A simulation study for handling two-way treatment switching in rare event scenarios
10.3760/cma.j.cn112338-20240522-00295
- VernacularTitle:罕见事件场景下处理双向治疗转换方法的模拟研究
- Author:
Wenkai WU
1
;
Qiao HE
;
Minghong YAO
;
Jiayue XU
;
Wen WANG
;
Xin SUN
Author Information
1. 四川大学华西公共卫生学院，成都 610041
- Publication Type:Journal Article
- Keywords:
Rare event;
Treatment switching;
Simulation study;
Switching rate;
Treatment effect
- From:
Chinese Journal of Epidemiology
2025;46(2):334-344
- CountryChina
- Language:Chinese
-
Abstract:
Objective:Drug safety assessments based on real-world data are often challenged by both treatment switching and rare events. In this study, we used statistical simulations to investigate the effects of switching rates and treatment effects on the statistical performance of commonly used analytical strategies and methods under overlapping scenarios of treatment switching and rare events.Methods:The simulation scenario was set up as a bidirectional treatment switching (allowing the control group to switch to the treatment group and the treatment group to switch to the control group), and the event rates were set at approximately 2%, 5%, and 20%. Different simulation scenarios were generated with sufficient sample size to consider switching rate and relative treatment effect. The simulated datasets were analyzed using three types of analysis strategy, i.e. intention to treat (ITT), per protocol (PP), and as treated (AT). The performance of five indicators, namely percentage bias, mean square error, empirical standard error, coverage, and rejection rate, were compared among the different methods in different scenarios, and recommendations for method selection were given.Results:In terms of analytical strategies and methods, AT analysis were relatively optimal in terms of percentage bias and accuracy, followed by PP analysis and ITT analysis. When the relative treatment effects converged (e.g. HR=1.0), both the ITT analysis and the time-dependent AT approaches (marginal structural model, time-dependent Cox regression model or time-dependent propensity score matching) performed well; when the relative treatment effects were small (e.g. HR=0.8), the marginal structural model was the most optimal; when the relative treatment effects were large (e.g. HR=0.6 or 0.4), the approaches of using a censored treatment for switchers in the AT analysis were more accurate. In addition, the time-dependent AT approaches had the highest rejection rate when there was a difference in treatment effect between the two groups, and the ITT analysis had the lowest rejection rate. Conclusions:For the dual challenges of bidirectional switching and rare events in real-world drug safety evaluations, adequate sample size is a prerequisite for accurate estimation of treatment effects, while switching rates and effect sizes of switched drugs might also affect estimation accuracy. Appropriate strategies and methods should be selected for the analysis. It is necessary to consider whether the event is rare or not, the switching rate and the expected treatment effect size of the two types of treatment to select appropriate analysis strategies and methods.
