Surveillance data from stroke registries across the Asia-Pacific region consistently demonstrate gaps between evidence-based treatment recommendations and actual clinical practice. A multicentre, cluster randomised clinical trial (RCT) published in The BMJ offers population-level evidence that a clinical decision support system (CDSS) can meaningfully close that gap, reducing new vascular events among patients with acute ischaemic stroke (AIS) by a statistically significant margin across a large, geographically diverse hospital network in China.

The trial, designated GOLDEN BRIDGE II, enrolled 21,603 patients with AIS across 77 hospitals, randomising 38 hospitals to a CDSS intervention arm and 39 hospitals to usual care. Patients were eligible for inclusion if they were admitted within seven days of symptom onset. The scale of the study, spanning enrollment from January 2021 through June 2023, positions its findings among the more robust datasets available for evaluating health information technology interventions in acute cerebrovascular care.

CDSS Architecture and Intervention Design

The CDSS evaluated in GOLDEN BRIDGE II incorporated three functional components: artificial intelligence (AI)-assisted imaging analysis, classification of stroke etiology, and generation of evidence-based treatment recommendations. This integration of AI-driven diagnostic support with clinical guideline adherence tools represents a substantial departure from passive decision support modalities, which have historically demonstrated modest effectiveness in changing clinician behavior at the point of care.

Hospitals assigned to the intervention group received CDSS support embedded within the clinical workflow, enabling attending physicians to receive real-time guidance informed by imaging findings and stroke classification. The control group received no supplemental decision support beyond standard institutional protocols.

Primary Outcome: New Vascular Events at Three Months

The primary outcome was a composite measure of new vascular events at three months following initial symptom onset, defined as the occurrence of ischaemic stroke, haemorrhagic stroke, myocardial infarction, or vascular death. Among the 11,054 patients in the intervention group, new vascular events occurred in 2.9% of cases (320 patients). Among the 10,549 patients in the control group, the rate was 3.9% (416 patients). The adjusted hazard ratio (aHR) was 0.74 (95% confidence interval [CI]: 0.58 to 0.93; p=0.01), indicating a statistically significant reduction in composite vascular events associated with CDSS use.

The intervention effect remained statistically significant at the cluster level, with an estimated difference of -0.01 (95% CI: -0.02 to -0.004; p=0.003). This cluster-level confirmation is methodologically noteworthy, as cluster-randomised trials are susceptible to within-cluster correlations that can attenuate apparent treatment effects when analyses are conducted only at the individual patient level.

For clinicians practicing in Hawaii’s stroke networks, particularly those serving rural communities on neighbor islands where specialist consultation may be delayed or unavailable, this magnitude of risk reduction carries practical relevance. A 26% relative reduction in three-month vascular events within a real-world hospital network, rather than a controlled efficacy trial, suggests that CDSS tools could function as a form of distributed specialist knowledge across facilities with varying levels of neurological subspecialty coverage.

Care Quality Measures

Secondary outcomes included composite and all-or-none measures of evidence-based performance for AIS care quality. Patients in the intervention group demonstrated higher rates of adherence to evidence-based care standards, with a composite measure of 91.4% (77,049 of 84,276 eligible measure applications) compared with 89.8% (70,794 of 78,834) in the control group. The adjusted odds ratio (aOR) was 1.21 (95% CI: 1.17 to 1.26; p<0.001).

This improvement in process adherence is consistent with prior literature on CDSS interventions in cardiovascular and cerebrovascular care, which generally demonstrates that structured decision support tools increase the likelihood that clinicians will order guideline-concordant therapies, particularly in time-sensitive clinical environments where cognitive load is high and protocol recall is variable.

Long-Term Vascular Events

The benefit observed at three months persisted at twelve months. New vascular events at twelve months occurred in 4.0% of intervention group patients (440 of 11,054) compared with 5.5% of control group patients (576 of 10,549), with an aHR of 0.73 (95% CI: 0.56 to 0.95; p=0.02). This durable separation in event rates between the two groups over a twelve-month horizon suggests that the CDSS effect was not limited to acute-phase decision-making but may have influenced secondary prevention practices, including antiplatelet and anticoagulation management, as well as risk factor modification protocols.

