGBoost.

Results

Of 87 proteins, 13 were important in PLS-DA and XGBoost modelling for detection of left ventricular (LV) remodelling, LV diastolic dysfunction and/or left atrial reservoir dysfunction: placenta growth factor, kidney injury molecule-1, prostasin, angiotensin-converting enzyme-2, galectin-9, cathepsin L1, matrix metalloproteinase-7, TNFR superfamily members 10A, 10B and 11A, interleukins-6 and 16 and alpha-1-microglobulin/bikunin precursor. Based on these proteins, the clustering algorithm divided the cohort into two distinct phenogroups, with each cluster grouping individuals with a similar protein profile. Participants belonging to the second cluster (n=118) were characterized by an unfavourable cardiovascular risk profile and adverse cardiac structure and function. The adjusted risk of presenting cardiac maladaptation was higher in this phenogroup than in the other cluster (P<0.0001).

Conclusion

We identified proteins reflecting renal function, extracellular matrix remodelling, angiogenesis and inflammation to be associated with echocardiographic signs of early-stage heart failure. Focused proteomic phenomapping discriminated individuals at high risk for cardiac maladaptation in the community.

Funding Acknowledgement

Type of funding sources: Public grant(s) – National budget only. Main funding source(s): Research Foundation Flanders

Temporal shift and accuracy of machine learning in heart transplant outcomes

Background

Accurate prediction of outcomes following a heart transplant is critical to explaining risks and benefits to patients and decision-making when considering potential organ offers. Given the large number of potential variables to be considered, this task may be most efficiently performed using machine learning (ML).

Purpose

We trained and tested different ML algorithms to accurately predict outcomes following a cardiac transplant using the United Network of Organ Sharing (UNOS) database.

Methods

We included 67 939 adult and pediatric patients enrolled in the UNOS database between January 1994 and December 2016 who underwent cardiac transplantation (median age 53 [IQR 38 – 60], 72.7% males). In our models, as an input, we included 114 features that have been collected from recipients and donors prior to transplant. The primary outcome was all-cause mortality at one-year post-transplant. We evaluated three different ML methods: XGBoost, Random Forest (RF) and L2 regularized logistic regression. Algorithms were trained and tested using shuffled 10-fold cross-validation (CV) as well as rolling CV. In the rolling CV, to mimic prospective procedure, ML models were trained by incrementally adding patients according to transplant year and testing models on the data in the following year. The hyperparameters, controlling the learning process, were tuned using Bayesian optimization. Prognostic accuracy for one-year all-cause mortality was characterized using the area under the receiver-operating characteristic curve (AUC).

Results

In total, 8,394 patients died within 1 year of transplant. We observed a substantial difference in prognostic accuracy between the shuffled 10-fold CV and the rolling CV. In the 10-fold CV, XGBoost and RF achieved high predictive performance with AUC of 0.848 (95% CI: 0.842–0.854) and 0.891 (95% CI: 0.886–0.896), respectively. In the rolling CV, which is a more realistic setting, AUC dropped to 0.673 (95% CI: 0.661–0.684) for XGBoost and 0.670 (0.657–0.683) for RF. Predictive performance of L2 regularized logistic regression remained stable across the two CV procedures, achieving AUC 0.669 (95% CI: 0.662–0.676) in the 10-fold shuffled CV and 0.665 (95% CI: 0.649–0.680) in the rolling CV procedure (Figure).

Conclusions

Our study suggests that ML models could be used to predict mortality in the first year post-transplant. We also show that the choice of CV procedure is crucial for evaluating ML models, particularly in data collected over a long period of time. The difference between the shuffled and rolling CV in the predictive performance of the tree-based ML models might indicate temporal dataset shift. In the rolling CV, all three methods achieved similar predictive performance.

Funding Acknowledgement

Type of funding sources: Public grant(s) – National budget only. Main funding source(s): Research Foundation Flanders (FWO)

Prognostic value of strain rate during isovolumic relaxation in a general population

Background

Left ventricular (LV) strain rate (SR) during isovolumic relaxation (SRIVR) has been shown to correlate with invasive measurements of diastolic function, namely the time constant of LV relaxation (τ), and has thus been proven useful in the assessment of diastolic function. Tissue Doppler imaging (TDI) has an adequate frame rate to resolve the SR during a short-lived mechanical event such as IVR.

Purpose

The purpose of this study was to assess the additive prognostic value of SRIVR on top of conventional cardiovascular risk factors in a general population.

