Development and validation of a prediction model for cryptogenic ischemic stroke risk in migraine patients with right-to-left shunt: A 10-year study
Lixia Zhu, Yingqi Xing, Boyu Fang, Lisi An, Lijuan WangAim
Long-term risk stratification for cryptogenic ischemic stroke (CIS) remains uncertain in patients with migraine and contrast-enhanced transcranial Doppler ultrasound-confirmed right-to-left shunting (RLS). Therefore, we aimed to develop and externally validate a simple baseline score for identifying individuals at low risk.
Methods
This 10-year study included 978 and 237 patients in the development and external validation cohorts, respectively. Internal validation involved a five-fold cross-validation strategy. Multivariable Cox proportional hazards modeling was used to develop the Migraine Risk Stratification Score (MRSS). Model performance was comprehensively evaluated using time-dependent receiver operating characteristic (ROC) curves at 3, 5, and 7 years, calibration plots, and decision curve analysis. Model interpretation was based on the Cox regression coefficients, hazard ratios, and integer points assigned to each predictor. Survival divergence was assessed by Kaplan–Meier analysis.
Results
Age ≥50 years (1 point), migraine with aura (2 points), and permanent and large RLS (2 points each) were identified as predictors. Patients were stratified into low- (0–4 points) and high-risk (5–7 points) groups based on total score. The high-risk group exhibited a significantly higher 10-year CIS risk (21.77% vs. 4.68%; hazard ratio: 5.06; 95% confidence interval: 3.31–7.74). The areas under the curve (AUCs) at 3, 5, and 7 years were 0.771, 0.777, and 0.751 in the development cohort; 0.753, 0.768, and 0.744 after five-fold cross-validation; and 0.917, 0.863, and 0.808 in external validation, respectively. The 3-year external AUC was based on only five events; therefore, it should be interpreted cautiously. Across evaluated horizons, the negative-predictive value was high (0.96–0.99), whereas the positive-predictive value (0.05–0.19) and F1 score (0.10–0.30) were low. Kaplan–Meier curves confirmed significant long-term stroke-free survival stratification (log-rank
Conclusion
MRSS exhibited characteristics consistent with a rule-out-oriented baseline risk stratification tool rather than a precise individualized absolute risk assessment for treatment selection. Further prospective validation and recalibration are required.