真实世界研究中协变量组间均衡性的诊断指标研究

A Research on Metrics for Covariate Balance in Real World Study

目的评价真实世界研究(real world study,RWS)组间协变量均衡性的诊断指标。方法模拟不同的组间均衡性程度、不同的协变量与暴露、结局关系等RWS模拟数据场景,通过构建各诊断指标与估计偏差的相关性模型,评价不同的单一协变量、全局协变量均衡性诊断指标的准确性、稳健性。结果除L1测度外,标准化差值法、重叠系数、K-S距离、Lévy距离、马氏距离和一般加权差均能识别不同程度的均衡性。基于倾向得分的C统计量和一般加权差估计相关性模型的R2值均大于0.8,截距值逼近原点,对于组间均衡性的诊断最为准确和稳定。结论单一协变量诊断指标可以评估RWS数据组间协变量的均衡性,但全局诊断指标的准确性、灵敏度和稳健性更好,其中倾向得分C统计量的诊断效果最佳。

Objective To compare the diagnostic index which used to evaluate the balance of covariates between groups in real world study(RWS).Methods Simulate RWS simulation data scenarios such as different inter group equilibrium degree,different covariate and exposure,outcome relationship,etc.by constructing the correlation model between each diagnostic index and estimation deviation,evaluate the accuracy and robustness of different single covariate and global covariate diagnostic indexes.Results In addition to L1 measure,standardized difference,overlap coefficient,K-S distance,Lévy distance,Mahalanobis distance and general weighted difference can distinguish different degrees of inter group covariate equilibrium.The R 2 of the correlation models estimated by the C-statistic based on propensity score and the general weighted difference diagnostic index is greater than 0.8,and the intercept value approaches the origin,which is the most accurate and stable for the diagnosis of inter group equilibrium.Conclusion The single covariate diagnosis index can evaluate the balance of covariates between groups of RWS data,but the accuracy,sensitivity and robustness of the global diagnosis index are better.Among them,the diagnosis effect of L1 measure is poor,while of the C-statistic based on propensity score is the best.