Reproducibility of Use by Physicians and Patients of a Smart Stethoscope for Artificial-Intelligence Enhanced Diagnosis of Heart Failure
Abstract
Introduction: Early detection of heart failure (HF) is a global health system priority. An artificial intelligence-enhanced stethoscope (‘Eko DUO’), is capable of recording and analysing a single-lead ECG to detect HF with reduced left ventricular ejection fraction (LVEF≤40%) after only 15 seconds of apposition to the chest. Given this simplicity, patient-administered remote HF diagnosis and screening is feasible if performance and reproducibility of use of the Eko DUO is adequate.
Hypothesis: We tested hypotheses that:
- Eko DUO predictions for LVEF≤40% have good intra- and inter- operator reproducibility (intra-class correlation coefficient > 0.75) for both patient and clinician deployment.
- Eko DUO performance for detection of LVEF≤40% is comparable between patients and clinicians.
Methods: 98 hospital inpatients undergoing echocardiography at Imperial College Healthcare NHS Trust, UK, performed self-examination with the Eko DUO, using instructions to take three recordings from the left parasternal 2nd intercostal position. A clinician then repeated three recordings. The intra-class correlation coefficient was calculated using the raw numerical artificial intelligence predictions for LVEF≤40%. Performance for detection of LVEF≤40% was measured by calculating the area under the receiver-operating characteristic curves (AUC) using a reference standard of echocardiography-derived percentage LVEF.
Results: There was good intra-operator reproducibility for both patients (ICC = 0.817) and for clinicians (ICC = 0.888), and good inter-operator reproducibility between patient and clinicians (ICC = 0.887). The performance of the Eko DUO was comparable between patients and clinicians (AUC = 0.887 vs. AUC = 0.800, respectively).
Conclusions: The Eko DUO predictions for LVEF≤40% have good intra- and inter-operator reproducibility between patients and clinicians.
