基于异质性检验方法的统计模拟研究
Statistical Simulation Study Based on Heterogeneity Test Methods

目的：在流行病学研究中，分层因素可能会影响研究因素与结局变量之间的关系。因此异质性检验在分析暴露与结局的关联性时尤为重要，以确保研究结果的科学性。常用的异质性检验方法包括Breslow-Day法、Tarone法和Logistic回归分析法，但它们在不同分层数据下的I类错误控制和检验效能表现存在差异。本研究旨在通过蒙特卡洛模拟实验评估Breslow-Day法、Tarone法和Logistic回归法在不同样本量和分层因素下的表现，以找出在分层异质性检验中最稳健的统计方法，为流行病学研究提供参考。方法：利用R软件生成模拟数据，分别针对单分层(性别)和双分层(性别和年龄)条件下多次模拟不同样本量的数据，评估各方法在一类错误和检验效能方面的表现。结果：在单分层情况下，Breslow-Day和Tarone法在小样本时的I类错误控制较弱，随着样本量增大逐渐稳定；Logistic回归法在小样本条件下控制能力较好，表现更为稳健。在双分层情况下，Breslow-Day和Tarone法不再适用，Logistic回归法在校正后可有效控制I类错误。结论：本研究揭示了不同异质性检验方法的适用条件和表现差异，为流行病学和分层分析的实践提供了数据支持。Breslow-Day和Tarone法适用于简单分层和小样本条件，而Logistic回归法在大样本和多分层条件下表现更优，需适当校正以控制误差。这为后续研究和应用提供了更为科学的异质性检验方法选择依据。
Objective: In epidemiological research, stratification factors may influence the relationship between study factors and outcome variables. Therefore, heterogeneity testing is crucial for analyzing the association between exposure and outcomes to ensure the scientific validity of research findings. Common heterogeneity test methods include the Breslow-Day test, Tarone’s test, and Logistic regression analysis, but their performance in terms of Type I error control and test power varies under different stratified data conditions. This study aims to evaluate the performance of the Breslow-Day test, Tarone’s test, and Logistic regression method under varying sample sizes and stratification factors through Monte Carlo simulation experiments, to identify the most robust statistical method for stratified heterogeneity testing, and provide references for epidemiological research. Methods: Simulated data were generated using R software under single-stratification (gender) and double-stratification (gender and age) conditions. Multiple datasets with varying sample sizes were simulated to evaluate the performance of each method in terms of Type I error control and test power. Results: Under single-stratification, the Breslow-Day and Tarone tests showed weak Type I error control for small samples but stabilized as the sample size increased. The Logistic regression method demonstrated better robustness and control in small sample conditions. Under double-stratification, the Breslow-Day and Tarone tests became inapplicable, while the Logistic regression method, after adjustment, effectively controlled Type I errors. Conclusion: This study highlights the applicability and performance differences of various heterogeneity testing methods, providing data-driven support for practice in epidemiology and stratified