Skip to main content
Premium Trial:

Request an Annual Quote

Algorithm-Enabled 'Opportunistic' Heart Disease Screening Boosts Statin Therapy Prescriptions

Premium

NEW YORK – An algorithm developed by Stanford University researchers to detect coronary artery calcium from unrelated chest CT scans is being rolled out at various healthcare centers, with researchers having shown it can nudge patients and their physicians to consider preventive statin therapy.

Using artificial intelligence for such opportunistic screenings is an active research area right now, according to Nishith Khandwala, cofounder and CEO of Bunkerhill Health, which focuses on translating academic AI research into the clinic. Khandwala noted that data collected on a patient being evaluated for one reason could also highlight risk of another condition. In this case, the presence of coronary artery calcium (CAC) on a chest CT is a strong predictor of future atherosclerotic cardiovascular disease (ASCVD) events.

"Our capability for developing algorithms using granular data easily and at scale is just so vastly expanded," said Alexander Sandhu, a cardiologist at Stanford Medicine and codeveloper of the CAC algorithm. "But that puts even more burden on the next step, which is making sure that these algorithms are usable [and] understanding how and if they can be used to improve the quality of care, which is the real goal."

According to Sandhu, fewer than a million scans for CAC are conducted each year in the US, in part due to a lack of insurance coverage but also due to a lack of motivation by asymptomatic people to get tested. By contrast, he said that there are about 19 million non-cardiac chest scans taken each year, such as for lung cancer screening.

CAC also shows up on those scans. Sandhu and his colleagues developed an algorithm to spot CAC on non-gated chest CTs saved within patients' electronic health records. This way, they could identify patients at risk of disease and notify them and their physicians about the presence of the biomarker and, perhaps, push them toward preventive treatment.

"We know that statin therapies, as one example, lower the risk of heart attack or stroke by about 25 percent, but are used by less than half of the high-risk patients around the country," Sandhu said. "And so there is an opportunity to … motivate preventive therapies amongst those at high risk."

Sandhu and his colleagues in computer science and radiology used chest CTs from Stanford Hospital and the Multi-Ethnic Study of Atherosclerosis (MESA) study to develop a deep learning model to accurately and quickly quantify patients' CAC scores. The tool, as they reported in NPJ Digital Medicine, had a high sensitivity and positive predictive value when applied to outside validation datasets.

"The algorithm worked quite well," Sandhu said, but what "really excited" him was considering whether it could be used to modify clinical care and behavior.

That's the question he and his colleagues sought to answer in the NOTIFY-1 trial. In that study, they used their algorithm to screen 2,113 Stanford Health Care System patients' non-gated chest CTs from within their EHR for evidence of CAC. While about 80 percent of the people screened had no signs of calcium buildup, 20 percent, or 424 people, did. The researchers included 173 of those patients in their study cohort after excluding people who were already on statin therapy, had a cancer diagnosis, or for whom a radiologist could not confirm the presence of CAC.

Sandhu noted that they wanted a radiologist to confirm CAC, especially as their analysis was focusing on quality improvement and they did not want to cause undue anxiety among patients.

The patients were then randomized so that half were notified of these findings and half received usual care. Patients' physicians were first contacted, then the patients themselves. The notification included a picture of the calcium, which Sandhu noted is "like a bright white circle in the middle of your heart. It's pretty obvious."

Six months later, patients who were notified of their CAC status were much more likely to have received a statin prescription than patients in the usual care arm, as the researchers reported last year in Circulation.

In particular, 51.2 percent of the notification arm was on statin therapy after six months, while 6.9 percent of the usual care arm was. That boost was more than Sandhu expected. By speaking with other experts and based on other notification studies, he predicted closer to a 10 percent rise in statin prescription rates.

"I was incredibly, pleasantly surprised," Sandhu said.

He noted that in addition to that uptick in statin prescription rates, there was also an increase in the discussion of possible preventive therapies. Patients who received a notification were also more likely to have discussed statin therapy with their healthcare provider — 77.9 percent versus 12.0 percent for the usual care arm.

"We really want this to lead to an informed discussion between the clinician and the patient," he said.

This algorithm to spot CAC has now been further validated and is being commercialized by Bunkerhill Health, which specializes in connecting researchers at disparate institutions to one another's datasets by smoothing the way with the needed legal and other agreements for data access. It has established a consortium of 25 research institutions, including Stanford, Harvard University, Johns Hopkins University, and the University of California, San Francisco.

Bunkerhill's Khandwala noted that the Stanford developers trained the algorithm only using data from Stanford, as that data was easily accessible to them, but then it had to be tested using inputs from outside Stanford to ensure, for instance, it could be applied to scans taken on equipment made by other manufacturers and in different patient populations. He said the Stanford CAC algorithm has consistently performed well.

Such validation, he noted, is also needed for regulatory clearance, which his firm also facilitates. This algorithm has since garnered clearance from the US Food and Drug Administration, and Bunkerhill has provided it to more than 200 hospitals as part of its larger platform that also includes algorithms developed by other members of its consortium.

"It's a very interesting algorithm from the viewpoint that it leads to more patients getting access to care," Khandwala said.

He noted that they found that about 5 percent to 10 percent of patients that Bunkerhill has run the algorithm on have high calcium scores and no record of being on preventive treatments. As some health systems perform about 15,000 chest CT scans a year, that represents a sizable number of patients, Khandwala said. Though the initial NOTIFY-1 study excluded patients with known atherosclerotic cardiovascular disease, a follow-up study that the Stanford researchers are now undertaking, dubbed NOTIFY-ASCVD, is focusing on that group.

"In some ways, it's counterintuitive. We're using the algorithm to identify individuals at risk to help them understand their risk. These are all patients that already know they're at risk," Sandhu said.

However, nearly half of ASCVD patients are not on statin therapy, even though it is a guideline-recommended treatment, and those who are taking statins could potentially benefit from a change in dosage and from lifestyle or other interventions. Sandhu added that there is often a flurry of preventive measures taken right after a patient has a heart attack or stroke, but that the intensity often fades with time. Further, some patients with peripheral artery disease or cerebrovascular disease may not know their risk extends to their heart.

In the NOTIFY-ASCVD trial, the researchers are enrolling 1,000 individuals with coronary artery disease, peripheral artery disease, or cerebrovascular disease whose prior CT chest scans show signs of CAC. Those with cardiac calcium buildup, as determined by the algorithm, will then be randomized into three groups: notification with picture of that coronary artery calcium, notification without the picture, and usual care.

Here, too, the researchers will be examining how notification, and which type, affects statin prescription rates after six months.

"The question that we're trying to answer is: Is the image, even in these individuals who know that they have disease of their vasculature, motivating for intensification of therapy?" Sandhu said. "And I think it's an interesting question."