Objective To retrospectively analyze the risk factors for recurrence within 1 year after discontinuation of antiepileptic drugs (AEDs) in epilepsy patients, and to construct and validate a columnar diagram model for individualized prediction of recurrence risk. The aim is not only to provide decision-making assistance for clinicians but, more importantly, to offer a quantitative tool for nursing staff to identify high-risk patients, implement stratified management, and deliver precision interventions. Methods A retrospective study was conducted on 271 epilepsy patients treated at Beijing Tiantan Hospital, Capital Medical University, from October 2018 to October 2023, who ultimately attempted AED withdrawal. Patients were divided into a recurrence group (n=67) and a non-recurrence group (n=204) based on whether they experienced recurrence within 1 year after AED discontinuation. Key predictive factors were screened from candidate clinical variables using least absolute shrinkage and selection operator (LASSO) regression, followed by multivariate logistic regression analysis to identify independent risk factors. A columnar diagram predictive model was constructed based on these independent factors. Internal validation was performed using bootstrap repeated sampling (1000 times), and the model's discriminative power, calibration, and clinical utility were comprehensively evaluated by calculating the consistency index (C-index), plotting receiver operating characteristic (ROC) curves, calibration curves, and decision curve (DCA). Results LASSO regression and multivariate analysis ultimately identified three independent predictors: epileptiform discharges on electroencephalogram (EEG) (OR=3.15, 95%CI (1.72, 5.78)], presence of structural etiology [OR=2.89, 95%CI (1.51, 5.54)], and duration of epilepsy before drug withdrawal (≥5 years) OR=2.46, 95%CI (1.30, 4.66)]. The columnar diagram model constructed based on these predictors achieved a C-index of 0.786 [95%CI (0.728, 0.844)] in the original cohort. After internal validation, the adjusted C-index was 0.773. ROC curve analysis demonstrated an area under the curve (AUC) of 0.786. Calibration curves showed good consistency between predicted and observed probabilities, with a Hosmer-Lemeshow test P-value of 0.452. Decision curve analysis indicated that the model had clinical net benefit when applied within a threshold probability range of 2% to 58%, with the maximum net benefit observed at a threshold probability of approximately 30%. Conclusion This study established a columnar diagram model incorporating three readily available clinical indicators, enabling rapid calculation of individualized relapse risk during outpatient drug withdrawal assessments. The model identifies three core targets for nursing stratification (EEG discharges, structural etiology, and long disease duration), providing objective evidence for transitioning from "routine education" to "risk-based precision intervention" for nursing staff. This approach contributes to optimizing the entire withdrawal process management and enhancing patient safety.