Key result
Machine learning using routine blood data achieves 0.68 AUROC for stratifying AIS risk in hypertension.
Why the study?
Hypertensive patients are at risk for acute ischemic stroke, but the absence of accessible tools for near-term risk stratification hinders prevention.
Does a machine learning model using routine blood data accurately stratify acute ischemic stroke risk in hypertensive patients?
Cohort (n=4,230)
Yes
Does a machine learning model using routine blood data accurately stratify acute ischemic stroke risk in hypertensive patients?
Effect estimate: AUROC 0.68 (95% CI 0.64-0.72)
A machine learning model using routine blood biomarkers showed moderate accuracy (AUROC 0.68) in stratifying near-term acute ischemic stroke risk among hypertensive patients.
Low sensitivity cautions against routine clinical use in AIS risk stratification; leaves open prospective validation and model refinement.
Background and aims Hypertensive patients are at risk of developing acute ischemic stroke (AIS). The absence of accessible tools for near-term risk stratification hinders prevention. An AIS risk stratification model is independently validated in a hypertensive cohort in this study. Methods A model for AIS risk stratification in hypertensive patients by repurposing demographic data and 19 routine blood biomarkers from the preceding 30 days (territory-wide cohort, n=273,780) was developed. It was then applied to an independent cohort of 4,230 hypertensive patients, comprising 151 AIS cases and 4,079 non-AIS controls. Performance was evaluated via AUROC. A dual-threshold method stratified patients into low-, medium-, and high-risk groups, using thresholds pre-defined in the development cohort at 97.5% sensitivity and 90% specificity. Results Cases were older (75.6 ± 12.4 vs. 70.2 ± 14.4 years; p-value <0.05) and had a similar male proportion (55.6% vs. 53.2%; p-value = 0.62) as controls. The model achieved an AUROC of 0.68 (95% CI: 0.64–0.72) and stratified patients into low- (n=619; 14.6%), medium- (n=2,787; 65.9%), and high-risk groups. The high-risk group identified 39.7% (60/151) of AIS events, with a positive predictive value of 7.3%. The low-risk group excluded 14.6% (619/4230) patients, achieved a negative predictive value of 98.9% and missed 7 (<5%) AIS cases. Conclusions The model that repurposes routine blood test data within 30 days to stratify AIS risk in a >4k hypertensive cohort was tested. The practicality and scalability of the model for opportunistic screening in primary care is validated, enabling targeted preventive therapy and bridging a critical gap in patient management. Conflict of interest
No takes yet. Share an insight, caveat, or question.
Lau et al. (2026) conducted a cohort in Hypertension and Acute Ischemic Stroke (n=4,230). Machine learning model using repurposed routine blood data was evaluated on Acute ischemic stroke (AIS) risk stratification performance via AUROC (AUROC 0.68, 95% CI 0.64-0.72). A machine learning model using repurposed routine blood data stratified acute ischemic stroke risk in hypertensive patients, achieving an AUROC of 0.68 (95% CI 0.64-0.72).
Synapse has enriched 5 closely related papers on similar clinical questions. Consider them for comparative context: