Screening with a “smart,” artificial-intelligence (AI)-enhanced investigational digital stethoscope (Eko Duo) that provides phonocardiogram and electrocardiogram (ECG) readings doubled the detection of peripartum cardiomyopathy in a large study of obstetric patients in Nigeria.
Demilade A. Adedinsewo, MD, MPH, from Mayo Clinic, Florida, reported these findings from the Screening for Pregnancy Related Heart Failure in Nigeria (SPEC-AI Nigeria) trial in a press briefing and in a late-breaking trial session at the American Heart Association (AHA) Scientific Sessions 2023.
“The key takeaway,” Adedinsewo told theheart.org | Medscape Cardiology, “is recognizing that a simple, low-impact tool like a digital stethoscope can dramatically improve the diagnosis of a life-threatening condition, and we can treat it. A large proportion of the women will recover; if we identify them early and treat them appropriately, we can reduce the risk of dying.”
If the device predicted low ejection fraction, the patient went on to have an echocardiogram to confirm cardiomyopathy, defined as a left ventricular ejection fraction (LVEF) <50%.
Peripartum cardiomyopathy was detected in 4% of the women who were screened with this tool, compared with 1.8% of those who received usual care, which included a traditional ECG.
“I believe that the control arm also has about 4% of cardiomyopathy cases, but because they didn’t have the same screening and echo, we’re missing them,” Adedinsewo said.
Diagnosis of peripartum cardiomyopathy is challenging, she noted, owing to overlap of common symptoms in pregnancy, such as lower-extremity swelling, fatigue, and shortness of breath with mild activity, which are also cardinal symptoms of heart failure.
“We were really impressed by the effectiveness of the tool, looking at how accurate it was when it comes to the sensitivity,” she added. She noted that the digital stethoscope correctly identified 92% of women with LVEF <50% and 100% of those with LVEF <40%.
This was the first large, clinical trial to evaluate an AI intervention in pregnancy. The investigators used a portable, battery-operated device that yielded AI results in real time.
Nigeria has the highest rate of pericardium cardiomyopathy of any country. However, one study showed a 16-fold higher rate of cardiomyopathy among African American women compared to White women in the United States, Adedinsewo noted. “It will be important to identify who we should be screening to identify more cases,” she said.
A digital stethoscope that provides an ECG is currently available, but the algorithm that powers detection of cardiomyopathy is not yet commercially available.
Findings “Absolutely Startling”
The study discussant in the press briefing, Alexander Tarlochan Singh Sandhu, MD, from Stanford University, California, congratulated the authors on this “valuable study that uses AI tools to solve a real health problem.”
Finding that 4% of the women in the intervention arm had reduced ejection fraction is “absolutely startling,” he said, “and speaks to how important improving our diagnosis in this space is.
“Where the burden of disease is high, a tool like this can be so incredibly valuable,” he said. He noted that the investigators identified 2% more patients with peripartum cardiomyopathy.
“This is an example of the potential of AI tools that can actually improve access to care and improve quality of care in resource-limited settings,” he said. “We need to move to understanding how to implement this into subsequent care [and] figure out what the next steps are to improve their outcomes.”
“The main takeaway is that in areas where there is a very high prevalence of a morbid condition, a prescreening tool like this may be helpful” for diagnosis, the assigned discussant in the session, Marco Perez, MD, also from Stanford University, told theheart.org | Medscape Cardiology.
The number of women needed to screen to detect peripartum cardiomyopathy by echocardiography alone is 1 in 23 in Nigeria and 1 in 970 in the United States, he said.
With an AI tool such as this one (sensitivity, 92%; specificity, 80%), the number needed to screen would be 1 in 5.7 in Nigeria and 1 in 194 in the United States, he estimates on the basis of incidence data.
“Because it is so common in Nigeria, a screening method makes a lot of sense,” Perez said. “The big question that remains is, what is the best screening modality?
“Certainly, this tool helped in bringing down the number of echoes needed to find a case, from the mid 20s down to about five or six, so it certainly does seem to be helpful.”
However, the investigators did not say whether this tool is better than a clinical review of ECG or an AI analysis of ECG alone. It’s not clear whether the phonocardiogram component is significant in conjunction with the ECG component.
Nevertheless, “In a place where there’s a very high prevalence of peripartum cardiomyopathy, like Haiti, like Nigeria, doing something like this makes a lot of sense.
“For the US and the rest of the world, where the prevalence is much lower, even with a tool like this you still would need to do a lot of echoes to find one case, and that may end up not being cost-effective. You would need to screen 200 women with echo to find one case.”
AI-Guided Screening Study
Nigeria has the highest reported incidence of peripartum cardiomyopathy mortality (1 in 100 live births) and the highest number of maternal deaths.
In the United States, where rates of peripartum cardiomyopathy are much lower, maternal deaths are nevertheless higher than in other developed countries and have trended up over the past three decades; cardiomyopathy is a key contributor.
The investigators enrolled 1195 women who were pregnant or had given birth in the past 12 months. The patients were from six teaching hospitals in Nigeria (two in the north and four in the south). They were randomly assigned in a 1:1 ratio to the intervention group (587) or the control group (608).
In the intervention group, clinicians used a smart stethoscope to record a phonocardiogram and a single-lead ECG reading in the V2 position and in an angled position on the patient’s chest wall and to record an ECG from the patient’s fingers. The recordings were sent to a Bluetooth-enabled mobile device (tablet or smartphone), which displayed the phonocardiogram and ECG images and that indicated whether the ejection fraction was normal or low. All patients in the intervention group received an echocardiogram.
In the control group, patients received usual care plus a traditional ECG. They were not required to have an echocardiogram because undergoing an echocardiogram is not part of usual care; however, they could receive an echocardiogram if the ECG suggested that they might need further testing.
The mean age of all the patients was 31 years, and all were Black. At study entry, 73% were pregnant, and 26% were post partum. They had similar comorbidities.
The primary outcome, cardiomyopathy (LVEF <50%) was detected in 24 of 587 patients (4.1%) in the intervention group and in 11 of 608 patients (1.8%) in the control group (odds ratio, 2.3; 95% CI, 1.1 – 4.8; P = .02).
For the detection of LVEF <50%, the sensitivity was 92% and the specificity was 80%. For the detection of LVEF <40% (a secondary outcome), the sensitivity was 100% and the specificity was 79%.
Adedinsewo is supported by the Mayo Building Interdisciplinary Research Careers in Women’s Health (BIRCWH) Program, which is funded by the National Institutes of Health. The trial was funded by Mayo Clinic (Centers for Digital Health and Community Health and Engagement Research) and in part by the Mayo Clinic BIRCWH Program. Portable ECG, phonocardiogram recordings, and AI predictions using the digital stethoscope were extracted by the Eko Health team and were sent to the coordinating center for analysis. Eko Health had no role in study design, data collection, data analysis, or data interpretation.
Source: Read Full Article