Recent advances in machine learning (ML) have made it possible to analyze high-dimensional and complex data-such as free text, images, waveforms, videos, and sound-in an automated manner by successfully learning complex associations within these data. Cardiovascular medicine is particularly well poised to take advantage of these ML advances, due to the widespread digitization of medical data and the large number of diagnostic tests used to evaluate cardiovascular disease. Various ML approaches have successfully been applied to cardiovascular tests and diseases to automate interpretation, accurately perform measurements, and, in some cases, predict novel diagnoses from less invasive tests, effectively expanding the utility of more widely accessible diagnostic tests. Here, we present examples of some impactful advances in cardiovascular medicine using ML across a variety of modalities, with a focus on deep learning applications.

Background

The ECG remains the most widely used diagnostic test for characterization of cardiac structure and electrical activity. We hypothesized that parallel advances in computing power, machine learning algorithms, and availability of large-scale data could substantially expand the clinical inferences derived from the ECG while at the same time preserving interpretability for medical decision-making.

Methods and results

We identified 36 186 ECGs from the University of California, San Francisco database that would enable training of models for estimation of cardiac structure or function or detection of disease. We segmented the ECG into standard component waveforms and intervals using a novel combination of convolutional neural networks and hidden Markov models and evaluated this segmentation by comparing resulting electrical intervals against 141 864 measurements produced during the clinical workflow. We then built a patient-level ECG profile, a 725-element feature vector and used this profile to train and interpret machine learning models for examples of cardiac structure (left ventricular mass, left atrial volume, and mitral annulus e-prime) and disease (pulmonary arterial hypertension, hypertrophic cardiomyopathy, cardiac amyloid, and mitral valve prolapse). ECG measurements derived from the convolutional neural network-hidden Markov model segmentation agreed with clinical estimates, with median absolute deviations as a fraction of observed value of 0.6% for heart rate and 4% for QT interval. Models trained using patient-level ECG profiles enabled surprising quantitative estimates of left ventricular mass and mitral annulus e' velocity (median absolute deviation of 16% and 19%, respectively) with good discrimination for left ventricular hypertrophy and diastolic dysfunction as binary traits. Model performance using our approach for disease detection demonstrated areas under the receiver operating characteristic curve of 0.94 for pulmonary arterial hypertension, 0.91 for hypertrophic cardiomyopathy, 0.86 for cardiac amyloid, and 0.77 for mitral valve prolapse.

Conclusions

Modern machine learning methods can extend the 12-lead ECG to quantitative applications well beyond its current uses while preserving the transparency that is so fundamental to clinical care.

Background

Pulmonary arterial capacitance (PAC) is one of the strongest predictors of clinical outcomes in patients with pulmonary hypertension (PH). We examined the value of an echocardiographic surrogate for PAC (ePAC) as a predictor of mortality in patients with PH.

Methods

We performed a retrospective study of 302 patients with PH managed at a PH comprehensive care center over a cumulative follow-up time of 858 patient-years. Charts from 2004 to 2018 were reviewed to identify patients in whom a right heart catheterization (RHC) was performed within two months of an echocardiogram. Standard invasive, non-invasive, functional, and biochemical prognostic markers were extracted from the time of RHC. The primary outcome was all-cause mortality. Cox proportional hazards models were used to model the time from RHC to the primary outcome or last medical contact.

Results

Variables associated with all-cause mortality included ePAC [standardized hazard ratio (HR) 0.68, 95% CI 0.48-0.98, p = 0.036], RHC-PAC (HR 0.68, 95% CI 0.48-0.96, p = 0.027), echocardiographic pulmonary vascular resistance (HR 1.29, 95% CI 1.05-1.60, p = 0.017), six-minute walk distance (HR 0.43, 95% CI 0.23-0.82, p = 0.01), and B-type natriuretic peptide (HR 1.29, 95% CI 1.03-1.62, p = 0.027). In multivariable-adjusted Cox analysis, ePAC predicted all-cause mortality independently of age, gender, and multiple comorbidities. There was a graded and stepwise association between low (<0.15 cm/mmHg), medium (0.15-0.25 cm/mmHg), and high (>0.25 cm/mmHg) tertiles of ePAC and all-cause mortality.