Methods

We included 670 subjects (age: 51.2±14.2, 48.8% males) that were already recruited in the Flemish Study on Environment, Genes and Health Outcomes (FLEMENGHO), from May 2005 to February 2009. Subjects were followed up on average 5 years after their recruitment, either by a follow-up visit or by telephone. Exclusion criteria at baseline were atrial fibrillation, presence of an artificial pacemaker, more than mild valvular disease and segmental wall motion abnormalities. All patients underwent echocardiographic examination with a state of the art ultrasound machine. Using an in-house developed software package (SPEQLE), we extracted the velocity, strain and SR curves from the color TDI images (FR >180Hz) and imposed timing information on the IVR based on valve opening/closing as determined from PW Doppler data. The sample volume was positioned at the mid portion of the inferolateral wall, in an apical 3 chamber view, manually tracked over the cardiac cycle and all curves were averaged over 3 subsequent cardiac cycles. Then, SRIVR was estimated as the peak SR value during IVR (Fig. 1). Outcome data consisted of major adverse cardiac events (MACE) during the follow-up period. The hazard ratio (HR) associated with SRIVR values was estimated using Cox regression analysis; we included age, sex, body mass index, systolic blood pressure, smoking and serum cholesterol as co-variables in the model.

Results

An accurate assessment of the SRIVR in the inferolateral wall was not possible in 34 participants, so further analysis was confined to 636 subjects. In total, 65 adverse cardiac events were recorded over the period of 8.7 years of follow-up. Figure 2 demonstrates the cumulative incidence estimates (1-Kaplan-Meier survival estimates) of composite cardiac events in quartiles of SRIVR measured in the inferolateral wall (log-rank test p=0.005). Overall, after adjustment for the important cardiovascular risk factors, SRIVR of the inferolateral wall analyzed as a continuous variable was a significant predictor of fatal and nonfatal cardiac events (HR 1.94 (95% CI 1.09–3.47); p=0.025).

Conclusion

SRIVR measured in the inferolateral wall is an important biomarker not only in assessing diastolic function but also as a significant predictor of future adverse outcomes.

Funding Acknowledgement

Type of funding sources: None.

The association of soluble ACE2 change with metabolic health, body composition and proteome dynamics during a weight loss diet intervention: implications for the COVID-19 pandemic

Background

Angiotensin-converting enzyme 2 (ACE2) serves protective functions in metabolic, cardiovascular, renal and pulmonary diseases and is linked to COVID-19 pathology. We explored the association between soluble AC2 (sACE2) and metabolic health and proteome dynamics during a weight loss diet intervention.

Methods

We analyzed 457 healthy individuals (mean age 39.8±6.6) with BMI 28–40 kg/m2 who participated in the Diet Intervention Examining the Factors Interacting with Treatment Success (DIETFITS). Biochemical markers of metabolic health and 236 proteins measured by Olink CVD II, CVD III and Inflammation arrays were available at baseline and 6 months following dietary intervention. We determined clinical and routine biochemical correlates of the diet-induced change in sACE2 (ΔsACE2) using stepwise linear regression. We then combined feature selection models and multivariable-adjusted linear regression to identify protein dynamics associated with ΔsACE2.

Results

sACE2 decreased significantly on average at 6-months in the diet intervention. A stronger decline in sACE2 during the diet intervention was independently associated with female sex, lower HOMA-IR and LDL cholesterol at baseline, and a stronger decline in HOMA-IR, triglycerides, HDL-cholesterol and fat mass. In line, participants with decreasing HOMA-IR and triglycerides had significantly higher odds for a decrease in sACE2 during the diet intervention than those who did not (P≤0.0073 for both). Feature selection models linked ΔsACE2 to changes in AMBP, E-selectin, HAOX1, KIM-1, MERTK, PGF, thrombomodulin and TRAIL-R2. ΔsACE2 remained independently associated with these protein changes in multivariable-adjusted linear regression.

Conclusion

Decrease in sACE2 during a weight loss diet intervention was associated with improvements in metabolic health, fat mass and markers of angiotensin peptide metabolism, vascular injury, renal function, chronic inflammation and oxidative stress. Our findings may improve the risk stratification, prevention, and management of cardiometabolic and COVID-19-related complications.

Funding Acknowledgement

Type of funding sources: Public grant(s) – National budget only. Main funding source(s): NIHResearch Foundation Flanders

Left atrial reservoir strain in relation to metabolic and inflammatory biomarkers: a community-based study

Background

The pathophysiological mechanisms that underlie progressive left atrium (LA) remodelling and dysfunction are only partially understood. Metabolic disturbances and chronic inflammation might mediate LA dysfunction. To date, population data investigating the contribution of these processes to LA reservoir dysfunction remain scarce.

Purpose

In a large population sample, we investigated the association between LA reservoir function and a panel of 38 metabolic and inflammatory biomarkers.

Methods

In 1236 community-dwelling individuals (mean age, 51.0 years; 51.5% women), we echocardiographically assessed LA reservoir strain (LARS) using 2D speckle-tracking analysis. LA reservoir dysfunction was defined as having LARS <23%. We applied partial least squares-discriminant analysis (PLS-DA) to identify biomarkers associated with LA dysfunction. We further explored the associations between LARS and selected biomarkers that were the most influential in PLS-DA, while adjusting for important clinical correlates such as age, sex, body mass index (BMI), heart rate, systolic blood pressure (BP) and antihypertensive treatment. We applied stepwise regression to identify the clinical features and circulating biomarkers most valuable for prediction of abnormal LARS.

