AI-Based Decision Support Boosts Post-Stroke Prevention

PHOENIX — An artificial intelligence decision support intervention for acute ischemic stroke patients at Chinese hospitals reduced subsequent vascular events, the GOLDEN BRIDGE II trial showed.

The proportion of patients with ischemic or hemorrhagic stroke, myocardial infarction, or vascular death within 3 months after their initial stroke was a relative 25.6% lower with the intervention compared with usual care (2.9% vs 3.9%, P=0.013).

The difference appeared to be driven by modest improvements in the rate of patients being prescribed dual antiplatelet therapy, statins, and anticoagulation for atrial fibrillation acutely and at discharge, Zixiao Li, MD, PhD, of Beijing Tiantan Hospital and Capital Medical University in Beijing, reported at the American Stroke Association’s International Stroke Conference.

“It’s really exciting to see some difference from that type of system,” commented Larry B. Goldstein, MD, chair of neurology at the University of Kentucky in Lexington. “This is one of the first [trials] using decision support that really showed an effect on outcomes.”

The absolute impact was fairly small but significant due to the large numbers in the trial — 21,603 patients randomized at 77 hospitals, he pointed out. “Whether the benefit of instituting that system makes sense on a wide scale level, I think that also needs to be worked out.”

For China’s health system, it made sense. “China faces a huge stroke burden,” Li noted. “One of the effective strategies for the fight is to translate guideline into clinical practice.”

His group’s prior trial had shown the potential for a multi-faceted intervention to get hospital personnel to adhere to guidelines in China.

The intervention in the new trial revolved around the artificial intelligence (AI) decision support system. The algorithm analyzed the patient’s MRI, identified infarct lesions, and provided imaging analysis and etiology classification for physicians. Then it integrated clinical information from the the electronic medical record and other hospital information systems. That information was put together with a stroke clinical knowledge base to provide physicians with guideline-based recommendations for treatment of stroke.

The trial cluster randomized hospitals to either use that AI system or continue usual care. Centers were included if they were tertiary or secondary-grade (not community hospitals), had an emergency department and neurological wards, and were equipped with 1.5 or 3 T MRI scanners. These criteria provided healthcare setting diversity that would contribute to generalizability, according to Li.

Adults seen at those centers with ischemic stroke within 7 days of symptom onset and confirmed by brain imaging were included if they consented.

The reduction in the primary endpoint was driven by a relative 29% reduction in recurrent ischemic stroke with the intervention (2.5% vs 3.4%, P=0.008).

Acute quality process measures that were significantly improved with the intervention were utilization of dual antiplatelet therapy (76.2% vs 69.6%), anticoagulation for atrial fibrillation (77.0% vs 69.3%), dysphagia screening (98.5% vs 91.2%), and deep vein thrombosis prophylaxis (37.1% vs 30.0%). The only discharge performance measure met significantly more often in the intervention group was prescription of anticoagulation for atrial fibrillation (77.3% vs 67.5%).

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

One key question, Goldstein said, is whether the findings are specific to the health system and patient population studied. “They do have a unified health system, which at least we in the United States don’t. So that could be a major issue,” he noted.

Li agreed that the AI intervention needs to be examined in different healthcare settings.

Please enable JavaScript to view the

comments powered by Disqus.