MRI-based risk stratification for recurrent ischemic stroke in embolic stroke of undetermined source.

Abstract

Leukoaraiosis and other brain MRI-assessed parameters were shown to be associated with recurrent stroke in this population. We aimed to develop an MRI-based predictive tool for risk stratification of ESUS patients. We retrospectively assessed consecutive patients who were diagnosed with ESUS and underwent brain MRI and performed a multivariable analysis with the outcome of recurrent stroke/TIA. Based on the coefficient of each covariate, we generated an integer-based point scoring system. The discrimination and calibration of the score were assessed using the area under the receiver operator characteristic curve, net reclassification improvement, integrated discrimination improvement, calibration curve, and decision curve analysis. Also, we compared the new score with a previously published score (ALM score). Among 176 patients followed for an overall period of 902.3 patient-years (median of 74 months), there were 39 recurrent ischemic stroke/TIAs (4.32 per 100 patient-years). Fazekas score (HR: 1.26, 95% CI: 1.03-1.54), enlarged perivascular space (EPVS) (HR: 2.76, 95% CI: 1.12-6.17), NIHSS at admission (HR: 1.11, 95% CI: 1.02-1.18), and infarct subtypes (HR: 2.88, 95% CI: 1.34-6.17) were associated with recurrent stroke/TIA. Accordingly, a score (FENS score) was developed with AUC-ROC values of 0.863, 0.788, and 0.858 for 1, 3, and 5 years, respectively. These were significantly better than the AUC-ROC of ALM score (0.635, 0.695, and 0.705, respectively). The FENS score exhibited better calibration and discrimination ability than the ALM score (Hosmer-Lemeshow test χ2 : 4.402, p = 0.819). The MRI-based FENS score can provide excellent predictive performance for recurrent stroke/TIA and may assist in risk stratification of ESUS patients.