Results

The three latent factors constructed from the panel of 38 biomarkers during PLS-DA explained 16.9% of the variation between the normal and the impaired LA function group. The PLS-DA model discriminated between normal and abnormal LA reservoir strain with 79% accuracy (P<0.0001). In PLS-DA, serum uric acid, serum insulin, γ-glutamyl transferase, interleukin-6, D-dimer and triglycerides were the top biomarkers responsible for class discrimination. On average, these top biomarkers were higher in the LA dysfunction group as compared to their normal counterparts (P<0.0001 for all). In multivariable-adjusted continuous analyses, LARS decreased significantly with the level of serum insulin, serum uric acid and γ-glutamyl transferase (P≤0.0035 for all). Of the clinical correlates and the top biomarkers selected in PLS-DA, stepwise regression models highlighted age, BMI, systolic BP, serum insulin, serum uric acid and interleukin-6 as the main predictors of an impaired LA reservoir function (see figure). Conjointly, these clinical and biochemical features identified LA reservoir dysfunction with an overall accuracy of 85%.

Conclusions

Circulating markers of insulin resistance, hyperuricemia and chronic inflammation were independently associated with impaired LA reservoir function. These markers may help to further unravel the pathophysiological processes behind LA maladaptation and improve the management of early LA dysfunction in the community.

Funding Acknowledgement

Type of funding source: Public grant(s) – National budget only. Main funding source(s): Research Foundation Flanders

Echocardiographic phenogrouping by machine learning for risk stratification in the general population

Background

Improved taxonomical classification of routinely measured echocardiographic parameters is needed for better phenotypic characterisation of the asymptomatic stages of cardiac maladaptation. This would create opportunities to intervene early in the course of the heart disease and prevent progression to more advanced stages and adverse events.

Purpose

We tested the hypothesis that an unbiased clustering analysis utilizing echocardiographic indexes reflecting left heart structure and function and hemodynamic measurements could identify phenotypically distinct groups of asymptomatic individuals in the general population.

Methods

We prospectively studied 1407 community-dwelling individuals (mean age, 51.0 years; 51.5% women), in whom we performed clinical and echocardiographic examination at baseline and collected cardiac events on average 8.5 years later. Cardiac phenotypes that were correlated at r>0.8 were filtered, leaving 15 less redundant echocardiographic and hemodynamic features for phenogrouping. We employed two methods of unsupervised machine learning: agglomerative hierarchical clustering and Gaussian mixture using expectation minimization algorithm. The optimal number of phenogroups was chosen based on the combination of the cohesion/separation approach (silhouette index) and stability. Cox regression was used to demonstrate the clinical validity of phenogroups.

Results

Overall, both methods agreed with respect to cluster assignment (RI=0.75). Unbiased clustering analyses classified study participants into 3 distinct phenogroups that differed markedly in cardiac structure/function indexes and hemodynamics used for cluster analysis (Figure, left panel). Indeed, phenogroup 3 had the worst left ventricular diastolic function (ie, lowest e' velocity and left atrial reservoir strain, but highest E/e', deceleration time, and left atrial volume index), highest left ventricular mass index, as well as highest systolic blood pressure and pulse pressure (Figure, left panel). The phenogroups were also different in other clinical characteristics and incidence of cardiac events (Figure, right panel). Even after adjustment for traditional risk factors, phenogroup 3 had the highest risk of cardiac events as compared to cluster 1 (Figure, right panel).

Conclusions

Unsupervised learning algorithms integrating routinely measured cardiac imaging and hemodynamic data can provide a clinically meaningful classification of cardiac health in asymptomatic individuals. This approach might facilitate early detection of cardiac maladaptation and improve risk stratification.

Funding Acknowledgement

Type of funding source: Public grant(s) – National budget only. Main funding source(s): Research Foundation Flanders

Subclinical left atrial dysfunction profiles for prediction of cardiac outcome in the general population

Background

Echocardiographic screening for subclinical left atrial (LA) dysfunction might enhance the prediction of cardiac diseases such as atrial fibrillation (AFib) in the community. To date, however, outcome-driven definitions of early-stage LA dysfunction remain scarce.

Purpose

In a large population sample, we sought to extract outcome-driven thresholds for echocardiographic indexes of LA function to define subclinical left atrial (LA) dysfunction and assess the prognostic value of these echocardiographic criteria for prediction of cardiac events.

