Quick answer
Most of the evidence in stroke care comes from drugs and devices. GOLDEN BRIDGE II tested something different: a clinical decision support system that integrates AI-assisted imaging, automatic stroke subtyping and evidence-based treatment prompts into the workflow. Seventy-seven Chinese hospitals were cluster-randomised, 21,603 patients were enrolled, and the intervention cut new vascular events within three months from 3.9% to 2.9% — a 26% relative reduction — with measurable gains in care quality. The results appeared in The BMJ on 21 March 2026 [1][2][4].
Key takeaways
- Design: multicentre, cluster-randomised controlled trial across 77 Chinese hospitals [2].
- Population: 21,603 adults with acute ischaemic stroke within seven days, enrolled January 2021 to June 2023 [2].
- Allocation: 38 hospitals to the CDSS, 39 to usual care [2].
- Primary outcome: new vascular events (ischaemic stroke, haemorrhagic stroke, myocardial infarction, vascular death) at three months — 2.9% versus 3.9%, adjusted HR 0.74 (95% CI 0.58–0.93, P=0.01) [2][3].
- Care quality composite: 91.4% versus 89.8%, adjusted OR 1.21 (95% CI 1.17–1.26) [4].
What the system does
The CDSS is not a chatbot. It reads the imaging, classifies the stroke mechanism (large-artery atherosclerosis, cardioembolism, small-vessel disease and so on), and then surfaces the guideline-recommended actions for that subtype — anticoagulation for atrial fibrillation, dual antiplatelet therapy for minor stroke within 24 hours, swallowing assessment, venous thromboembolism prophylaxis for immobile patients. The point is to convert guidelines that clinicians know into actions that happen [2][4].
The key performance indicators moved most where the gap was largest: dual antiplatelet therapy within 24 hours for non-disabling stroke rose from 69.6% to 76.2%; anticoagulation for patients with atrial fibrillation from 69.3% to 77.0%; swallowing assessment from 91.2% to 98.5%; and thromboprophylaxis for immobile patients from 30.0% to 37.1% [4].
What did not change
Disability and all-cause mortality at 3, 6 and 12 months did not differ significantly, and neither did moderate-to-severe or any bleeding. That pattern — fewer recurrent vascular events, no mortality difference, no excess bleeding — is a plausible result for better secondary prevention rather than better acute reperfusion [4].
Cluster randomisation by hospital means the unit of analysis is the institution, and the effect includes whatever enthusiasm and attention came with implementing a new system. Hospitals that volunteered may not represent the country. The absolute risk reduction of 1 percentage point at three months is small per patient, though it is a cheap intervention to scale. Eight years elapsed between product development and device approval, which shows how long this kind of translation takes [2][4][5].
The condition in context: stroke in China
China has the world's largest stroke burden, and secondary prevention is where outcomes are most improvable.
| Region | What the data shows | Source |
|---|---|---|
| Recurrence | Recurrent vascular events drive much of the long-term burden after stroke [2]. | China Stroke Association [2] |
| Guideline gap | Baseline adherence varied widely across indicators, from 30% to over 90% [4]. | BMJ China [4] |
| Scale | 77 hospitals and 21,603 patients make this one of the largest AI implementation trials in stroke [2]. | China Stroke Association [2] |
Because the intervention is software rather than a drug, the cost of spreading it is far lower than for a new medicine [5].
Why it matters
| Potential | What the report suggests | Source |
|---|---|---|
| Outcomes | 26% relative reduction in new vascular events [2][3]. | [2][3] |
| Quality | Adherence improved across most indicators [4]. | [4] |
| Caution | No mortality benefit; cluster design [4][5]. | [4][5] |
Availability and limits
- This is an editorial summary of the publication, not a treatment recommendation [1][2].
- The system requires hospital IT integration and clinician training [5].
- Effects in non-Chinese health systems are untested [2].
- Absolute benefit per patient is modest [3].
- It supports, rather than replaces, clinical judgement [4].
What to ask before booking
| Question | Why it matters |
|---|---|
| Does this hospital use a stroke care pathway with decision support? | Adherence to secondary prevention differs widely [4]. |
| Will I be assessed for swallowing and thromboprophylaxis? | These were the largest quality gaps [4]. |
| Which department handles foreign patients, and in what language? | Tiantan is a working hospital we do not book; confirm international-department coordination and language support. |
| What is my secondary prevention plan on discharge? | The benefit came from implementing prevention, not acute therapy [2]. |
Questions patients ask
What was tested?
A stroke clinical decision support system integrating AI imaging analysis, automatic stroke subtyping and guideline prompts, compared with usual care in a cluster-randomised trial across 77 hospitals [2].
What was the result?
New vascular events within three months fell from 3.9% to 2.9% (adjusted HR 0.74, P=0.01), a 26% relative reduction [2][3].
Did mortality improve?
No — disability and all-cause mortality at 3, 6 and 12 months did not differ significantly [4].
Where was it published?
The BMJ, online 21 March 2026 [1][2].
Is this relevant to international patients?
It is useful context for stroke care quality in China. Tiantan is a working hospital we do not book — contact its international department directly [1].