Ischaemic stroke survivors who received care recommendations from an artificial intelligence (AI)-based system experienced fewer recurrent strokes and heart attacks, and a lower rate of vascular death, within three months—as compared to people whose stroke treatment was not guided by AI tools—in the recently presented GOLDEN BRIDGE II study. Lead study author Zixiao Li (Beijing Tiantan Hospital/Capital Medical University, Beijing, China) delivered findings from the trial in a late-breaking science session at the ongoing International Stroke Conference (ISC; 7–9 February, Phoenix, USA).
“This research showed that an AI-based clinical decision support system for stroke care was effective and feasible in clinical settings in China, and improved patient outcomes,” Li said. “This type of technology aids neurologists by facilitating the sharing of information between humans and AI, using their combined strengths.”
In the GOLDEN BRIDGE II clinical trial, 77 hospitals across various regions of China were randomly assigned to deliver diagnosis and treatment for ischaemic stroke patients based on either recommendations from the AI technology system, or assessments and recommendations by the hospitals’ stroke care teams. The AI system integrated participants’ brain imaging scans—interpreted by AI—with established clinical knowledge for stroke diagnosis, stroke classification, and guideline-recommended treatment and strategies to prevent secondary stroke.
Across 21,603 hospitalised, adult acute ischaemic stroke patients (average age, 67 years; one-third women) included in the study, researchers then measured the number of vascular events—ischaemic and haemorrhagic strokes, heart attacks, or death due to a vascular event—among all patients after their initial ischaemic stroke during a three-month follow-up period. Some 11,054 patients received AI evaluation and treatment, while 10,549 received standard stroke care, and the vast majority of patients (21,579) were included in the final data evaluation after completing the three-month follow-up.
The analysis found that using an AI-based clinical decision support system reduced the chances of new vascular events by 25.6% during this initial three-month period, and also improved stroke care quality, with patients more likely to be treated with guideline-directed medical therapy. At three months, participants treated at hospitals using AI support also experienced fewer total vascular events compared to people receiving standard post-stroke evaluation and treatment (2.9% vs. 3.9%). Li and colleagues observed no statistically significant differences in physical disability levels between patients in either of the two groups at three months, as assessed using modified Rankin scale (mRS) scores.
“The reduction in new vascular events is a significant finding, because it shows that AI has the potential to make a real difference in stroke care and benefit this large population of stroke survivors,” Li added. “In the future, we hope to have more AI applications validated through clinical research and hope that the clinical decision support system can be expanded to include more aspects of stroke care, including reperfusion therapy and long-term secondary prevention, rehabilitation, and so on. At the same time, we also hope that AI applications can be broadened to apply to other health conditions.”
Li further noted that neurologists in the hospitals testing the AI decision support technology completed training on the system before the study began. He also detailed that stroke care quality was measured by internationally recognised composite scores of evidence-based performance measures for acute ischaemic stroke care quality, including eight measures at the beginning of hospitalisation and five measures at discharge.
Notable limitations of the GOLDEN BRIDGE II study include the fact that entire hospitals were randomised to the AI-based strategy or standard care, rather than individual patients, and differences in care patterns and outcomes between hospitals—and subsequent outpatient care—may have impacted the results too, according to Li.
The researchers believe that questions remain over whether or not these AI-led improvements in care and outcomes can be sustained, thus requiring further evaluation, and the functionality of the AI-based clinical decision support system may need to be constantly updated to keep pace with evidence-based clinical guideline revisions. More extensive and sustainable clinical application models of the system for health conditions beyond stroke, and for use in other countries, need to be explored as well.
