乌鸦传媒視頻

CHICAGO – Core lab interpretation of coronary CT angiography (CTA) classified significantly fewer patients as having significant coronary artery disease (CAD) compared with on-site interpretation, but without a loss of predictive power for cardiovascular events, researchers reported here.

Readers at the core lab judged that 13% of the images represented a significant coronary artery stenosis, defined as ≥50% occlusion. On-site CTA deemed 23% of the lesions as having ≥50% occlusions, according to Michael Lu, MD, MPH, of Massachusetts General Hospital and Harvard Medical School in Boston, and colleagues.

Core lab and site interpretations were discordant in 16% of cases, and discordance was most commonly because of significant CAD by site but not by core lab read (80%), the authors reported at the Radiological Society of North America (RSNA) annual meeting. The study was simultaneously published in.

When catheter-quantitative coronary artery stenosis ≥50% was used as a reference standard, core laboratory readers had an accuracy of 0.69 area under the curve (AUC) compared with an accuracy of 0.57 AUC for on-site readers (P<0.001), Lu noted.

The C statistic for future MI or cardiovascular death was 0.61 (95% CI 0.54, 0.68) for the core laboratory and 0.63 (95% CI 0.56, 0.70) for the sites, the authors reported.

"A more parsimonious reading of coronary CT angiography at the institutional sites would have resulted in reduction of patients being sent for catheterization," Lu said. "However, there are conscious or unconscious biases that may push the reader to call more disease when trying to determine if a lesion is 45% or greater than 50%."

"These findings were robust across clinical reader expertise, the threshold for significant coronary artery disease and the extent of coronary artery calcium," Lu added. "The core lab was more accurate with quantitative coronary angiography as the reference standard. Despite calling 41% less disease, the core lab had similar prognostic value for cardiovascular death and myocardial infarction."

Over 25 months follow-up, 1.3% of the patients experienced a coronary event. There was a correlation of 0.67 with significant findings and an event in the core lab and a correlation of 0.68 for significant findings and an event in the local site interpretation of the scans (P=0.71).

In the trial, researchers at 193 North American sites interpreted coronary CTA results as part of the clinical evaluation of stable chest pain. Coronary CTA was also interpreted retrospectively by a central core lab blinded to clinical data, site interpretation, and outcomes.

The sites evaluated 4,347 patients, 51.5% of whom were women. The mean patient age was 60.4.

RSNA discussant Jonathan Leipsic, MD, of the University of British Columbia, Vancouver, said that the environment under which the decisions were made as to whether to call a lesion significant or not likely played a role in the results.

Calling on his own experience as a core lab researcher, he noted, "The drivers of the discordance may be that in the core lab, we have a goal of greater accuracy and the core lab readers are not biased by the symptoms of the patients. The core lab reader is more risk tolerant because we are not dealing with a patient but with a trial; we have no time constraints. We have 48 hours to give an interpretation, but that is very different than someone calling out from the ward asking about the patient they are clinically worried about. We have an optimal reading environment with no interruptions."

But, "The physician reader is more focused on the patients, is biased by the patient symptoms, is risk averse especially with medical-legal concerns, and has constant interruptions," he noted.

He said the PROMISE results raises questions, such as, "Can we really expect core labs to make all our decisions, or are we going to accept overdiagnoses?"

Leipsic suggested further testing of blood perfusion could help define which patients need to go to the cath lab. He added that there is also the promise of artificial intelligence or deep learning algorithms that may eventually play a role in decision-making.