Chest-Vibration Algorithm Shows Promise for Heart Failure Screening, but Specificity Trails Existing Tests

In a quiet room at a Danish hospital, a clinician placed a small accelerometer against a patient's chest, at the xiphoid process near the bottom of the breastbone, and recorded the vibrations of the beating heart. Only after that did the patient have a resting ECG, a blood draw for NT-proBNP, and a full echocardiogram. The order was deliberate. The person recording the vibrations did not yet know what the later tests would show.

That sequence is the backbone of a feasibility study published in PLOS Digital Health by Agam and colleagues. They set out to test seismocardiography, or SCG, as a way of detecting heart failure. The device, called CADScor and made by Acarix, uses an ADXL 354 accelerometer sampling at 8000 samples per second. It picks up the mechanical activity of the heart: muscle contraction, valve movement, and turbulence in blood flow. Earlier work had linked SCG signals to changes in preload and to early diastolic velocity on tissue-Doppler echocardiography. It had also separated compensated from decompensated heart failure patients.

The researchers enrolled patients referred by their general practitioners for suspected heart failure to Aalborg University Hospital or Odense University Hospital. In all, 218 patients were included: 198 in the main study and 20 newly diagnosed patients with heart failure with reduced ejection fraction (HFrEF), recruited at Aalborg afterward. The main study had only nine HFrEF patients, so the extra group was added to test the finished algorithms on that subtype. Those 20 were never used to train the models. Exclusions included a known prior heart failure diagnosis, atrial fibrillation, acute coronary syndrome as the main cause of the episode, pacemakers and defibrillators, implanted donor hearts, ventricular assist devices, significant scarring in the recording area, and pregnancy.

From the recordings the team built two scores. The AnyHF score aims to flag any type of heart failure, and the HFrEF score targets the reduced ejection fraction type. The AnyHF score reached an area under the curve of 82%, with sensitivity of 90.9% and negative predictive value of 87.5%. Its specificity was only 43.8%, and its positive predictive value 52.6%. In practice it caught most people who had heart failure but also flagged many who did not.

The second comparison was more balanced: the HFrEF score against NT-proBNP, the blood marker that guidelines already recommend, for separating HFrEF from no heart failure. NT-proBNP had an AUC of 94.7%, sensitivity of 94.1%, specificity of 68.8%, negative predictive value of 99%, and positive predictive value of 27.1%. The HFrEF score had an AUC of 92.9% and sensitivity of 88.2%. Its specificity was 92% and its positive predictive value 57.7%, both reported as significantly higher than the blood test's. On AUC and sensitivity, NT-proBNP came out ahead.

The scores also sorted patients who were referred on suspicion of heart failure but turned out not to have it. They classed 71 of them, 51% of that no-heart-failure group, as minimal risk. The authors argue that this kind of rule-out could ease pressure on echocardiography, which has long waiting lists and operator-dependent interpretation. NT-proBNP has its own limits, including delays in central lab analysis and the influence of non-cardiac factors on levels. Access to both tests is limited in many low- and middle-income countries, and ageing populations strain health systems in wealthier ones.

The study carries disclosures. Acarix funded it and owns the data, which Danish data protection law and re-identification risk also keep from public release. The company was allowed to review the manuscript, though the authors say final content and the decision to submit were theirs. Two authors, P.S. and S.E.S., previously transferred patent rights for the technology to Acarix, and S.E.S. is a part-time consultant for the company. The authors state that none received salary or personal payment from Acarix for this work.

The results are preliminary. A larger, independent validation, with more HFrEF patients than the 29 in this work, would be needed before the authors' vision of simple wearable or bedside screening in primary care or the community becomes realistic.