Conclusions

We have demonstrated that ePAC is a readily available echocardiographic marker that independently predicts mortality in PH, and have provided clinically relevant ranges by which to risk-stratify patients and predict mortality.

Background

Pulmonary arterial capacitance (PAC) is a strong hemodynamic predictor of outcomes in patients with pulmonary hypertension (PH). Its value across subgroups of race/ethnicity, sex, and PH etiologies is unclear. We hypothesized that the association of PAC with outcomes would not vary across World Health Organization (WHO) PH group, race/ethnicity, or sex.

Methods

We performed a retrospective study in patients with PH diagnosed and managed at the Pulmonary Hypertension Comprehensive Care Center of a tertiary care hospital (n = 270). Demographic, diagnostic, treatment, and outcome data were extracted from the electronic medical record. Cox proportional hazards models were used to model time from right heart catheterization to event in univariate and multivariable models. Our primary outcome was all-cause mortality and our secondary outcome was PH hospitalization.

Results

The median age of the cohort was 56 years (±14.6), and 67% were female. In multivariable Cox models adjusted for significant covariates, decreased PAC remained independently and significantly associated with both all-cause mortality (p = 0.029) and hospitalization for PH (p = 0.010). No significant interactions were observed between PAC and race, sex, or WHO group. Hispanic patients exhibited a significant independent association with increased hospitalizations (p = 0.030), and there was a trend toward increased all-cause mortality in African Americans. WHO group 2 PH was associated with more frequent hospitalization (p = 0.004).

Conclusions

Decreased PAC is significantly associated with mortality and hospitalization in PH patients independent of race, sex, and PH subgroups. Further investigation is required to characterize the effects and determinants of racial disparities in PH.

Background

Many individual echocardiographic variables have been associated with heart failure (HF) in patients with stable coronary artery disease (CAD), but their combined utility for prediction has not been well studied.

Methods

Unsupervised model-based cluster analysis was performed by researchers blinded to the study outcome in 1,000 patients with stable CAD on 15 transthoracic echocardiographic variables. We evaluated associations of cluster membership with HF hospitalization using Cox proportional hazards regression analysis.

Results

The echo-derived clusters partitioned subjects into four phenogroupings: phenogroup 1 (n = 85) had the highest levels, phenogroups 2 (n = 314) and 3 (n = 205) displayed intermediate levels, and phenogroup 4 (n = 396) had the lowest levels of cardiopulmonary structural and functional abnormalities. Over 7.1 ± 3.2 years of follow-up, there were 198 HF hospitalizations. After multivariable adjustment for traditional cardiovascular risk factors, phenogroup 1 was associated with a nearly fivefold increased risk (hazard ratio [HR] = 4.8; 95% CI, 2.4-9.5), phenogroup 2 was associated with a nearly threefold increased risk (HR = 2.7; 95% CI, 1.4-5.0), and phenogroup 3 was associated with a nearly twofold increased risk (HR = 1.9; 95% CI, 1.0-3.8) of HF hospitalization, relative to phenogroup 4.

Conclusions

Transthoracic echocardiographic variables can be used to classify stable CAD patients into separate phenogroupings that differentiate cardiopulmonary structural and functional abnormalities and can predict HF hospitalization, independent of traditional cardiovascular risk factors.

