中风队列研究中多重插补法拟合量表缺失数据效果评价

Evaluation of the Effects of Multiple Imputation Method in Fitting Missing Data of Stroke Cohort Study Scale

目的评价多重插补法拟合中风队列研究量表缺失数据的效果,为今后开展相关临床研究提供方法学支撑。方法选取2017年1月-2020年12月陕西省5所三级甲等医院实施的多中心、前瞻性队列研究中的400例中风患者数据。应用R4.0.1软件分析美国国立卫生研究院卒中量表(NIHSS)、日常生活能力量表(ADL)、Fugl-Meyer运动功能评分量表(FMAS)、汉密顿抑郁量表(HAMD)和汉密顿焦虑量表(HAMA)的数据缺失特征;选取最优多重填补方法,采用SAS 9.4拟合缺失数据;通过标准误和95%置信区间(CI)宽度比较多重插补法与删除法的数据结果,评价多重插补法的填补效果。结果所有量表的数据缺失比例均小于20%,均为任意缺失模式,其中NIHSS为完全随机缺失机制,其余量表均为随机缺失机制。故选取马尔科夫链蒙特卡罗法(MCMC)作为最优多重填补方法。相较于删除法,MCMC产生更小的标准误及更窄的95%CI宽度,可更有效利用其他信息预测缺失数据并提高数据的利用率。结论通过应用多重填补的方法处理量表数据缺失后可有效提高中医综合方案治疗中风队列研究数据的利用率,减少浪费,使统计结果最大程度接近真实测量情况,同时提高研究数据完整性,进一步提升数据质量。

Objective To evaluate the effects of multiple imputation method on fitting missing data of scales of stroke cohort studies;To provide methodological support for future clinical studies.Methods The data of 400 stroke patients in a multicenter,prospective cohort study from January 2017 to December 2020 in five tertiary hospitals in Shannxi Province were selected.R 4.0.1 was used to analyze the data missing characteristics of NIHSS,ADL,FMAS,HAMD,and HAMA.The optimal multiple imputation method was selected and SAS 9.4 was used to fit the missing data.The data results of multiple imputation method and deletion method were compared by standard error and the width of 95% confidence interval(CI)to evaluate the filling effect of multiple imputation method.Results The percentage of missing data on all scales was less than 20% and all in arbitrary missing patterns.The NIHSS was based on the missing completely at random,and the remaining scales were based on the missing at random.Therefore,the MCMC was selected as the optimal multiple filling method.Compared with the deletion method,the MCMC produced a smaller standard error and a narrower 95% CI width,and could more effectively use other information to predict missing data and improve data utilization.Conclusion Application of multiple imputation method to deal with the missing scale data can effectively improve the utilization rate of cohort study data of TCM comprehensive treatment for stroke,reduce waste,make the statistical results approach the real measurement to the maximum extent,improve the completion of the data in the study,and further improve the data quality.