2026-10-12T13:51:46Z https://www.scidb.cn/oai

10.57760/sciencedb.j00253.05072 2026-05-29T16:45:44Z

2026-05-29 Comparison and application of Mendelian Randomization-based Mediation Analysis doi:10.57760/sciencedb.j00253.05072 en Objective To evaluate the performance of four Mendelian randomization-based (MR-based) mediation analysis methods, including two-step epigenetic MR (two step MR), multivariable MR (MVMR), mediation analysis framework through GWAS summary data integration with bias correction (MAGIC), and path specific effect MR (PSE-MR), through simulation and real-world validation. Method Simulations varied different direct effects, indirect effects, and pleiotropic effects. Performance was assessed using average causal effect estimates, standard deviation, root mean square error, type I error rate, statistical power, and 95% confidence interval (CI) coverage. The mediating role of smoking in the relationship between educational attainment and attention-deficit/hyperactivity disorder (ADHD) was explored. Results Without pleiotropy, MAGIC performed well, achieving 95% CI coverage rates of 94.0%~95.7% for direct and indirect effects with approximately 100% power. When pleiotropy affected mediator or outcome, MAGIC maintained 95% CI coverage rate of 94.5%~96.1% across all three effects. When pleiotropy affected both the mediator and outcome, PSE-MR demonstrated >95% power for direct effects. MAGIC revealed that smoking mediated 9.006% (3.381%, 14.632%) of the association between educational attainment and ADHD. Conclusion MAGIC is recommended for scenarios without pleiotropy or with mediator/outcome-specific pleiotropy. PSE-MR is preferred when pleiotropy affects both mediator and outcome. Improving educational attainment and reducing smoking may help lower the risk of ADHD. Mendelian Randomization; Mediation analysis; Causal inference; Pleiotropy Shijia.Wang Jingwen.Fan Yongli.Yang Xuezhong.Shi Nana.Wang Xiaocan.Jia PUBLIC https://creativecommons.org/publicdomain/zero/1.0/ dataset Science Data Bank