中国实用神经疾病杂志2025，Vol.28Issue(12)：1453-1459,7.DOI:10.12083/SYSJ.250688
基于机器学习构建视神经脊髓炎谱系疾病复发预测模型
Building of relapse prediction model for neuromyelitis optica spectrum diseases based on machine learning
摘要
Abstract
Objective To summarize the clinical characteristics of neuromyelitis optica spectrum disorder(NMOSD)patients,and develop and evaluate a machine learning-based clinical prediction model for NMOSD relapse.Methods This is a single-center cohort retrospective study.NMOSD patients were recruited from the Department of Neurology,Beijing Tongren Hospital,Capital Medical University between December 1,2017 and December 1,2019 forming the initial cohort.Clinical data from enrollment until the first relapse were collected.Clinical characteristics and relapse risk factors were analyzed.Statistically significant clinical indicators differentiating the relapse group from the non-relapse group were selected as candidate predictors.Additionally,NMOSD patients admitted between December 2,2019 and December 25,2020 were enrolled as an independent test set cohort.A random survival forest model was used to develop and validate the relapse prediction model.Results Totally 145 NMOSD patients were identified,with a median follow-up time of 718 days.There were 73(50.3%)relapse cases and 72(49.7%)non-relapse cases.Analysis revealed several potential predictors:MOG-IgG serostatus,baseline EDSS score,sensory score,ANA antibody positivity,optic neuritis at enrollment,non-optic nerve-spinal type at enrollment,irregular preventive immunotherapy,and rituximab as preventive immunotherapy.The initial cohort was randomly split into training sets(n=73)and validation sets(n=72),and 36 NMOSD patients were enrolled in the test set.The model demonstrated good discrimination:training set AUC(95%CI)=0.880(0.806-0.954),validation set AUC=0.833(0.737-0.930),and test set AUC=0.706(0.525-0.887).It also showed good calibration.Baseline EDSS score,irregular preventive immunotherapy,and sensory score were the top three significant predictors in the model.Conclusion The random survival forest-based prediction model for NMOSD relapse exhibits good discrimination and calibration.Baseline EDSS score,irregular preventive immunotherapy,and sensory score are identified as the most significant risk factors.关键词
视神经脊髓炎谱系疾病/临床特征/复发/临床预测模型/机器学习/随机生存森林/预测效能Key words
Neuromyelitis optica spectrum disease/Clinical characteristics/Relapse/Clinical prediction model/Machine learning/Random survival forest/Predictive effectiveness分类
医药卫生引用本文复制引用
颜榕,郭燕军,王佳伟..基于机器学习构建视神经脊髓炎谱系疾病复发预测模型[J].中国实用神经疾病杂志,2025,28(12):1453-1459,7.基金项目
国家自然科学基金项目(编号:82271384) （编号:82271384）
