Acarix has reported peer-reviewed exploratory evidence that the seismocardiography technology underlying its CADScor System may help detect heart failure, including a machine-learning algorithm that produced substantially fewer false-positive classifications than the widely used NT-proBNP blood test when identifying heart failure with reduced ejection fraction. The study enrolled 218 patients and was published in PLOS Digital Health, but the new heart-failure algorithms remain investigational and are not included in the current United States Food and Drug Administration clearance for CADScor.
Researchers developed two algorithms using chest-vibration signals recorded through seismocardiography. The AnyHF score was intended to identify heart failure broadly, while a separate HFrEF score focused on heart failure with reduced ejection fraction, a phenotype in which the heart’s pumping function is impaired.
The AnyHF algorithm achieved sensitivity of 90.9% and a negative predictive value of 87.5%. For HFrEF, the acoustic algorithm generated an area under the receiver operating characteristic curve of 92.9%, compared with 94.7% for NT-proBNP, while delivering specificity of 92% compared with 68.8% for the blood biomarker. Positive predictive value was 57.7% with the seismocardiography approach versus 27.1% with NT-proBNP.
Those results make the technology worthy of further study, particularly because unnecessary echocardiography referrals create cost and capacity burdens in cardiovascular care. They do not establish that CADScor is ready to diagnose heart failure in clinical practice. The analysis was exploratory, performed in a relatively small two-center cohort, and the researchers themselves concluded that larger external validation is necessary.
Why could fewer false-positive heart failure tests matter in everyday clinical practice?
Heart failure is not diagnosed from a single symptom or measurement. Patients may present with shortness of breath, fatigue, edema or exercise intolerance, symptoms that overlap with pulmonary disease, obesity, anemia, kidney disease and other conditions. Primary-care physicians therefore need tools that can help determine which patients are sufficiently likely to have heart failure to justify echocardiography and specialist assessment.
NT-proBNP is widely used because concentrations of the biomarker rise when cardiac chambers experience increased wall stress. A low value can be particularly useful for ruling out heart failure, but elevated concentrations are not specific to one disease. Age, renal dysfunction, atrial fibrillation and other conditions can increase natriuretic peptide levels, potentially generating positive tests in patients who do not ultimately meet diagnostic criteria for heart failure.
Acarix’s feasibility study suggests that acoustic cardiac signals might provide complementary information. The HFrEF algorithm’s specificity of 92% was significantly higher than the 68.8% reported for NT-proBNP in the studied population, meaning fewer people without HFrEF were classified as positive. Positive predictive value also increased from 27.1% to 57.7%.
If those results are reproduced prospectively, the operational impact could be meaningful. A more specific first-line test could reduce the number of patients referred for echocardiography despite ultimately having no heart failure, allowing imaging capacity to be directed toward individuals with higher disease probability.
The clinical objective would not be to eliminate echocardiography. Cardiac imaging remains essential for characterizing ventricular function, valve disease, chamber dimensions and other structural abnormalities once heart failure is suspected. The potential role of seismocardiography would instead be upstream triage.
How does seismocardiography identify signals associated with heart failure?
Seismocardiography measures very small mechanical vibrations generated by cardiac contraction and blood movement. A sensitive accelerometer placed on the chest can capture patterns linked to events during the cardiac cycle, producing information that is different from the electrical signals recorded by an electrocardiogram.
Acarix already uses related acoustic and mechanical information in CADScor, its FDA-cleared diagnostic aid for symptomatic patients with suspected coronary artery disease. The existing system records heart sounds and vibrations, processes those signals computationally and combines them with clinical risk factors to calculate a CAD score intended to help rule out significant coronary stenosis.
The heart-failure study applies machine learning to the signals in a different way. Changes in ventricular contractility, filling, timing and mechanical function can alter the vibration patterns reaching the chest wall. Researchers therefore trained algorithms to identify combinations of features associated with heart failure and with reduced ejection fraction.
That approach is attractive because data collection is rapid and non-invasive. CADScor produces its current coronary assessment in less than 10 minutes, and an acoustic heart-failure assessment could theoretically be incorporated into a similar point-of-care workflow.
The challenge is generalizability. Machine-learning algorithms can perform strongly in the dataset used to develop them yet lose accuracy when exposed to new hospitals, operators and patient populations. External prospective validation is therefore particularly important before the reported accuracy can be generalized to routine care.
What are the biggest weaknesses in the 218-patient heart failure study?
The study included 198 patients referred by general practitioners because heart failure was suspected, plus another 20 participants with previously established HFrEF who were enrolled separately to strengthen algorithm development. That design supplied researchers with enough affected patients to explore the signal, but it also means the dataset was not simply an unselected real-world primary-care population.
The study was conducted at two centers, limiting geographic and institutional diversity. Patient characteristics, referral thresholds and background disease prevalence may differ in other health systems, potentially changing both algorithm performance and predictive values.
Another limitation is that the algorithms were developed and evaluated within a feasibility research program rather than a completed registration study designed to support a new regulatory indication. The strong specificity comparison with NT-proBNP is therefore hypothesis-generating rather than evidence that the device should replace biomarker testing.
Predictive values also vary with prevalence. A positive predictive value of 57.7% in the study cannot simply be assumed in a lower-risk primary-care population where HFrEF is much less common. Sensitivity and specificity tend to be more transferable, but these metrics can also shift when case mix changes substantially.
