Cardiology

In EAGLE, new diagnoses of EF of 50% or less within 90 days rose from 1.6% to 2.1% (odds ratio 1.32). The SPEC-AI Nigeria authors describe the U.S. FDA-cleared algorithm, available from Anumana, as a modification of the original Mayo Clinic 12-lead model retrained to detect LVEF below 40%; with it, left ventricular systolic dysfunction was detected in 3.4% of the intervention arm versus 2.0% of controls (odds ratio 1.75, P=0.125).

A retrospective validation of 16,000 records (13,960 evaluable) at four U.S. sites, funded by Anumana, co-authored by four of its employees and used in its FDA submission, found an area under the receiver operating characteristic curve of 0.92, sensitivity of 84.5% and specificity of 83.6% for LVEF of 40% or less, with a negative predictive value of 98.4% at 7.9% prevalence. The FDA summary reports the same sensitivity, specificity and predictive values and says the algorithm was developed on records of 93,722 Mayo Clinic patients who had an ECG and an echocardiogram within two weeks; LVEF of 40% or less (11% of the cohort) was the target.

In SPEC-AI Nigeria, at six hospitals, the intervention arm was screened with an Eko DUO stethoscope; detection was 24 of 587 versus 12 of 608 (odds ratio 2.12, P=0.032), with no serious adverse events. In TRICORDER, the incidence rate ratio for heart failure detection was 0.94 (95% confidence interval 0.86 to 1.02). Press coverage reported that patients examined with the device were 2.33 times as likely to be diagnosed with heart failure, that two-thirds of suspected heart failure cases it flagged were ruled out, and that a significant number of practices stopped using it after a year.

In a prospective study of 1,050 patients in London, independent of Eko, the AI had an area under the curve of 0.85 at the best recording position (sensitivity 84.8%, specificity 69.5%). For murmurs, two accuracy studies by authors including current or former Eko employees found 76.3% sensitivity and 91.4% specificity in 962 patients and, at three primary care clinics, 92.3% sensitivity for moderate or greater valve disease versus 46.2% for physicians, with lower specificity (86.9% versus 95.6%).

The PRECISE composite of death, nonfatal myocardial infarction or catheterization without obstructive coronary artery disease was driven by fewer catheterizations without obstructive disease (2.6% versus 10.2% with usual testing); nonfatal myocardial infarction occurred in 1.2% with the precision strategy and 0.5% with usual testing, a difference that was not significant. In FORECAST, invasive angiography showing no obstructive disease was 52% lower in the group assigned to CT angiography with selective FFRct.

In an observational study of England's national program, covering 90,553 patients who had coronary CT angiography at 27 NHS hospitals, there was no difference in all-cause mortality (adjusted hazard ratio 1.00) and fewer invasive coronary angiograms (adjusted hazard ratio 0.93) after FFRct became available than before. In the nonrandomized PLATFORM study at 11 European sites, invasive angiography found no obstructive disease in 12% of patients for whom it was planned in the FFRct-guided arm, versus 73% with usual care. For Plaque Analysis, the REVEALPLAQUE study of 237 patients, co-authored by HeartFlow employees, found a correlation of 0.91 with intravascular ultrasound for total plaque volume.