Stroke and bleeding risks in atrial fibrillation (AF) are often assessed at baseline to predict outcomes years later. We investigated whether dynamic changes in CHA2DS2-VASc and HAS-BLED scores over time modify risk prediction.We included patients with AF who were stable while taking vitamin K antagonists. During a 6-year follow-up, all ischemic strokes/transient ischemic attacks (TIAs) and major bleeding events were recorded. CHA2DS2-VASc and HAS-BLED were recalculated every 2-years and tested for clinical outcomes at 2-year periods.We included 1361 patients (mean CHA2DS2-VASc and HAS-BLED 4.0±1.7 and 2.9±1.2). During the follow-up, 156 (11.5%) patients had an ischemic stroke/TIA and 269 (19.8%) had a major bleeding event. Compared with the baseline CHA2DS2-VASc, the CHA2DS2-VASc recalculated at 2 years had higher predictive ability for ischemic stroke/TIA during the period from 2 to 4 years. Integrated discrimination improvement (IDI) and net reclassification improvement (NRI) showed improvements in sensitivity and better reclassification. The CHA2DS2-VASc recalculated at 4 years had better predictive performance than the baseline CHA2DS2-VASc during the period from 4 to 6 years, with an improvement in IDI and an enhancement of the reclassification. The recalculated HAS-BLED at 2-years had higher predictive ability than the baseline score for major bleeding during the period from 2 to 4 years, with significant improvements in sensitivity and reclassification. A slight enhancement in sensitivity was observed with the HAS-BLED score recalculated at 4 years compared with the baseline score.In AF patients, stroke and bleeding risks are dynamic and change over time. The CHA2DS2-VASc and HAS-BLED scores should be regularly reassessed, particularly for accurate stroke risk prediction.