Despite AI Stethoscope Rollout, Heart Failure Diagnosis Rates Remain Unchanged

A study published in the international journal The Lancet found that introducing AI stethoscopes in primary care settings (clinics and health centers that serve as patients' first point of contact) did not significantly increase the rate of new heart failure diagnoses (a condition in which the heart fails to pump blood effectively). However, patients who actually underwent AI stethoscope testing showed higher rates of cardiac disease diagnoses.

The device used was the Eko DUO from U.S.-based Eko Health. It records a single-lead electrocardiogram and heart sounds over 15 seconds, then uses an AI algorithm to analyze the likelihood of reduced left ventricular ejection fraction of 40% or below (heart failure), atrial fibrillation, and valvular heart disease.

Over 12 months, the AI stethoscope group recorded 1,342 new heart failure diagnoses, or 2.58 per 1,000 patients per year, with no statistically significant difference from the control group's 2.76 (1,984 cases). The adjusted incidence rate ratio (IRR) was 0.94 (95% confidence interval, 0.86–1.02). No significant differences were found for atrial fibrillation (adjusted IRR, 0.98) or valvular heart disease (adjusted IRR, 1.00) either.

The researchers analyzed where heart failure was first diagnosed, dividing locations into community and hospital settings. The adjusted IRR was 0.90 for community diagnoses and 0.92 for hospital diagnoses. No effect of AI stethoscope adoption in shifting diagnosis from hospitals to the community was observed.

A different picture emerged in the analysis of actual users. Comparing 8,942 patients who underwent at least one AI stethoscope test using propensity score matching, the rate of new heart failure diagnosis was 2.33 times higher than in the control group, with atrial fibrillation 3.45 times higher and valvular heart disease 1.92 times higher. All three conditions were statistically significant. The researchers explained that this may reflect an actual increase in diagnoses rather than merely an earlier timing of diagnosis. They added that since test selection was left to clinicians' judgment, selection bias and other factors could not be ruled out.

During the study period, 972 clinicians performed a total of 12,725 tests, but at the 12-month mark, 38 of 96 sites (40%) had not used the device at all. Another 38 sites (40%) used it only 1–9 times per month, while just 6 sites (6%) used it 31 or more times per month. The researchers noted that the study demonstrated that AI stethoscopes can uncover additional cardiovascular disease when actually put to use.