At six months, the directional trend was consistent, though the precise figures were not fully disaggregated in the published abstract. The consistency of effect across the three-month and twelve-month time points strengthens the plausibility of a causal relationship between CDSS implementation and reduction in long-term vascular risk.

Null Findings: Disability and Mortality

No statistically significant differences were identified between the intervention and control groups for disability, defined as a modified Rankin Scale (mRS) score of 3 to 6, or for all-cause mortality at three, six, or twelve months. These null findings warrant careful interpretation rather than dismissal.

The absence of a detectable effect on disability and mortality may reflect several factors. First, the composite vascular event rate, even in the control group, was relatively low at 3.9% at three months, which may limit statistical power to detect downstream differences in functional outcomes that depend on event occurrence. Second, disability and mortality at twelve months are influenced by a broad range of post-discharge factors, including rehabilitation access, social determinants of health, and outpatient follow-up quality, domains that a hospital-based CDSS is unlikely to directly modify. Third, the modified Rankin Scale, while widely used, has measurement characteristics that may obscure clinically meaningful gradations of functional recovery in large heterogeneous cohorts.

Public health researchers tracking stroke burden in Hawaii should consider these null findings in the context of what CDSS tools are structurally positioned to accomplish. Reducing recurrent vascular events is a measurable and consequential outcome; it does not require co-occurring reductions in disability rates to represent a meaningful population health benefit.

Safety Profile

Safety outcomes, comprising moderate or severe bleeding events and all bleeding events at three, six, and twelve months, did not differ statistically between the intervention and control groups. This finding is relevant to clinical implementation discussions, as increased adherence to guideline-recommended antiplatelet and anticoagulation therapies carries a theoretical bleeding risk. The absence of a detectable increase in bleeding events across more than 21,000 patients provides reassurance that improved process adherence facilitated by CDSS did not translate into a measurable increase in hemorrhagic complications.

Methodological Considerations

The cluster randomisation design addresses several confounders inherent in individual-level randomisation of a system-level intervention. Because the CDSS functions at the hospital level, randomising individual patients within the same institution would introduce contamination bias. The 77-hospital design, with balanced allocation between intervention and control arms, provides adequate cluster-level power to detect the observed effect sizes.

Reporting considerations include the possibility of differential ascertainment of vascular events across hospital sites. Hospitals with more active clinical decision support may also be more likely to conduct follow-up imaging and laboratory assessment, potentially identifying events that would go undetected in control facilities. The published abstract does not specify the follow-up methodology applied uniformly across all sites, a question with epidemiological relevance to outcome rate comparisons.

Additionally, the trial was conducted entirely within the Chinese hospital system, which carries specific implications for generalisability. Hospital volume, staffing ratios, imaging infrastructure, and electronic health record interoperability differ substantially across health systems. Hawaiian hospitals, which operate within a United States regulatory and reimbursement framework with different electronic health record adoption patterns, would require adaptation of any similar CDSS tool to local workflow and documentation environments.

Implications for Hawaii Stroke Care Infrastructure

The Hawaii Department of Health and affiliated academic medical centers have documented ongoing challenges in achieving consistent stroke care quality benchmarks across the state’s geographically fragmented healthcare environment. Neighbor island facilities, which serve populations with higher proportions of Native Hawaiian and Pacific Islander patients who carry elevated stroke risk due to hypertension, diabetes, and obesity prevalence, operate with more constrained specialist access than facilities in Honolulu.

The GOLDEN BRIDGE II findings contribute to a growing body of evidence supporting the integration of AI-assisted CDSS tools into acute stroke care pathways. For administrators and clinical informatics teams evaluating technology investments, the trial offers cluster-level effect estimates grounded in a real-world hospital network rather than a single-center proof-of-concept study. The 21,603-patient cohort and the 2.5-year enrollment window provide a degree of