A novel quality improvement project that notified physicians and patients about incidental coronary artery calcium (CAC) on prior nongated chest CT scans led to significantly more statin prescriptions than usual care.
Six months after randomization, statins were prescribed to 51.2% of patients who received messages that contained a personalized image of their chest CT with coronary calcium circled in red compared with 6.9% of patients given usual care (P < .001).
Among patients receiving a statin in the notification group, 72.7% received a moderate-intensity statin and 18.2% a high-intensity statin.
“We found opportunistic CAC screening with non-gated chest CT with subsequent clinician and patient notification increases statin initiation rates. This is really important because it leverages existing data without additional radiation at minimal cost,” said study author Alexander Sandhu, MD, MS, Stanford University, California.
He noted that ECG-gated CT scanning, the reference standard for CAC testing, is predictive of cardiovascular outcomes but rarely performed, whereas an estimated 19 million nongated chest CTs are done annually in the United States and also predict cardiovascular events. “This creates an incredible opportunity to extract data on coronary calcium from millions of scans and improve our risk prediction.”
The study was published online in Circulation and presented at the American Heart Association (AHA) Scientific Sessions 2022.
“I’m a big fan of this study and the space overall — incidental risk information that’s actually really powerful — and this is kind of the first real implementation work in this space,” Parag Joshi, MD, a preventive cardiologist at the University of Texas Southwestern Medical Center in Dallas, who was not part of the research, said in an interview. “The effect seen was really large in this somewhat small study, but it’s a really large effect size that’s plausible and believable in my opinion.”
How the Project Unfolded
The NOTIFY-1 project was motivated by efforts at Stanford to increase primary prevention statin therapy among high-risk patients, and by their recent development of a deep-learning algorithm that automated CAC scoring of routine nongated chest CTs, with a sensitivity for CAC >0 of 82% to 94% and positive predictive value of 87% to 100%, Sandhu explained.
The investigators identified 2113 patients under age 85 with a Stanford primary care visit and nongated, noncontrast chest CT between 2014 and 2019 who did not have a history of atherosclerotic cardiovascular disease (ASCVD), prior statin therapy, or coronary angiography. The algorithm identified CAC on chest CTs of 424 patients, of whom 230 were excluded after manual chart review.
Of the remaining 194 patients, CAC was confirmed by a radiologist in 89% and those 173 patients were randomized. Their median age was 70.8 years, slightly more than half were women, and 94% had a 10-year ASCVD risk of 7.5% or higher.
In the notification arm, primary care clinicians were sent an electronic health record message that included an image of the patient’s chest CT with a circle around the CAC, a reference to the 2018 AHA/American College of Cardiology cholesterol guidelines, and told that their patient would be given the same information within 2 weeks unless they thought it would be inappropriate. “There were no objections to notifying the patients,” Sandhu said.
All 86 notification patients were sent the same image, cautioned about the increased risk for a heart attack with CAC, and told to speak with their primary care physician. Reminders were sent after 2 weeks if the notice wasn’t opened and after 3 months if there was no documented discussion.
The most common reason for the CT was to evaluate a pulmonary nodule (36%), followed by lung cancer screening (12%). Although most radiology reports (84.4%) noted CAC in the report, it was mentioned in the final impression in only three instances, he pointed out.
Downstream Effects
At 6 months, 78% of the notification group had a documented statin discussion or new statin prescription vs 12% in the usual care group (P < .001).
There was no significant difference in aspirin treatment, hemoglobin A1c, or blood pressure in follow-up. However, lipid measurement was significantly higher at 58% in the notification group vs 33% with usual care (P = .002).
There was an increase in healthcare utilization among the notification group, including more primary care visits per patient (2.2 vs 1.4; P = .011), more new cardiology visits (16.3% vs 4.6%; P = .015), and more coronary artery disease testing (15.1% vs 2.3%; P = .008). The latter was predominantly driven by the 12% of patients in the notification group who underwent stress testing, Sandhu said.
Limitations are that this is a small proof-of-concept study, a radiologist reviewed each scan to reduce the risk for false positives, and a small absolute error in the estimated CAC score might be less important at a higher threshold than zero, he said.
“Now we need further studies to evaluate how this impacts clinical outcomes,” Sandhu concluded. “This project provides an example of how we can and should evaluate [the impact of artificial intelligence algorithms] on clinical decision making, clinical processes, and outcomes, rather than solely classification accuracy.”
Joshi, who is also an associate editor for Circulation, noted that the recent DANCANVAS trial of comprehensive screening with noncontrast ECG-gated CT was hypothesis-generating because it missed its primary endpoint of reducing all-cause mortality in men aged 65 to 74, but showed a mortality benefit in those younger than 70.
“There are tons of data that a high CAC score is higher risk, so it makes a lot of sense that this [is] going to impact outcomes,” he said.
Nevertheless, NOTIFY-1 was a labor-intensive, quality-improvement intervention that included clinician education prior to randomization, provider and patient notifications, use of a novel algorithm, and a radiologist to read all CT scans, Joshi added.
“In terms of a first step, I think it’s a huge first step,” he said. “But there’s still obviously limitations here. How do you run this? How scalable is it in terms of what they did in this study?”
There are also the important considerations of anxiety caused by the CAC notifications and the potential for unnecessary downstream testing.
“I think many of us think and, guidelines would agree, that symptoms should drive stress testing,” Joshi said. “So if just by notifying someone that ‘hey you have this calcium’ and get a bump in stress testing, you wonder did all 12% of these stress tests have symptoms or was that just because of the calcium score itself? And we don’t have that kind of granularity from this trial.”
Circulation. Published November 7, 2022. Full text
The study was supported by the Stanford University Human-Centered Artificial Intelligence Seed Grant. Sandhu receives research support from the National Heart, Lung, and Blood Institute.
Follow Patrice Wendling on Twitter: @pwendl. For more from theheart.org | Medscape Cardiology, follow us on Twitter and Facebook.
Source: Read Full Article