136 fewer secondary vascular events among 11,054 patients over one year
When a stroke strikes, the window for sound clinical judgment is narrow and the stakes are absolute. Across 77 hospitals in China, researchers asked whether artificial intelligence could help physicians make better decisions in that compressed moment — and over 21,603 patients and two and a half years, the answer proved measurable: a 27 percent reduction in secondary vascular events at one year. The trial, published in The BMJ in March 2026, does not herald the replacement of human judgment, but rather its augmentation — a quiet demonstration that well-designed tools can help medicine keep its own promises more consistently.
- Stroke remains one of China's leading causes of death and disability, and the gap between best-practice guidelines and actual bedside care is wide enough to cost lives.
- An AI clinical decision support system, deployed across 77 hospitals, analyzed brain scans and patient histories to recommend evidence-based treatments — injecting consistency into a process prone to variation.
- The intervention cut secondary vascular events from 5.5% to 4% at twelve months, sparing roughly 136 patients from another stroke, a heart attack, or cardiovascular death.
- Crucially, the tool required no major infrastructure overhaul, slotting into existing hospital networks and raising care quality scores from 89.8% to 91.4% — suggesting scalability is real, not theoretical.
- Disability rates and overall mortality did not shift, focusing the tool's demonstrated value precisely where it operated: preventing the dangerous cascade that follows a first stroke, not rewriting every outcome downstream.
A stroke arrives without warning, and the neurologist who receives the patient must move quickly — choosing a treatment pathway that will either reduce or compound the risk of what comes next. In 77 hospitals across China, researchers spent two and a half years testing whether an AI-assisted decision support system could make that choice more reliably correct.
The trial enrolled 21,603 patients with acute ischemic stroke between January 2021 and June 2023. Half were treated at hospitals where physicians used an AI tool that analyzed brain scans and recommended evidence-based interventions, accounting for age, medication history, lifestyle, and regional care patterns. The other half received standard care. Both groups were comparable: average age 67, roughly one-third women.
The results, published in The BMJ in March 2026, showed a consistent and widening gap. At three months, new vascular events — another stroke, a heart attack, or cardiovascular death — occurred in 2.9 percent of AI-supported patients versus 3.9 percent in the control group. By twelve months, that spread had grown to 4 percent versus 5.5 percent, a 27 percent relative reduction translating to approximately 136 fewer secondary events among the 11,054 patients who received AI-guided care.
The system also lifted care quality scores, with AI-assisted hospitals reaching 91.4 percent adherence to best practices compared to 89.8 percent in standard care. Notably, the tool required no major infrastructure changes — it integrated into existing hospital information systems with physician training alone, a detail the researchers emphasize as critical for scalability in resource-constrained settings.
What the AI did not change is equally telling: disability outcomes, overall mortality, and bleeding risk remained statistically similar between groups. The benefit was specific — a reduction in the secondary vascular cascade that so often follows a first stroke, not a broad rewriting of every downstream outcome.
The authors are candid about limitations, including the fact that hospitals rather than individual patients were randomized, leaving room for variation in outpatient follow-up to influence results. Still, they frame the trial as proof of concept for a scalable model: AI not as a replacement for clinical judgment, but as a tool that helps physicians apply evidence more consistently — particularly in rural and under-resourced hospitals where specialist expertise is scarce and the burden of cerebrovascular disease is heaviest.
A stroke arrives without warning. The patient reaches the hospital within hours, scans are ordered, and a neurologist must decide: which treatment, which pathway, which prevention strategy will give this person the best chance at recovery? In 77 hospitals across China, researchers tested whether an artificial intelligence system could sharpen that decision-making process. The results, published in The BMJ in March 2026, suggest it can—substantially.
The trial followed 21,603 patients with acute ischemic stroke admitted between January 2021 and June 2023. Half received care guided by a clinical decision support system that used AI to analyze their brain scans and recommend evidence-based treatments. The other half received standard care. The patients were evenly split: average age 67, about one-third women, distributed across 77 hospitals in a country where stroke remains a leading cause of death and disability.
The difference emerged quickly. At three months, 2.9 percent of patients in the AI-supported group had experienced a new vascular event—another stroke, a heart attack, or death from cardiovascular causes. In the standard care group, that figure was 3.9 percent. By the one-year mark, the gap had widened slightly: 4 percent versus 5.5 percent. That 27 percent reduction at 12 months translated to roughly 136 fewer secondary vascular events among the 11,054 patients who received the AI system's guidance. The reduction held steady across all three measurement points.
The system also improved what researchers call care quality performance measures—the degree to which hospitals followed established best practices for stroke management. Hospitals using the AI tool achieved a 91.4 percent quality score compared to 89.8 percent in the control group. Importantly, the tool proved simple to integrate. Physicians received training, but the system slotted into existing hospital information networks without requiring major infrastructure overhaul. It accounted for variables like patient age, medication history, lifestyle factors, and regional differences in care patterns.
What the AI system did not change was disability outcomes or overall mortality at three, six, or twelve months. Nor did it affect bleeding risk—neither moderate nor severe bleeding differed between groups. This matters because it suggests the tool's benefit came not from preventing death across the board, but from preventing a specific, measurable harm: the cascade of secondary vascular events that often follows a first stroke.
The researchers acknowledge limitations. They randomized hospitals rather than individual patients, meaning variations in outpatient care and follow-up practices between facilities could have influenced results. Yet they emphasize that the system's ease of use and compatibility with existing hospital infrastructure make it scalable—particularly valuable in regions where stroke burden is heavy but resources are thin. China faces exactly this challenge: a large population, high rates of cerebrovascular disease, and hospitals in rural and less-developed areas that lack the specialist expertise available in major urban centers.
The authors frame their finding as a proof of concept for a broader approach: AI-assisted management that focuses on in-hospital care and secondary prevention strategies. Not a replacement for human judgment, but a tool that helps clinicians apply evidence more consistently, catch details they might miss, and guide patients toward the treatments most likely to prevent the next stroke. In a field where most AI stroke tools remain unevaluated and their clinical use limited, a rigorous trial showing a 26 to 27 percent reduction in secondary vascular events represents a meaningful step forward.
Citas Notables
The stroke CDSS offers a promising approach to providing high quality care for patients with acute ischemic stroke admitted to hospital, particularly for resource constrained regions with a heavy burden of cerebrovascular diseases like China.— Study authors
La Conversación del Hearth Otra perspectiva de la historia
Why does it matter that the AI system didn't change overall mortality or disability? Doesn't that seem like a missed opportunity?
It's actually the opposite. If the system had reduced mortality broadly, we'd wonder if something else was happening—better overall care, different patient populations. What it did instead was very specific: it prevented the second and third strokes, the heart attacks that follow. That's the real work of secondary prevention, and it's harder to achieve than you'd think.
So the patients who got the AI support didn't live longer, but they had fewer new strokes?
Exactly. They lived the same length of time, but with fewer catastrophic events in that year. For someone who's already had one stroke, that's the difference between independence and another hospitalization, between staying home and losing more function.
The study was done in China. Does that limit what we can learn from it?
Not really. The burden of stroke is global, and resource constraints are global. What matters is that the system was tested in real hospitals with real workflows, not in a controlled lab. It worked within existing infrastructure, which is the only way something like this scales.
What's the actual mechanism? How does the AI improve care if doctors still make the final decision?
It's about consistency and speed. A radiologist in a rural hospital might miss a subtle finding on a scan. The AI flags it. A doctor might forget the latest guideline for secondary prevention. The system reminds them. It's not replacing judgment—it's making sure good judgment gets applied every time, not just when the right specialist is on shift.