Background: In atrial fibrillation (AF) patients, breakthrough stroke is not uncommon and represents an important subgroup due to its high stroke recurrence rate and mortality. However, no reliable tool exists for assessing stroke risk in anticoagulated paroxysmal AF (PAF) patients. While LOXL2 is implicated in atrial fibrosis, a key pathological substrate for atrial thrombus, its predictive value for stroke remains unclear. Objective: To investigate predictive value of serum LOXL2 levels for breakthrough stroke risk in PAF patients. Methods: We consecutively enrolled 197 anticoagulated PAF patients. The serum level of LOXL2 were quantified via ELISA. Patients were stratified into LOXL2+ and LOXL2− groups based on the median of their baseline LOXL2 (275.9 pg/mL). Stroke events were recorded over a median follow-up of 3.9 years. Predictive models incorporating LOXL2, age, and CHA2DS2-VASc were evaluated using ROC, NRI, and DCA. Results：During follow-up, 24 patients (12.2%) experienced stroke. LOXL2 levels were significantly higher in stroke cases (P = 0.006). Multivariable Cox analysis identified LOXL2 as an independent risk factor (P < 0.001). Kaplan-Meier and Nelson-Aalen cumulative hazard analyses further confirmed the contribution of LOXL2 for elevated stroke risk. A prediction model, incorporating both LOXL2 and age, achieved the highest predictive accuracy (AUC = 0.842), significantly improving risk stratification over CHA2DS2-VASc (NRI = 15.0%, P < 0.001). Conclusion: Elevated LOXL2 is independently associated with breakthrough stroke risk in PAF patients. Incorporating LOXL2 and age enhances predictive accuracy, offering a novel tool for personalized stroke risk stratification in AF patients despite anticoagulation medication.