Recently, researchers compared ejection fraction values derived from conventional two-dimensional (2D) transthoracic echocardiography to those generated by a wearable device that uses acoustic sensors, electrocardiographic signals, and machine learning to generate a value. How did the measurements from the wearable device MOST likely compare to those derived from 2D transthoracic echocardiography?

  • □ (A) Overestimated
  • □ (B) Underestimated
  • □ (C) Highly correlated

The left ventricular ejection fraction is an important measure of cardiac function that predicts mortality risk and influences treatment decisions. Currently, modalities for the measurement of left ventricular ejection fraction require the use of specialized equipment and high levels of provider and interpreter skill. Previous studies have found that heart sounds correlate with cardiac pressure gradients, blood flow, and myocardial tissue motion. Researchers used these correlates to develop a wearable device that uses acoustic biosensors, electrocardiogram (ECG) electrodes, and machine learning to generate a numeric value for the ejection fraction. A recent study compared the ejection fraction produced by this product (Cardiac Performance System [CPS]; Sensydia) against the ejection fraction calculated by a standard 2D transthoracic echocardiogram (TTE).

The device consists of two belt-like straps that wrap around the thorax, positioning acoustic sensors at the aortic, pulmonic, tricuspid, and mitral auscultation sites. Each sensor receives signals that characterize physiologic cardiac functions such as pressure gradients, flow, and tissue motion. ECG electrodes are applied at the conventional monitoring sites. These data are then input into a previously trained and calibrated neural network, and an ejection fraction value is generated. The device records the acoustic and ECG signals and outputs 1 ejection fraction value for a three-minute recording session.

A total of 121 patients were evaluated for study inclusion. However, 25 patients were excluded for low-quality acoustic or ECG data, and TTE imaging was unacceptable for another 15 patients. Eighty-one patients were ultimately enrolled in the study. Seventy-eight percent of the patients had known cardiac pathology at the time of the study (i.e., cardiomyopathy, valvular disease, arrhythmias). The mean patient age was 53 ± 16 years, and the mean body mass index was 27 ± 5 kg/m2.

The ejection fraction measurements produced by the CPS device were compared to the TTE ejection fraction measurements and were found to have a very high level of correlation as determined by a Deming regression slope of 0.9981 (95% CI, 0.8714-1.1263) and an intercept of 0.03415% (95% CI, –6.4253% to 6.4936%). A slope value near 1 and an intercept value near 0 indicate a high level of correlation between the CPS and TTE values. The CPS device was also evaluated for its ability to identify patients with an ejection fraction below 50% or 35% and for its ability to produce similar results during repeated assessments. The CPS device was found to have a very high ability to correctly identify patients in both of the lower ejection fraction groups. The repeatability assessment found low variability in the ejection fraction value produced by the device when measurements were performed by the same operator on the same patient. The operator was the individual responsible for appropriate device placement. A separate assessment was performed to determine the level of interoperator variability in ejection fraction measurements. Three different operators performed one ejection fraction measurement with the CPS device on eight patients. This assessment showed a high level of reproducibility between providers.

In summary, a wearable device that uses acoustic cardiac signals, ECG data, and machine learning to produce an ejection fraction value demonstrated a very high level of correlation with ejection fraction values obtained by TTE. The device also identified patients with an ejection fraction below 50% and 35% with a high level of accuracy. However, the device was not usable in approximately 20% of the initially enrolled patients due to poor quality acoustic or ECG signals.

CORRECT ANSWER: C