Chest pain is a common clinical complaint for which myocardial injury is the primary concern and is associated with significant morbidity and mortality. To aid providers' decision-making, we aimed to analyze the electrocardiogram (ECG) using a deep convolutional neural network (CNN) to predict serum troponin I (TnI) from ECGs. We developed a CNN using 64,728 ECGs from 32,479 patients who underwent ECG within 2 h prior to a serum TnI laboratory result at the University of California, San Francisco (UCSF). In our primary analysis, we classified patients into groups of TnI < 0.02 or ≥ 0.02 µg/L using 12-lead ECGs. This was repeated with an alternative threshold of 1.0 µg/L and with single-lead ECG inputs. We also performed multiclass prediction for a set of serum troponin ranges. Finally, we tested the CNN in a cohort of patients selected for coronary angiography, including 3038 ECGs from 672 patients. Cohort patients were 49.0% female, 42.8% white, and 59.3% (19,283) never had a positive TnI value (≥ 0.02 µg/L). CNNs accurately predicted elevated TnI, both at a threshold of 0.02 µg/L (AUC = 0.783, 95% CI 0.780-0.786) and at a threshold of 1.0 µg/L (AUC = 0.802, 0.795-0.809). Models using single-lead ECG data achieved significantly lower accuracy, with AUCs ranging from 0.740 to 0.773 with variation by lead. Accuracy of the multi-class model was lower for intermediate TnI value-ranges. Our models performed similarly on the cohort of patients who underwent coronary angiography. Biomarker-defined myocardial injury can be predicted by CNNs from 12-lead and single-lead ECGs.

Background

Regular physical activity is an important determinant of cardiovascular health and quality of life. Previous investigations examining the association between exercise and atrial fibrillation (AF) have been limited by self-reported, retrospectively collected activity data.

Objective

The purpose of this study was to objectively quantify differences in daily physical activity among individuals with and those without AF using electronic wearable activity trackers.

Methods

Daily exercise data were directly obtained from wrist-worn activity trackers (Fitbit, San Francisco, CA) among participants in the Health eHeart (HeH) study. Average daily step count was compared between individuals with and those without AF both before and after adjusting for comorbidities. AF severity was quantified using the Atrial Fibrillation Effect on QualiTy of Life (AFEQT) survey.

Results

Among 171,284 HeH study participants, 3333 individuals (234 with AF [7%]) submitted activity data. In unadjusted analysis, AF participants ambulated an average of 723 fewer steps per day (95% confidence interval [CI] 292-1154; P = .001) compared to individuals without AF. After adjustment for demographics and comorbid diseases, participants with AF demonstrated 591 fewer steps per day (95% CI 149-1033; P = .009). Among AF patients, AF severity was associated with less physical activity. For each single point decrease in AFEQT score (corresponding to more symptomatic AF), physical activity decreased by a mean 24 steps per day (95% CI 1-46; P = .04).

Conclusion

Objective, automatically collected step count data demonstrate that individuals with AF engage in significantly less average daily physical activity. In addition, worsening AF symptom severity is associated with reduced daily exercise.

The integrity of endothelial function in major arteries (EFMA) is a powerful independent predictor of heart attack and stroke. Existing ultrasound-based non-invasive assessment methods are technically challenging and suitable only for laboratory settings. EFMA, like blood pressure (BP), is both acutely and chronically affected by factors such as lifestyle and medication. Consequently, laboratory-based measurements cannot fully gauge the effects of medical interventions on EFMA. EFMA and BP have, arguably, comparable (but complementary) value in the assessment of cardiovascular health. Widespread deployment of EFMA assessment is thus a desirable clinical goal. To this end, we propose a device based on modifying the measurement protocol of a standard electronic sphygmomanometer. The protocol involves inflating the cuff to sub-diastolic levels to enable recording of the pulse waveform before and after vasodilatory stimulus. The mechanical unloading of the arterial wall provided by the cuff amplifies the distension that occurs with each pulse, which is measured as a pressure variation in the cuff. We show that the height of the rising edge of each pulse is proportional to the change in lumen area between diastole and systole. This allows the effect of vasodilatory stimuli on the artery to be measured with high sensitivity. We compare the proposed cuff flow-mediated dilation (cFMD) method to ultrasound flow-mediated dilation (uFMD). We find significant correlation (r = 0.55, p = 0.003, N = 27) between cFMD- and uFMD-based metrics obtained when the release of a 5 min cuff occlusion is employed to induce endothelial stimulus via reactive hyperemia. cFMD is approximately proportional to the square of uFMD, representing a typical increase in sensitivity to vasodilation of 300-600%. This study illustrates the potential for an individual to conveniently measure his/her EFMA by using a low-cost reprogrammed home sphygmomanometer.