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.
Key Findings
Results
Four baseline predictors were identified for long-term cryptogenic ischemic stroke (CIS) risk in migraine patients with contrast-enhanced transcranial Doppler-confirmed right-to-left shunt (RLS), forming the Migraine Risk Stratification Score (MRSS).
Age ≥50 years was assigned 1 point.
Migraine with aura was assigned 2 points.
Permanent RLS and large RLS were assigned 2 points each.
Points were derived from Cox regression coefficients and hazard ratios in multivariable Cox proportional hazards modeling; total score ranged 0-7.
Results
The score separated patients into low-risk (0-4 points) and high-risk (5-7 points) groups with markedly different 10-year CIS risk.
10-year CIS risk was 21.77% in the high-risk group vs. 4.68% in the low-risk group.
The MRSS showed moderate-to-good discrimination in the development cohort and after five-fold cross-validation.
Development cohort time-dependent AUCs were 0.771, 0.777, and 0.751 at 3, 5, and 7 years.
After five-fold cross-validation, AUCs were 0.753, 0.768, and 0.744 at 3, 5, and 7 years.
Performance was also evaluated with calibration plots and decision curve analysis.
Results
In the external validation cohort (n=237), the MRSS had AUCs of 0.917, 0.863, and 0.808 at 3, 5, and 7 years.
The 3-year external AUC was based on only five events and should be interpreted cautiously.
External AUCs were numerically higher than in the development cohort at all horizons.
Results
The MRSS had high negative predictive value but low positive predictive value and F1 score across time horizons.
Negative predictive value ranged 0.96-0.99.
Positive predictive value ranged 0.05-0.19.
F1 score ranged 0.10-0.30.
Conclusions
The authors characterize MRSS as a rule-out-oriented baseline stratification tool rather than a precise absolute-risk tool for treatment selection.
This interpretation follows from the high NPV and low PPV/F1.
The authors state that further prospective validation and recalibration are required.
What This Means
This study looked at people with migraine who also have a right-to-left shunt (an abnormal opening that lets blood bypass the lungs, detected by a special ultrasound), and asked who is more likely to have a stroke of unknown cause over 10 years. The researchers built a simple point score from 978 patients and tested it in a separate group of 237 patients. The score gives 1 point for age 50 or older, 2 points for migraine with aura, and 2 points each for a permanent shunt and a large shunt.
People scoring 5-7 points had about a 22% chance of this type of stroke over 10 years, compared with about 5% for those scoring 0-4. The score was reasonably good at telling higher-risk from lower-risk patients, including in the separate validation group, though the earliest external result rested on only five stroke events. Its main strength was in identifying people who were unlikely to have a stroke: a low score was very reliable (negative predictive value 0.96-0.99). When it flagged someone as high risk, most of those people still did not have a stroke (positive predictive value 0.05-0.19).
This research suggests the score could be useful for identifying patients at low risk, rather than for predicting exactly who will have a stroke or deciding on treatment such as shunt closure. The authors note that the tool needs further testing in prospective studies and recalibration before it can be relied upon in practice.
Zhu L, Xing Y, Fang B, An L, Wang L. (2026). Development and validation of a prediction model for cryptogenic ischemic stroke risk in migraine patients with right-to-left shunt: A 10-year study.. Cephalalgia : an international journal of headache. https://doi.org/10.1177/03331024261499744