Methods

In 1306 community-dwelling individuals (mean age, 50.7 years; 51.6% women), we assessed LA function and left ventricular (LV) global longitudinal strain (LS) by echocardiography. Using receiver-operating curve threshold analysis, we derived outcome-driven cut-offs for LA emptying fraction (LAEF) and LA reservoir strain (LARS) that best balanced the cardiac event prediction (i.e. cut-offs yielding the highest Youden index (=sensitivity+specificity-1)). Next, we constructed LA dysfunction profiles and integrative LA/LV strain profiles based on the extracted cut-offs for LAEF and LARS and a validated definition of impaired LV global LS. We assessed the prognostic performance of these profiles in predicting the incidence of cardiac events and AFib (mean follow-up, 8.5 years).

Results

During follow-up, 93 participants experienced a cardiac event (8.3 events/1000 person-years) and 27 developed AFib (2.3 events/1000 person-years). LAEF<55% and LARS<23% yielded the highest Youden indexes and thus provided the most balanced prediction of incident AFib. When applying these cut-offs, abnormal LAEF and LARS were respectively present in 27.0% and 18.1% of the cohort. Abnormal LARS was independently associated with higher risk for cardiac events (hazard ratio (HR) versus normal LA phenotype: 2.11, P=0.0021). Both abnormal LAEF (HR: 2.57) and abnormal LARS (HR: 3.28) predicted incident AFib (P≤0.029). As compared to subjects free from any LA dysfunction, those with both LAEF<55% and LARS<23% had a significantly higher risk to develop cardiac events (HR: 2.10; P=0.014) and AFib (HR: 6.45; P=0.0036). Of the integrative LA/LV strain profiles, the concomitant presence of an impaired LARS and LV global LS independently elevated the risk for cardiac events (HR: 2.81; P=0.0012) and AFib (HR: 4.36, P=0.0071) as compared to normal counterparts. Both the degree of LA dysfunction and the integrative LA/LV strain profiles improved the prognostic accuracy beyond clinical risk models and risk scores.

Conclusions

We validated population-based and outcome-driven definitions of subclinical LA dysfunction predicting cardiac events independent of conventional risk factors. Echocardiographic screening for subclinical LA and LV systolic dysfunction might enhance the prediction of cardiac diseases such as AFib in the community, empowering clinicians to timely intervene with the disease development.

Prediction of cardiac events

Funding Acknowledgement

Type of funding source: Public grant(s) – National budget only. Main funding source(s): Research Foundation Flanders, Internal Funds KU Leuven

P2488The 2013 ACC/AHA pooled cohort equations and insulin resistance status for detection of early-stage heart failure in the community

Objectives

Detection of heart failure (HF) in its subclinical phase would allow timely initiation of preventive measures that counter its pathophysiology. Here, we assessed the usefulness of traditional cardiovascular (CV) risk assessment and insulin resistance status to detect early-stage HF.

Methods

In 984 participants (mean age, 57.0 years, 52.3% women), we derived echocardiographic indexes of left ventricular (LV) structure and function and calculated the 10-year risk for a first atherosclerotic CV disease (ASCVD) using the 2013 ACC/AHA risk score. We assessed the discriminatory value of this risk score to detect LV maladaptation and the improvements in reclassification by insulin resistance status (HOMA-IR).

Results

The probability for LV maladaptation rose progressively with the 10-year ASCVD risk increasing. Participants at high 10-year ASCVD risk (>7.5%) had indeed significantly higher odds for LV concentric remodeling (odds ratio, 4.84), LV hypertrophy (OR, 5.93), abnormal LV longitudinal strain (OR, 2.04) and LV diastolic dysfunction (OR, 25.3) as compared to those at low ASCVD risk (<2.5%; P≤0.0003). Adding markers of insulin resistance to the ACC/AHA risk score moderately improved the integrated discrimination and net reclassification of all LV maladaptive phenotypes (P≤0.022) except LV diastolic dysfunction (P≥0.059). LV remodeling and abnormal LS was particularly more likely in insulin-resistant participants with a 10-year ASCVD risk between 5% and 15% than in their insulin-sensitive counterparts.

Prediction of early-stage HF profiles 2013 ACC/AHA risk score Addition of insulin resistance status to the 2013 ACC/AHA risk score AUC (95% CI) Integrated Discrimination Improvement Net Reclassification Improvement Absolute IDI (%) P value NRI (95% CI) P value LV concentric remodeling 0.70 (0.66 to 0.74) 0.0083 (11.3%) 0.022 0.23 (0.067 to 0.39) 0.0058 LV hypertrophy 0.70 (0.66 to 0.74) 0.017 (20.7%) 0.0033 0.27 (0.11 to 0.43) 0.0011 Abnormal LV LS 0.56 (0.53 to 0.62) 0.022 (202.0%) <0.0001 0.33 (0.18 to 0.49) <0.0001 LV diastolic dysfunction 0.82 (0.78 to 0.86) 0.0007 (0.45%) 0.84 0.093 (−0.11 to 0.30) 0.38 ≥2 LV abnormalities 0.76 (0.72 to 0.80) 0.0087 (7.3%) 0.071 0.22 (0.042 to 0.40) 0.016 The integrated discrimination improvement (IDI) and net reclassification improvement (NRI) reflect the improvements in classification by adding insulin resistance (by HOMA-IR) to the 2013 ACC/AHA risk score. HOMA-IR, Homeostatic Model for Assessment of Insulin Resistance; LS, longitudinal strain; LV, left ventricular.

