Study: AI-powered stethoscope fails two-thirds of the time
Embedded Files
Occurred: 2023-
Page published: February 2026
Report incidentđĽ| Improve page đ| Access database đ˘
An AI-enabled stethoscope used in UK GP practices flagged large numbers of patients as having heart problems, but follow-up tests showed that around twoâthirds of those labelled with suspected heart failure actually did not have the condition, creating significant rates of false alarms and anxiety for patients.Â
What happened
What happened
In a UK study involving more than 12,000 primary care patients, an AI-powered âsmartâ stethoscope developed by US company Eko Health was used to help doctors to detect conditions such as heart failure, atrial fibrillation, and heart valve disease.â
Patients examined with the Eko DUO device were more than twice as likely to be newly diagnosed with heart failure and over three times as likely to be diagnosed with atrial fibrillation compared with usual care.
However, subsequent specialist investigations revealed that about twoâthirds of patients the AI flagged with suspected heart failure did not actually have heart failure, meaning many people went through worrying additional tests and fear of a serious diagnosis unnecessarily.
The study also discovered that around 70 percent of participating GP practices stopped using the device or used it only rarely after a year, suggesting frontline clinicians found the false positives and workflow impact problematic.
Why it happened
Why it happened
The AI system was tuned to be highly sensitive so it would miss as few serious heart problems as possible, but this came at the cost of lower specificity, generating many false positives.
The tool was deployed in busy primary care with varied patient populations and recording conditions, which can differ from the more controlled settings used to develop and validate the algorithm.
Communication and governance around limitations appear weak: the device was marketed as a way to improve early diagnosis, but patients and some clinicians were not clearly equipped to understand that a positive AI flag was only an initial risk signal, not a confirmed diagnosis.
Limited transparency about how the model works and a lack of clear accountability for downstream consequences (extra tests, anxiety, clinic time) meant that the harms of unnecessary referrals were not fully anticipated or mitigated.
What it means
What it means
For individual patients, the incident shows how AI tools in healthcare can subject people to distress, extra testing, and potential overâtreatment even when the underlying condition is absent, undermining trust in both clinicians and digital health technologies.
For clinicians, it highlights the burden of managing AI-generated false alarms, including extra workload, more referrals, and difficult conversations to âunwindâ alarming provisional labels like suspected heart failure.
For society, this case illustrates that âmore detectionâ is not automatically better: highâsensitivity AI screening tools can shift healthcare systems toward overâdiagnosis and resource strain if they are not carefully evaluated for realâworld tradeâoffs, not just headline accuracy gains.
For policymakers and regulators, it underlines the need for stronger preâ and postâmarket evaluation of clinical AI, requirements for clear communication of falseâpositive risks, explicit responsibility for managing downstream harms, and ongoing surveillance of algorithm performance once deployed in routine care.
System đ¤
System đ¤
Eko DUO
Documents đ
Documents đ
National Institute for Health and Care Research. âSmart stethoscopeâ AI heart disease detection trialled in UK GP surgeries
Developer: Eko Health
Country: UK
Sector: Health
Purpose: Detect heart conditions
Technology: Deep learning; Machine learning
Issue: Accountability; Accuracy/reliability; Safety; Transparency
AIAAIC Repository ID: AIAAIC2190
Page updated
Google Sites
Report abuse