交叉滞后路径模型建模策略的模拟研究  

A Simulation Research of Modeling Strategy of Cross-Lagged Path Model

作　　者：于文浩 高超男 范冰冰 吕嘉丽 李春霞 贺力 张涛 Yu Wenhao;Gao Chaonan;Fan Bingbing(Department of Biostatistics,School of Public Health,Cheeloo College of Medicine,Shandong University,250012,Jinan)

机构地区：[1]山东大学齐鲁医学院公共卫生学院生物统计学系,250012 [2]国家健康医疗大数据研究院

出　　处：《中国卫生统计》2022年第5期669-674,共6页Chinese Journal of Health Statistics

基　　金：国家自然科学基金(81973147)。

摘　　要：目的通过模拟试验,研究在不同样本量、随访间隔,以及不同交叉滞后路径系数组合的情形下,交叉滞后路径模型的表现。通过实例分析,比较不同建模情形下的模型结果。方法首先基于两断面交叉滞后路径模型生成模拟数据集,然后在总数据集中随机抽样进行模型估计。通过遍历不同样本量及自相关系数,比较模型估计的Ⅰ类错误、检验效能、偏倚、标准误和均方误差。基于美国全国女性健康研究,使用两种建模策略探究身体质量指数(BMI)与血压之间的时序关系。结果当样本量达到500后,模型拟合趋于稳定。当纳入到模型中样本的随访间隔不一致时,模型的检验效能未受到明显影响,但模型估计的Ⅰ类错误率达到0.12~0.15,标准误和均方误差也有所增大。当随访间隔一致时,自相关系数的大小对模型估计的误差影响较小,而模型估计的偏倚在自相关系数偏大或偏小时均有增大趋势。当两变量的自相关系数不相等时,模型的估计基本不受影响。实例分析结果显示,当纳入的研究对象随访间隔不一致时,BMI与血压之间为双向时序关系;而当纳入随访间隔一致的研究对象时,BMI与血压之间表现出BMI→血压的单向时序关系。结论在交叉滞后路径模型建模过程中,样本量以及随访断面的定义都会对模型的统计分析结果产生影响。建议在纳入研究人群时,尽量保持随访断面整齐或人群随访间隔相似。Objective To study the performance of the cross-lagged path model under different sample sizes,follow-up intervals,and different combinations of cross-lagged path coefficients through simulation experiments.To compare the results of model under different modeling scenarios by cohort analysis.Methods Firstly,simulated dataset was generated based on the two-wave cross-lagged path model,and then a subset was sampled from the total dataset randomly for model estimation.By traversing different sample sizes and autocorrelation coefficients,we compare typeⅠerror,power,bias,standard error,and mean square error of the estimated models.Based on The Study of Women′s Health Across the Nation,two modeling strategies were used to explore the temporal relationship between body mass index(BMI)and blood pressure.Results When the sample size reached 500,the model fitting tended to be stable.When the follow-up intervals included were inconsistent,the power of the model was not significantly affected,but the type I error reached 0.12~0.15,the standard error and mean square error were also increased.When the follow-up intervals were consistent,the autocorrelation coefficients had little effect on the standard error and mean square error of the model estimation,but the bias tended to increase when the autocorrelation coefficients were too large or too small.When the autocorrelation coefficients of the two variables were not equal,the estimation of the model was basically unaffected.As for the cohort analysis,there was a bidirectional temporal relationship between BMI and blood pressure when the follow-up intervals were inconsistent.However,when the follow-up intervals included were consistent,there was a unidirectional temporal relationship between BMI and blood pressure(BMI→blood pressure).Conclusion In the process of modeling the cross-lagged path model,sample sizes and the definition of follow-up will affect the results of the model.It is recommended to keep the follow-up waves consistent or the follow-up intervals similar.

关 键 词：交叉滞后路径模型 样本量 随访间隔 Ⅰ类错误 

分 类 号：R195.1[医药卫生—卫生统计学]