Risk enhancers of LV maladaptation

Conclusions

The 2013 ACC/AHA risk score adequately captured the risk for echocardiographic phenotypes of early-stage HF. As risk enhancer, insulin resistance might improve risk stratification of subclinical HF in subjects at intermediate risk.

Acknowledgement/Funding

The European Union, European Research Council and the Flanders Scientific Research Fund supported this study.

P3819Machine learning for predicting early left ventricular abnormalities in the general population

Background

Current heart failure guidelines emphasize the importance of timely detection of subclinical left ventricular (LV) remodelling and dysfunction for more precise risk stratification of asymptomatic subjects. Both LV diastolic dysfunction (LVDD) and LV hypertrophy (LVH) as assessed by echocardiography are known independent prognostic markers of future cardiovascular events in the community. However, selective screening strategies of individuals at risk who would benefit most from in-depth cardiac phenotyping are lacking.

Purpose

We assess the utility of several Machine Learning (ML) classifiers built on clinical and biochemical features for detecting subclinical LV abnormalities.

Methods

We included 1407 participants (mean age, 51 years, 51% women) randomly recruited from the general population. We used echocardiographic parameters reflecting LV diastolic function and structure to define LV abnormalities (LVDD, n=239; LVH, n=135). After that four supervised ML algorithms (Random Forest (RF), Gradient Boosting (GD), Stochastic Gradient Descent (SGD) and Support Vector Machines (SV)) were built based on routine clinical, hemodynamic and laboratory data (features; n=61) to categorize LVDD and LVH (two prediction tasks). We applied a 10-fold stratified cross-validation set-up.

Results

ML classifiers exhibited a high area under the ROC (AUC) for predicting LVDD with values between 88.5% and 93.1% (Figure, left panel). Age, BMI, different components of blood pressure, antihypertensive treatment, routine biomarkers such as serum electrolytes, creatinine, blood sugar, leptin, uric acid, lipid profile, as well as blood cell counts were the top selected features for predicting LVDD. Prediction AUC of ML algorithms for detection of LVH was somewhat lower than for LVDD and ranged from 72.5% to 78.7% (Figure, right panel). The top selected features for LVH classifier were similar to those of LVDD, but also included social class, serum gamma-glutamyl transferase, fasting insulin, plasma renin activity and cortisol.

ROC curves (sensitivity-1-specificity)

Conclusions

ML algorithms combining routinely measured clinical and laboratory data have shown high accuracy of LVDD and LVH prediction. These ML classifiers might be useful to preselect individuals at risk for further in depth echocardiographic examination, monitoring and implementation of preventive strategies in order to delay transition to disease symptoms.

6075Workload adjusted blood pressure response rather than peak systolic blood pressure is associated with increased all-cause mortality in males; results from 7097 treadmill exercise tests

Background

Systolic blood pressure (SBP) is routinely measured during exercise testing (ET) and is in part determined by cardiac output and peripheral vascular resistance. A frequently used threshold for defining hypertensive response to exercise is ≥210 mmHg but this does not account for the fact that SBP is related to workload, via cardiac output.

Purpose

To examine the prognostic implications of considering external workload (METs) adjusted SBP response to exercise.

Methods

We reviewed all symptom-limited treadmill ET in males between 1987 and 2007 at a single centre (inclusion/exclusion criteria detailed in figure 1A). SBP was measured standing at rest and at peak exercise. Workload adjusted BP response with exercise (SBP/MET slope) was calculated as ΔSBP/ΔMET. METs were calculated from peak speed and grade according to the standard American College of Sports Medicine (ACSM) formula. Age-predicted peak METs was calculated as: 18 - 0.15 × age. Ten-year Cox proportional hazard ratios (HR) with 95% confidence intervals were calculated and adjusted as outlined in figure 1B.

Results

7097 subjects were included, of which 1559 (22%) died within 10 years. Survivors were younger (57.2±10.6 y vs. 64.5±10.3 y, p<0.001) and reached higher % of age-predicted METs (97±33% vs. 82±33%, p<0.001). Survivors had higher peak SBP (181±26 vs. 176±27 mmHg, p<0.001) as well as greater ΔSBP (49±22 vs. 42±22 mmHg, p<0.001), while they had lower SBP/MET slope (7.0±4.4 vs. 8.9±6.5 mmHg/MET, p<0.001). A peak SBP ≥210 mmHg was associated with better survival; 10-yr adjusted HR: 0.76 (0.64–0.88, p<0.001). In contrast, a higher SBP/MET slope was associated with increased mortality (table 1).

