【摘要】 目的 基于轻度认知功能障碍(mild cognitive impairment, MCI)患者的基线临床资料和纵向神经心理学评分，构建阿尔茨海默病(Alzheimer’s disease, AD)发病风险的动态预测模型。方法 选取2005—2011年阿尔茨海默病神经影像学倡议1研究中的380例MCI患者作为研究对象，以7∶3随机将研究对象划分为训练集和测试集。以阿尔茨海默病评估量表-13项认知分量表(ADAS-Cog13)、Rey听觉词语学习测验即时分数(Rey auditory verbal learning test immediate score, RAVLT Immediate)、社会功能活动问卷(functional activities questionnaire, FAQ)和简易精神状态检查表(mini-mental state examination, MMSE)作为纵向神经心理学评分指标，利用具有多元纵向内生协变量的随机生存森林方法于训练集中构建MCI患者AD发病风险的动态预测模型。采用时间依赖的受试者工作特征曲线下面积(areas under the receiver operator characteristic curve, AUC)和布莱尔分数(Brier score, BS)于测试集中评估模型的预测性能。结果 对于MCI患者AD发病风险的预测，纵向神经心理学评分是比基线临床资料更为重要的预测因子，其中FAQ是最强的预测因子。动态预测模型的预测性能较高，在测试集中的AUC范围为0.7695～0.8987,BS的范围为0.1369～0.2184。结论 基于具有多元纵向内生协变量的随机生存森林方法可以合并多个纵向神经心理学评分构建MCI患者AD发病风险的预测模型，具有较高的预测性能，并能够实现个体化的动态预测。更多还原

【Abstract】 Objective To construct a dynamic prediction model for the risk of developing Alzheimer’s disease(AD) based on baseline clinical data and longitudinal neuropsychological scores in patients with mild cognitive impairment(MCI). Methods A total of 380 MCI patients from the Alzheimer’s Disease Neuroimaging Initiative 1 study from 2005 to 2011 were selected and were randomly divided into a training set and a test set in a 7∶3 randomization. The Alzheimer’s disease assessment scale-cognitive 13 items(ADAS-Cog13), Rey auditory verbal learning test immediate score(RAVLT Immediate), functional activities questionnaire(FAQ), and mini-mental state examination(MMSE) were used as longitudinal neuropsychological score metrics. Random survival forests with multivariate longitudinal endogenous covariates were used to construct a dynamic prediction model of the risk of developing AD in patients with MCI in the training set. The predictive performance of the model was evaluated in the test set using time-dependent areas under the receiver operator characteristic curve(AUC) and Brier score(BS). Results For the prediction of the risk of developing AD in patients with MCI, longitudinal neuropsychological scores were more important predictors than baseline clinical data, with FAQ being the strongest predictor. The dynamic prediction model had high predictive performance in the test set, with AUC ranging from 0.7695 to 0.8987 and BS ranging from 0.1369 to 0.2184. Conclusion Random survival forests with multivariate longitudinal endogenous covariates can be used to combine multivariate longitudinal neuropsychological scores to construct a prediction model for the risk of developing AD in patients with MCI, with high predictive performance and the ability to achieve individual dynamic prediction.更多还原