基于现实世界研究中临床随访数据的两种动态预测建模方法的实证研究

杨锋; 陈欣; 尤东方; 黄丽红; 魏朝晖; 赵杨

doi:10.12025/j.issn.1008-6358.2021.20212060

基于现实世界研究中临床随访数据的两种动态预测建模方法的实证研究

An empirical study on the two dynamic risk prediction methods based on clinical follow-up data in real-world study

摘要

摘要:
目的针对现实世界研究（real-world study，RWS）中常见的具有纵向测量属性的动态观察指标，探讨界标法和联合建模法2种动态预测方法的应用价值。
方法基于358例某重症肺炎患者预后数据，分别采用界标法和联合建模法，基于R软件，对于第5天、第10天、第15天尚处于观察期的某重症肺炎患者，预测其未来的死亡风险。
结果 2种方法均能在各时间点预测个体未来发生结局事件的概率。第5天、第10天和第15天，利用界标法进行动态预测的AUC分别为81.64%、85.89%和82.15%；而联合建模法的AUC分别为81.11%、85.07%和72.09%。
结论在针对动态历史数据的现实世界研究中，可采用动态预测模型分析法，从而获得更为丰富的信息。

Abstract:
Objective To explore the applications of landmarking method and joint modeling for dynamic risk prediction in the analysis of longitudinal dataset obtained from real-world study.
Methods Based on the longitudinal prognosis data of 358 pneumonia patients, we used landmarking method and joint modeling respectively by software R, to estimate the probability of survival for the pneumonia patients who are under observation on day 5, day 10, and day 15.
Results Both of the two methods can make dynamic risk predictions on the probability of outcome in the future at different time points. On day 5, day 10, and day 15, the AUCs of the prediction using landmarking methods were 81.64%, 85.89%, and 82.15%, respectively, while those obtained from the joint modeling were 81.11%, 85.07%, and 72.09%, respectively.
Conclusions In the real-world study for dynamic historical data, dynamic prediction model analysis can be used to obtain more information.

HTML全文

参考文献(14)

施引文献

资源附件(0)