Table 1. Ten year adjusted hazard ratios Variable HR (95% CI) P Variable HR (95% CI) P Variable HR (95% CI) P Peak SBP, Q1: 100–159 mmHg REF REF Delta SBP, Q1: 1–29 mmHg REF REF SBP/MET slope, Q1: 0.2–4.2 REF REF Peak SBP, Q2: 160–179 mmHg 0.81 (0.71–0.94) 0.006 Delta SBP, Q2: 30–46 mmHg 0.80 (0.70–0.91) 0.001 SBP/MET slope, Q2: 4.3–6.2 0.95 (0.81–1.12) 0.562 Peak SBP, Q3: 180–199 mmHg 0.68 (0.58–0.78) <0.001 Delta SBP, Q3: 47–61 mmHg 0.76 (0.66–0.88) <0.001 SBP/MET slope, Q3: 6.2–9.1 1.18 (1.01–1.37) 0.032 Peak SBP, Q4: ≥200 mmHg 0.60 (0.51–0.69) <0.001 Delta SBP, Q4: ≥62 mmHg 0.59 (0.50–0.69) <0.001 SBP/MET slope, Q4: ≥9.1 1.40 (1.22– 1.62) <0.001 HR, hazard ratio (adjusted according to figure 1B); SBP, systolic blood pressure; MET, metabolic equivalent of task; Q1–Q4, quartiles (Q1 as reference).

Figure 1

Conclusion

Workload adjusted blood pressure response to exercise in contrast to peak BP response was associated with increased mortality in male patients referred for ET. Of note, reaching a BP of at least 210 mmHg (suggested to define a hypertensive response to exercise) was associated with a 24% reduction in all-cause mortality.

Acknowledgement/Funding

K Hedman was supported by post-doc. grants from the Fulbright Commission, the Swedish Society of Medicine, County Council of Östergötland, Sweden

Development of antibodies that interfere with the collagen-VWF-GPIb axis as new antithrombotics

The epitope and the antithrombotic effect of 6B4, an antibody that inhibits GPIb, the receptor for von Willebrand Factor (VWF) on blood platelets, and of 82D6A3, an antibody against VWF that prevents the binding of VWF to collagen, were characterised. By using canine-human chimeras, alanine-scans, phage display, mutant analysis and modeling both the epitope of 6B4 in the N-terminal domain of GPIb, and of 82D6A3 in the VWF-A3 domain, could be mapped. As both epitopes furthermore are part of the ligand binding sites, this at once also explained the mechanism of the inhibition by the antibodies. Next both antibodies were tested in a thrombosis model in a stenosed artery in baboons, where they showed potent antithrombotic activities, without a noteworthy prolongation of the bleeding time. With this we thus could reveal two new strategies to prevent arterial thrombosis, which presumably may be safer than the currently available antiplatelet agents.

Epitope mapping of inhibitory antibodies against platelet glycoprotein Ibalpha reveals interaction between the leucine-rich repeat N-terminal and C-terminal flanking domains of glycoprotein Ibalpha

The interaction of von Willebrand factor (vWF) with the platelet receptor glycoprotein Ibalpha (GPIbalpha) is important for platelet adhesion at high shear stress. Two functionally important antigenic areas within GPIbalpha were identified through the characterization of 5 new inhibitory anti-GPIb monoclonal antibodies (mAbs). The binding sites of 3 of these anti-GPIb mAbs, which were intercompeting and potently inhibiting shear stress-induced binding of vWF, were mapped within the N-terminal amino acid (aa) 1-59 area by the use of canine-human chimeras. These antibodies, however, had little or no effect (approximately 40% inhibition) on the binding of vWF induced by either botrocetin or ristocetin. On the other hand, the anti-GPIb mAbs 24G10 and 6B4, which blocked GPIb-vWF binding under all conditions examined, bound to 2 different regions of GPIbalpha, aa 1-81 and aa 201-268, respectively. The epitope for 6B4 was further narrowed by phage display revealing 2 sets of peptide sequences aligning within aa 259-262 and aa 230-242. In the latter region of GPIbalpha, the gain-of-function platelet-type von Willebrand disease (PT-vWD) mutations have been identified. Alignment was partially confirmed because the binding of 6B4 to recombinant GPIbalpha fragments carrying either one of the PT-vWD mutations was considerably impaired but not completely abolished. In contrast, mAb 24G10 bound more strongly to mutant PT-vWD GPIbalpha. However, although 24G10 competed with 6B4 for binding to platelets, it bound to an epitope within aa 1-81 of GPIbalpha. In conclusion, 2 functionally important areas within GPIbalpha were identified: one localized within the leucine-rich repeat N-terminal aa 1-59 area and one composed of residues aa 1-81 in close contact with aa 201-268. Moreover, further support is provided for the existence of an intramolecular interaction between the N-terminal flanking (aa 1-81) and C-terminal flanking (aa 201-268) regions. (Blood. 2001;98:652-660)

