Peripheral artery disease (PAD) is one of the leading causes of limb loss in the United States. That makes September—PAD Awareness Month—an important one for the limb-loss community. PAD awareness is a primary focus of the Amputation Prevention Alliance and other, similar campaigns to reduce rates of limb loss. Neither patients nor clinicians pay sufficient attention to PAD risk factors, so the condition often goes undetected for months or years—and by the time it’s caught, limbs are in danger.
“A concerted effort is needed to institute policies to improve limb outcomes among patients with PAD,” the American Heart Association wrote in 2021, while kicking off its campaign to decrease nontraumatic lower-extremity amputation rates by 20 percent during the 2020s. The AHA’s wish-list of reforms includes wider access to PAD screening (especially within high-risk populations) and better tools for diagnosing and managing PAD.
Progress toward both those objectives could be enhanced through the use of artificial intelligence. “Machine learning models can accurately detect risk of PAD, and physicians are receptive to automated risk detection for PAD,” wrote a team of Stanford researchers in a widely cited 2022 paper. AI tools allow doctors to accurately assess patients’ PAD risk within seconds, which is huge given the limited duration (15 to 30 minutes) of most office visits. A reliable AI predictor would let them screen more patients, identify vulnerable patients sooner rather than later, and initiate treatment long before symptoms progress to the limb-threatening stage.
In the roughly 12 months since the Stanford study was published, a wave of additional research from around the world has reinforced and expanded on those findings. We’ve plucked some key conclusions from that cluster of papers, in the hope of understanding how soon AI PAD detection might become common — and what effect it might have.
Keeping It Simple: EHR Records
A paper published this month in the Journal of the American College of Cardiology took a similar approach to the Stanford study, using AI to extract data from electronic health records (EHR) and identify patients who require additional PAD screening. Because it uses existing data that requires no additional testing or intervention, this form of AI can be applied at very little cost and scaled up very quickly. The model relies on 60 common health indicators that affect the incidence of PAD, including factors such as weight, blood pressure, smoking status, kidney function, cholesterol, ankle-brachial index, and patient demographics. It was tested on more than 6,000 patients at Northwestern University and Vanderbilt University medical centers, and was significantly more accurate than existing PAD screening models.
The authors’ stated goal was “to develop a pragmatic model . . . to facilitate highly targeted screening or initiation of preventive strategies in an undiagnosed population.” While this model is a major step toward that objective, it still needs some fine-tuning. The full study is available at Science Direct.
Reading Doppler Waves
In a July preprint at MedRxiv, doctors from the Mayo Clinic shared evidence that AI can read Doppler arterial waveforms of PAD patients to identify individuals who are at elevated risk of limb loss. Doppler waveforms, which are acquired via a noninvasive ultrasound procedure, measure the force, rate, and volume of blood flow. Previous evidence showed that AI can read Doppler waveforms to identify the presence of PAD. The new study takes that approach a step further, enabling doctors to identify which PAD patients are at greatest risk of limb loss and get a head start on treatments, behavior modifications, and other measures that might affect outcomes and save limbs.
In a retrospective analysis of roughly 10,000 patients, the researchers found, “an AI algorithm assessment of posterior tibial arterial Doppler signal can accurately identify PAD patients at greatest risk for major adverse cardiac and limb events.” The statistical analysis is dense, so we won’t delve into it here; you can take a crack at it yourself at MedRxiv. The article hasn’t been peer-reviewed yet, so it’s not a finished product. But the paper illustrates another promising use of AI in preventing limb loss.
Shortly after ChatGPT was released in November 2022, cardiovascular specialists from Stanford and the Cleveland Clinic attempted to use the chatbot to construct an accurate, patient-friendly knowledge base about PAD. Their goal was to test “the potential of interactive AI to assist clinical workflows by augmenting patient education and patient-clinician communication around common CVD prevention queries.” To put that in a less long-winded phrasing: Can ChatGPT give doctors an additional tool to educate patients about PAD? Or is it likely to mislead them with inaccurate / incomplete / misleading information?
In their initial attempt, they had ChatGPT generate answers to 25 common questions about PAD prevention and treatment. Twenty-one of the answers (84 percent) passed muster, which we suppose is pretty good for an exploratory first try. With refinement for accuracy and readability, AI-generated PAD materials could be made available in various contexts, from general information resources on hospital websites to individual exchanges between patient and clinician via email or clinical messaging portals.
We’re extremely wary of AI-generated text. But given the widespread lack of awareness about PAD, it’s fair to ask whether a carefully curated AI-written knowledge base might be better than nothing. Read more at JAMA Network.
Addressing Health Care Disparities
Researchers at Baylor are experimenting with AI to reduce amputations at the extreme end of the PAD spectrum, critical limb-threatening ischemia (CLTI). PAD cases are most likely to porogress to the CLTI stage among low-income Americans who have poor access to routine healthcare. Women are also more prone to CLTI than men.
Social and environmental factors such as these make CLTI outcomes difficult to predict, in part because they’re affected by dozens of disparate factors, from patient-level health indicators to broad social and environmental factors. In other words, there’s a lot of noise built into every case, which can make it difficult for doctors to isolate the signals that actually affect limb-care outcomes. The Baylor team’s AI model provided impactful clinical guidance in that regard, helping caregivers assign proper weight to various risk factors and arrive at a more precise, individualized assessment of the threat facing each patient. Details available at Seminars in Vascular Surgery.