AI-Based System Boosts Quality of Stroke Care and Reduces Events

PHOENIX, AZ—Integration of an artificial intelligence (AI)-based clinical decision support system into hospital workflows not only enhances the quality of care delivered to patients with acute ischemic stroke, but also improves patient outcomes, the GOLDEN BRIDGE II trial shows.

The rate of new vascular events through 90 days was 2.9% among patients treated at hospitals using the AI-based system and 3.9% among those treated at hospitals delivering usual care, a relative 26% difference, Zixiao Li, MD, PhD (Beijing Tiantan Hospital, Capital Medical University, China), reported last week at the International Stroke Conference.

The cluster-randomized trial, conducted at 77 secondary or tertiary hospitals across China, demonstrates both the feasibility and efficacy of implementing AI-based support in this setting, Li said, telling TCTMD that these are “exciting findings for us.”

During his presentation, he said that “further research is needed to fully evaluate the [generalizability] of these findings in different settings.”

But Lee Schwamm, MD (Yale School of Medicine, New Haven, CT), a member of the trial steering committee, said these findings likely are applicable to healthcare systems in other parts of the world, including the United States.

“I think we provide both excellent and really inadequate care, sometimes at the patient level even within a great hospital. Patients get out of some of our hospitals without getting a lot of these evidence-based treatments,” he commented to TCTMD. By using an AI-based tool that can be embedded in electronic medical record (EMR) systems, and that shows up at the point of care, he added, “you’re likely going to be a lot more successful.”

GOLDEN BRIDGE II

Stroke incidence and prevalence continue to rise around the world, particularly in developing countries, which creates challenges for care and prevention, Li said. Previously, he and his colleagues conducted the GOLDEN BRIDGE-AIS cluster-randomized trial, showing that a quality-improvement initiative resulted in better adherence to evidence-based performance measures and reduced new vascular events through a year of follow-up in patients with acute ischemic stroke.

I think there is the potential role for AI tools such as this to sort of raise the [bar] for all hospitals. Seemant Chaturvedi

In China, however, there remains a disparity between guideline recommendations and what’s happening in daily practice due to a general shortage of quality resources, a high demand for stroke physicians, and heavy workloads, Li said. AI, which has seen increasing use within healthcare, might help close the gap.

To that end, Li’s team developed their AI-based clinical decision support system, which integrates inputs from various hospital information systems, including the EMR. It has three main features: AI-assisted imaging analysis to provide details on the infarct, automated classification of stroke etiology, and guideline-based treatment recommendations (including secondary prevention).

The investigators then designed the GOLDEN BRIDGE II trial to test the impact of the system, randomizing centers to treat patients with the assistance of the AI-based tool or to deliver usual care. The study ultimately included 21,603 patients (median age 67; 64% men) with acute ischemic stroke; the median NIHSS score at admission was 3.

The primary outcome was new vascular events (a composite of ischemic stroke, hemorrhagic stroke, MI, or vascular death) within 90 days, and those were reduced at centers using the AI tool (adjusted HR 0.74; 95% CI 0.59-0.94). The advantage was driven by a reduction in ischemic stroke (2.5% vs 3.4%; HR 0.71; 95% CI 0.55-0.91), with no significant differences in the other components of the primary outcome.

There were no significant differences in disability, all-cause mortality, or bleeding at 90 days.

Looking at the quality of stroke care, the researchers found the intervention slightly increased the proportion of evidence-based performance measures that were met compared with usual care (91.4% vs 89.7%; adjusted HR 1.26; 95% CI 1.19-1.33). And delving deeper, they showed that AI assistance increased use of dual antiplatelet therapy, dysphagia screening, and deep vein thrombosis prophylaxis acutely as well as anticoagulation for atrial fibrillation both acutely and at discharge.

What’s really important about the trial is not just that AI led to more patients receiving guideline-recommended medical therapy, but that it reduced new vascular events, Schwamm said. The absolute benefit was only 1%, but the advantage for AI assistance emerged early and was sustained through the end of follow-up, he noted. Moreover, he added, that reduction in events translates to a lot of people who will avoid an adverse outcome in a country the size of China.

“It’s actually quite striking to see that kind of an effect,” Schwamm said.

AI Will ‘Not Necessarily Replace You’

Speaking with TCTMD, Seemant Chaturvedi, MD (University of Maryland School of Medicine, Baltimore), said some of his colleagues, after seeing the GOLDEN BRIDGE II results, were joking about whether an AI tool that can process clinical and imaging information and generate evidence-based recommendations will put them out of business.

In fact, he said, there is concern among some specialties about the role of AI, including in radiology, where AI has become advanced in interpreting images. “But on the other hand, I think you always need to put the interpretation into the context, which the human physician can do better,” Chaturvedi said. “Use the AI tool to help you in clinician diagnosis and treatment, not necessarily replace you.”

I think there’s tremendous value in having these sorts of tools shine a spotlight on areas that are not controversial. Lee Schwamm

He pointed out that the quality of care was high even in the control group, and that event rates were low overall, with the AI tool providing a “slight improvement” over usual care. Nonetheless, there is variability in quality of care across centers, “so I think there is the potential role for AI tools such as this to sort of raise the [bar] for all hospitals and to make sure that they’re providing high-quality care,” he said.

Schwamm pointed out that a subgroup analysis indicated that the AI intervention had the greatest impact on new vascular events at hospitals in the less-developed western part of China.

“I think where these kinds of approaches will have the most profound effect is in parts of the country where you don’t have routine access to high-quality care, [where] you don’t have access to a vascular neurologist, or even a neurologist,” he said. “This could make a really profound difference.”

But even in major medical centers in the US, “I think there’s tremendous value in having these sorts of tools shine a spotlight on areas that are not controversial,” Schwamm added. “It’s a way of supporting the work that we all agree should be done. . . . This is really more like a reminder system than it is a replacement of a physician for the purposes of diagnosis.”

Looking ahead, Schwamm sees these types of AI tools being integrated into large EMR systems in the US within the next 5 years, although, he added, “I think the business models are going to dictate adoption more than the science.”