Fc-receptor dependent platelet aggregation induced by monoclonal antibodies against platelet glycoprotein Ib or von Willebrand factor

In this paper we describe two pathways leading to platelet activation by crosslinking glycoprotein (GP) Ibalpha to the platelet Fc-receptor (FcgammaRII). First the monoclonal antibody (MoAb) 9C8, raised against human platelet GPIbalpha, dose-dependently induced platelet aggregation of citrate-anticoagulated platelet-rich plasma, an effect that can be inhibited by several activation inhibitors. The FcgammaRII-inhibitory MoAb IV.3 was able to prevent the aggregatory effects of MoAb 9C8, indicating that crosslinking of the antigen GPIbalpha to the FcgammaII-receptor is necessary for the activating effect. Secondly we observed a synergistic activating effect of two anti-von Willebrand factor (vWF) MoAbs IC1E7 and B724, both known to enhance vWF binding to GPIbalpha in the presence of shear or ristocetin. When these antibodies are added together to PRP, platelet aggregation is induced without further need for an additional modulator. This effect can be blocked by either MoAb IV.3 or an inhibitory anti-GPIb MoAb, indicating that again the platelet activation results from signaling through FcgammaRII crosslinked to vWF bound to GPIbalpha. In addition, both the anti-GPIb MoAb 9C8, or the two anti-vWF MoAbs 1C1E7 and B724 induce genuine platelet activation, as evidenced by the secretion of ATP and protein tyrosine phosphorylation. These findings with both anti-GPIb and anti-vWF MoAbs add further proof to recent reports demonstrating an interaction between the platelet receptors GPIb and FcgammaRII, suggesting a role for the FcgammaII-receptor in GPIb-related signaling.

Structural determinants within platelet glycoprotein Ibalpha involved in its binding to von Willebrand factor

Platelet adhesion to subendothelial structures upon injury to a vessel wall is one of the first steps in a sequence of reactions critical for the formation of a haemostatic plug, or in diseased vessels for the development of an arterial thrombus. This adhesion process is mediated by an interaction between the glycoprotein (GP) Ib-V-IX complex on the platelet surface with von Willebrand Factor (vWF), associated with collagen on the subendothelial surface. After this initial adhesion, platelets will activate, resulting in recruitment of additional platelets and adherence to each other to form the platelet plug or developing thrombus. Several studies to date have attempted to identify the regions of the GPIb-V-IX complex that are critical for binding to vWF. The vWF binding site is contained in the 45 kDa N-terminal domain of the GPIbalpha chain. This N-terminal domain is characterized by a structural motif consisting of 7 leucine-rich repeats (LRRs), followed by a double disulphide-bonded loop and an anionic sulphated region. This review summarizes recent research efforts elucidating the characteristics of the GPIb-vWF interaction. Potential mechanisms that regulate the GPIb-vWF function are discussed, and advances in identifying functional sequences within GPIba involved in the binding to vWF are reviewed.

Acquired Bernard-Soulier syndrome: a case with necrotizing vasculitis and thrombosis

We describe a patient with positive antinuclear antibodies, polyclonal gammopathy and high level of circulating immunocomplexes, resulting in vascular purpura. In addition, the patient had a slightly prolonged bleeding time and an isolated defect of ristocetin-induced platelet aggregation (RIPA) in platelet-rich plasma (PRP). The patient's plasma also inhibited RIPA in normal PRP and in normal platelet suspension. The activity and multimeric structure of plasmatic von Willebrand factor showed no alteration. We could demonstrate an autoantibody against platelet membrane glycoprotein (GP) Ib, using an ELISA-type assay. These data suggest an acquired Bernard-Soulier syndrome. We suggest that the patient had an immunocomplex-mediated leukocytoclastic vasculitis accompanied by production of antinuclear autoantibodies as well as the presence of an autoantibody against GPIb. The titer of the anti-GPIb antibody, however, was too low to induce significant platelet-type bleeding tendency, only laboratory alterations were found. Moreover, in a later stage of her disease, she developed a severe necrotizing vasculitis which was followed by a deep venous thrombosis.

Characterization of murine anti-glycoprotein Ib monoclonal antibodies that differentiate between shear-induced and ristocetin/botrocetin-induced glycoprotein Ib-von Willebrand factor interaction

Platelet adhesion to vascular subendothelium under conditions of high shear stress is mediated by the platelet glycoprotein (GP) Ib-von Willebrand Factor (vWF) interaction. The aim of this study was to characterize the murine monoclonal antibodies (MoAbs) 27A10 and 28E6, both raised against purified GPIb. The MoAb 27A10 is a potent inhibitor of shear-induced platelet adhesion to collagen type I in a flow chamber at shear rates of 1,300 and 2,700 s(-1). 20 microg/ml of MoAb 27A10, furthermore, could completely block shear-induced aggregation in a modified Couette viscometer at shear rates of 1,000 and 4,000 s(-1). On the other hand, MoAb 27A10 had a negligible effect on botrocetin-induced GPIb-vWF binding and is only a poor inhibitor of the ristocetin-dependent interaction. In contrast, MoAb 28E6 did abolish both the ristocetin- and botrocetin-induced GPIb-vWF binding, whereas it did not block the shear-induced interaction. Thus, we identify here two anti-GPIb MoAbs 27A10 and 28E6 that either preferentially inhibit the shear-induced or the ristocetin/botrocetin-induced platelet-vWF interaction. With these tools it should be possible to more clearly define the mechanisms by which platelets bind to vWF in vivo.