Acarix consequently needs a larger independent cohort in which the algorithm is fixed before testing begins. Prospective validation should establish how often the system produces usable recordings, how performance varies across age, sex, body habitus and comorbidities, and whether adding the test to existing clinical assessment genuinely reduces unnecessary imaging without missing important disease.
Why is the HFrEF performance more commercially interesting than the AnyHF result?
Heart failure comprises several phenotypes. HFrEF involves clearly reduced left ventricular ejection fraction, while heart failure with preserved ejection fraction can produce symptoms and abnormal filling pressures despite apparently preserved systolic pumping function. Diagnosis of the latter can be particularly challenging.
The AnyHF algorithm achieved 90.9% sensitivity, showing that seismocardiography carries information relevant to heart failure more broadly. Its negative predictive value of 87.5%, however, may not yet be strong enough for clinicians to comfortably use the test alone as a general heart-failure exclusion tool in many settings.
The HFrEF result is more striking because diagnostic discrimination was close to that of NT-proBNP while specificity was markedly higher. HFrEF also represents a phenotype with well-established treatment pathways, creating a clear value in identifying affected patients efficiently.
That does not mean Acarix should necessarily develop an HFrEF-only commercial product. A point-of-care tool becomes more valuable if it can address the broader population presenting with suspected heart failure, particularly as preserved-ejection-fraction disease represents a substantial proportion of cases.
The eventual regulatory strategy may therefore involve balancing algorithm performance against the breadth of the proposed indication. A narrow claim could potentially be easier to validate, while a broad heart-failure triage indication could create a larger commercial opportunity if supported by sufficiently strong data.
Could one CADScor platform eventually screen for both coronary disease and heart failure?
This is perhaps the most important strategic implication of the study for Acarix. The existing CADScor System was developed as a first-line aid for assessing patients with suspected stable coronary artery disease. The company says the system has been used in more than 60,000 patient assessments and is supported by a larger coronary-disease clinical program.
Chest pain and shortness of breath often create overlapping diagnostic pathways. A platform able to extract several cardiovascular risk signals from one rapid chest recording could become more useful to primary-care practices, outpatient cardiology and other settings where clinicians need to determine which patients require more expensive imaging or specialist testing.
Acarix has said the investigational heart-failure algorithm has been integrated technically into the mechanics of its second-generation device. That does not confer regulatory authorization, but it suggests the company is developing the hardware with a broader diagnostic portfolio in mind.
A multi-indication platform could also improve the economics of installation. Healthcare providers may find it easier to justify purchasing a system if the same equipment can support several common cardiovascular triage decisions, while disposable patches can create recurring revenue as utilization expands.
The business case still depends on regulatory evidence and reimbursement. A technically capable device without sufficient payer coverage or physician confidence can struggle to reach meaningful utilization, particularly when competing against inexpensive blood tests and established imaging pathways.
What does Acarix’s recent share performance say about investor expectations?
Acarix is listed on the Nasdaq First North Premier Growth Market in Stockholm under the ACARIX ticker. Shares closed around SEK 0.2155 on September 11, down about 3.8% for the session, despite publication of the heart-failure research. The stock had traded at approximately SEK 0.235 at the beginning of the month, leaving it lower across the early-September period.
That subdued reaction is consistent with the developmental status of the program. The paper opens a potentially important new indication but does not immediately expand the device’s authorized market. Investors still have to factor in the cost and duration of additional validation, regulatory submission and commercial development.
The distinction is commercially important. If Acarix had already secured an expanded indication, investors could begin estimating revenue contributions from heart-failure testing. At present, the data are better understood as evidence supporting investment in a new development program.
Successful external validation could materially change that position because the potential heart-failure population is large and diagnostic demand is persistent. Failure to reproduce the reported specificity advantage, however, would reduce the strategic value of the exploratory findings.
What must happen before CADScor technology can be used clinically for heart failure?
Acarix first needs to lock the investigational algorithm and demonstrate performance in a larger independent population. Ideally, the validation study would include multiple geographies, diverse patient demographics and a broad range of comorbidities that commonly complicate interpretation of suspected heart failure.
The reference standard also needs to be rigorous. Diagnosis should incorporate echocardiography, specialist adjudication and appropriate clinical information rather than relying on one comparator blood test. Performance should be evaluated separately for HFrEF, preserved-ejection-fraction disease and other relevant phenotypes.
The company would then need to submit the new indication to the appropriate regulatory authorities. The existing FDA De Novo clearance for suspected coronary artery disease does not permit Acarix to market the device as an approved or cleared heart-failure diagnostic based on the new paper.
Clinical utility could become the final commercial hurdle. Even strong diagnostic accuracy may not be enough unless prospective studies show that use of the system changes referral patterns, shortens time to diagnosis or reduces unnecessary testing without compromising detection of disease.
The PLOS Digital Health study gives Acarix a credible signal to pursue. A 92% specificity result compared with 68.8% for NT-proBNP is difficult to ignore, particularly when false-positive testing contributes to pressure on echocardiography services. The next study, however, will determine whether the difference reflects a genuinely scalable diagnostic advantage or the favorable conditions of an early feasibility dataset.
