基于LASSO logistic回归模型的轻度认知障碍逆转预测模型

A Predictive Model for Reversion of Mild Cognitive Impairment based on Lasso Logistic Regression Model

目的探讨轻度认知障碍(mild cognitive impairment,MCI)患者逆转为认知正常(normal cognition,NC)的相关因素并建立预测模型。方法基于美国公共数据库NACC,数据包括社会人口学信息、体格检查、疾病史、认知功能、抑郁状况、精神症状和日常活动功能,构建LASSO logistic回归模型筛选自变量,通过十折交叉验证法选择模型中的最优调和系数λ;采用AIC和BIC与全变量logistic回归和逐步logistic回归进行比较,基于AUC、Brier评分和校准曲线分别评价预测模型的区分度和准确度,并绘制森林图和列线图。结果共纳入397例MCI患者,其中124例MCI患者逆转为NC,逆转率为31.23%。LASSO logistic回归模型(λ=0.044),纳入的自变量为年龄、BMI、高脂血症、维生素B12缺乏症、他人报告认知障碍、FAQ、MMSE、CDR和动物命名正确数;AIC=188.364,BIC=232.187,均低于全变量logistic回归(207.940/299.570)和逐步logistic回归(196.489/260.232);AUC、Brier评分和校准曲线均显示LASSO logistic回归模型的区分度和准确度更好。结论MCI患者逆转为NC受多个因素影响,应关注未患有高脂血症和维生素B12缺乏症、日常活动功能和认知功能较好的低龄MCI患者,对其进行健康管理干预和预防性护理,减少其未来疾病进展的风险。

Objective To identify characteristics of individuals with mild cognitive impairment(MCI)that are associated with reversion of MCI to normal cognition(NC),and to establish a prediction model for the reversion.Methods Based on the United States NACC public database,the data included sociodemographic information,physical examination,medical history,cognitive function,depressive status,psychiatric symptoms and daily activity function.A Lasso logistic regression model was used to select independent variables,where the penalty parameterλin the model selected by 10 fold cross validation.In addition,AIC and BIC were used to evaluate performance of Lasso logistic model,compared with full-variable logistic regression and stepwise logistic regression.AUC,Brier score and calibration curve were used to evaluate the discriminability and accuracy of the prediction model.The forest plot and the histogram were drawn.Results A total of 397 MCI were surveyed,among which 124 MCI were reversed to NC,with a reversion rate of 31.23%.Theλselected by cross validation was 0.044.Nine characteristics contributed significantly to the reversion prediction:age,BMI,hyperlipidemia,vitamin B12 deficiency,informant complaint,and FAQ,MMSE,CDR,and correct numbers of animals.The values of AIC and BIC were 188.364 and 232.187 respectively,which were lower than those of full-variable logistic regression(207.940/299.570)and stepwise logistic regression(196.489/260.232).AUC,Brier score and calibration curve all showed that Lasso logistic regression model had better discriminability and accuracy.Conclusion The reversion from MCI to NC is affected by multiple factors,and we should pay attention to the younger MCI patients without hyperlipidemia and vitamin B12 deficiency,and with good daily activity and cognitive function,so as to conduct health management intervention and preventive care to reduce the risk of future disease progression.