Antithrombotic effect of platelet glycoprotein Ib-blocking monoclonal antibody Fab fragments in nonhuman primates

Platelet adhesion in arterial blood flow is mainly supported by the platelet receptor glycoprotein (GP) Ib, which interacts with von Willebrand factor (vWF) that is bound to collagen at the site of vessel wall injury. Antibody 6B4 is a monoclonal antibody (MoAb) raised against purified human GPIb. MoAb 6B4 inhibits both ristocetin- and botrocetin-induced, vWF-dependent human platelet agglutination. MoAb 6B4 furthermore blocks shear-induced adhesion of human platelets to collagen I. We studied the antithrombotic effect of this inhibitory murine MoAb 6B4 in a baboon model of arterial thrombosis. When injected into baboons, intact IgG and its F(ab')(2) fragments caused almost immediate thrombocytopenia, whereas injection of the Fab fragments alone did not. Fab fragments were subsequently used to investigate their in vivo effect on platelet deposition onto a thrombogenic device, consisting of collagen-rich, glutaraldehyde-fixed bovine pericardium (0.6 cm(2)), at a wall shear rate ranging from 700 to 1000 s(-1). Baboons were either pretreated with Fabs to study the effect of inhibition on platelet adhesion or treated 6 minutes after placement of the thrombogenic device to investigate the effect on interplatelet cohesion. Pretreatment of the animals with bolus doses ranging from 80 to 640 microgram/kg Fab fragments significantly reduced (111)In-labeled platelet deposition onto the collagen surface by approximately 43% to 65%. Only the highest dose caused a significant prolongation (doubling) of the bleeding time. Ex vivo ristocetin-induced platelet agglutination was equally reduced. Treatment with a bolus of 110 microgram/kg Fab fragments after a thrombus was allowed to form for 6 minutes had no effect on further platelet deposition. We therefore conclude that Fab fragments or derivatives of inhibitory anti-GPIb antibodies may be useful compounds to prevent thrombosis.

A reliable and reproducible ELISA method to measure ristocetin cofactor activity of von Willebrand factor

The ristocetin induced binding of vWF to GPIb, which is routinely tested in a platelet agglutination assay, can be reproducibly studied in an ELISA where plasma vWF binds to a captured rGPIb alpha-fragment (His1-Val289) in the presence of ristocetin. This binding is specific since the vWF-GPIb interaction could (i) be blocked by inhibitory anti-GPIb or anti-(vWF A1 domain) monoclonal antibodies (mAbs) and (ii) be enhanced by an anti-vWF mAb that also facilitates ristocetin induced platelet agglutination. Further studies were undertaken to determine whether the test could be used to differentiate vWF from patients with different types of von Willebrand's disease. The median vWF:RiCof activity in controls (n = 24) was 0.75 U/ml, in type 1 vWD patients (n = 17) 0.28 U/ml, in type 2A (n = 18) 0.055 U/ml, in type 2B (n = 4) 0.094 U/ml and in type 3 (n = 3) <0.0005 U/ml. Moreover, the values correlated well with those obtained from the vWF:RiCof-agglutination assay (r = 0.873). The vWF:RiCof-ELISA has several advantages: the use of a recombinant fragment instead of donor platelets results in a more reproducible test with a low inter- and intra-assay variability (<14% CV), the test can further be readily automated and for a single determination, only minimal amounts of patient plasma are required (8 microl).

Effect of route and frequency of administration of apomorphine on growth hormone release in African catfish (Clarias gariepinus)

Apomorphine is known to stimulate growth hormone release in African catfish following an intraperitoneal (IP) injection. In the present study the effect of apomorphine (5 or 20 mg/kg body weight) on plasma GH levels was evaluated after gastro-intestinal or parenteral delivery. Apomorphine increased the plasma GH concentration regardless of the route of administration, indicating that apomorphine can be absorbed from the intestinal tract. The effect of repeated administration of apomorphine differed clearly between the tested doses. Although a single IP injection with 20 mg apomorphine/kg body weight resulted in a clear increase in plasma GH levels, a second injection given 12 hours later was ineffective. In contrast the last of 4 consecutive injections with 5 mg apomorphine/kg body weight given at intervals of 12 hours stimulated the plasma GH levels in a similar way to a single IP injection with the same dose.