AI-generated research brief — verify at source
AI-Integrated Stroke CDSS Cuts Three-Month Vascular Events by 26% in Chinese Trial
A large cluster randomised trial found that deploying an artificial intelligence–assisted clinical decision support system (CDSS) in hospitals across China reduced the rate of new vascular events within three months of acute ischaemic stroke by approximately 26% compared with usual care (2.9% vs 3.9%; adjusted HR 0.74, 95% CI 0.58–0.93, P=0.01). The result held at the cluster level and was accompanied by meaningful improvements in adherence to evidence-based stroke care standards, suggesting the effect reflects genuine improvements in care delivery rather than statistical artefact.
What Was Studied
The trial investigated whether a hospital-level CDSS—incorporating AI-assisted imaging analysis, automated stroke cause classification, and evidence-based treatment recommendations—could reduce the occurrence of new vascular events (a composite of ischaemic stroke, haemorrhagic stroke, myocardial infarction, and vascular death) within three months of symptom onset in patients with acute ischaemic stroke. The question is clinically important because evidence-to-practice gaps in acute stroke care remain a recognised driver of avoidable morbidity and mortality, particularly in settings where specialist expertise and standardised workflows may be inconsistently available.
How It Was Studied
GOLDEN BRIDGE II was a multicentre, cluster randomised clinical trial conducted across 77 hospitals in China, with 38 hospitals assigned to the CDSS intervention and 39 to usual care. Patients were eligible if they had acute ischaemic stroke and were admitted within seven days of symptom onset; a total of 21,603 patients were enrolled between January 2021 and June 2023, with 11,054 in the intervention group and 10,549 in the control group. The unit of randomisation was the hospital rather than the individual patient, which is appropriate for system-level interventions but introduces specific analytical considerations. Follow-up extended to 12 months, with the primary endpoint assessed at three months and secondary vascular, disability, and mortality outcomes assessed at three, six, and twelve months.
What Was Observed
- Primary outcome — reduced vascular events at three months: New vascular events occurred in 2.9% of intervention patients versus 3.9% of control patients, representing roughly a one-quarter reduction in risk (adjusted HR 0.74, 95% CI 0.58–0.93, P=0.01). This effect was confirmed in a cluster-level analysis, where the absolute difference was −0.01 (95% CI −0.02 to −0.004, P=0.003), reinforcing that the benefit was consistent across hospital units and not driven by outlier facilities.
- Stroke care quality — process measure improvement: Patients in the intervention group achieved a higher composite measure of evidence-based performance standards, with 91.4% of indicator-level assessments met (77,049 of 84,276 eligible instances), compared with lower adherence in the control group. This improvement in process quality provides a plausible mechanistic explanation for the reduction in vascular events.
- Longer-term outcomes: The beneficial effect on vascular events was reported to persist at six and twelve months, and reductions in disability (modified Rankin Scale score 3–6) and all-cause mortality were included among the secondary outcomes, indicating durability of the intervention effect beyond the primary three-month window.
Why This Matters
This trial provides some of the strongest evidence to date that AI-integrated decision support tools can improve not just process metrics but hard clinical outcomes—specifically recurrent vascular events—at meaningful scale across a heterogeneous hospital network. The findings support the hypothesis that systematising guideline adherence through real-time decision assistance can close evidence-practice gaps in stroke care. More broadly, the results are relevant to health systems evaluating digital health infrastructure as a means of achieving consistent stroke care quality across institutions with varying levels of specialist resources.
How to Read This Result
This is a high-quality, large-scale RCT with a statistically significant and directionally consistent primary result, though the cluster randomisation design may introduce residual hospital-level imbalances, and the findings were generated entirely within the Chinese healthcare system, which may limit direct generalisability to other health system contexts.
Limitations
The abstract does not explicitly report study limitations.