AI-guided stroke system cut new vascular events by 27% across 77 Chinese hospitals

BMJ Group

An AI-powered clinical decision support system that analyzes brain scans and pairs the results with treatment recommendations reduced new vascular events — strokes, heart attacks, and related deaths — by 27% over 12 months in a large randomized trial across China. The study, published in The BMJ, followed 21,603 patients with acute ischaemic stroke admitted to 77 hospitals between January 2021 and June 2023.

The numbers are specific and consistent. At three months, 2.9% of patients in the AI-supported group experienced a new vascular event, compared with 3.9% in the control group — a 26% relative reduction. By 12 months, the gap held: 4.0% versus 5.5%, a 27% reduction. In absolute terms, that translates to roughly 15 fewer vascular events per 1,000 patients treated.

How the system works in practice

The tool is a stroke clinical decision support system (CDSS) that combines AI-assisted imaging analysis with evidence-based treatment protocols. When a patient arrives with acute ischaemic stroke, the system classifies the stroke's cause from imaging data and generates treatment recommendations tailored to that classification.

Physicians in the intervention group received training on the system, which integrated directly into their hospital's existing information infrastructure. The design was deliberate: rather than requiring new hardware or workflows, the CDSS worked within systems clinicians already used. Patient factors including age, medication history, and lifestyle were incorporated into the recommendations.

The trial assigned 38 hospitals (11,054 patients) to the intervention group and 39 hospitals (10,549 patients) to usual care. This cluster randomization — randomizing hospitals rather than individual patients — is a practical necessity for testing system-wide tools, but it introduces a limitation worth noting.

Stroke care quality scores improved, but mortality did not change

Beyond vascular events, patients in the CDSS group scored higher on composite stroke care quality measures: 91.4% versus 89.8%. That difference, while modest in percentage terms, reflects systematic improvements in adherence to treatment protocols across dozens of hospitals.

But the system did not move every needle. There were no significant differences in disability rates or all-cause mortality between the two groups at three, six, or twelve months. Bleeding complications — both moderate-to-severe and total — also showed no difference. The AI tool appears to prevent secondary vascular events without affecting the initial stroke's neurological damage or broader mortality.

This is an important distinction. A tool that reduces new strokes and heart attacks by a quarter is valuable. A tool that reduces disability and death from the initial stroke would be transformative. This one does the former, not the latter.

The cluster randomization tradeoff

Because the trial randomized hospitals rather than patients, differences between the intervention and control sites could influence the results. Variations in hospital resources, regional patient demographics, physician expertise, and follow-up care patterns all represent potential confounders. The researchers adjusted for hospital region and grade, but residual differences are difficult to eliminate entirely.

There's also the question of outpatient care. The CDSS operated during hospitalization. What happened after discharge — whether patients received consistent follow-up, adhered to medications, or accessed rehabilitation — varied across sites and was not controlled by the intervention. Some of the 12-month benefit may reflect differences in post-discharge care rather than the AI system itself.

The trial was conducted entirely in China, a country with particular stroke demographics. China carries one of the world's heaviest burdens of cerebrovascular disease, and its hospital infrastructure ranges widely from urban tertiary centers to rural facilities. Whether similar results would emerge in healthcare systems with different baseline care quality, patient populations, or resource constraints remains untested.

Where resource-constrained hospitals stand to gain most

The researchers position the tool as especially relevant for hospitals with limited specialist access. In well-resourced stroke centers, experienced neurologists already make many of the decisions the CDSS automates. The system's value may be greatest where such expertise is thin — smaller hospitals, rural facilities, regions with high stroke burden and few vascular neurologists.

This framing matters because it sets realistic expectations. The CDSS is not replacing clinical judgment at top-tier academic hospitals. It's standardizing decision-making at facilities where evidence-based protocols might otherwise be applied inconsistently.

The average patient age was 67, and only 36% were female. Whether the system performs equally well across different demographic groups — younger patients, women, patients with atypical stroke presentations — is not addressed by this trial.

A 27% reduction in context

Stroke remains a leading cause of death and disability worldwide, and secondary prevention — stopping the next stroke after the first — is one of the most impactful interventions in vascular medicine. A 27% relative reduction in composite vascular events is clinically meaningful, comparable to the benefit seen with some antiplatelet and statin regimens.

That the benefit came from a software system integrated into existing hospital workflows, rather than a new drug or device, is notable. The tool requires no per-patient consumables, no surgical implantation, no ongoing pharmaceutical costs. Its scalability depends primarily on IT infrastructure and clinician training — both cheaper to deploy than most medical interventions.

But scalability is not the same as proven generalizability. The CDSS worked in this population, in these hospitals, in this healthcare system. Extending those results to other countries, other stroke subtypes, or other clinical workflows will require further validation. For now, the evidence supports a specific claim: in Chinese hospitals treating acute ischaemic stroke, this AI-guided system reduced the rate of subsequent vascular events without increasing harm.