首页|校正弱工具变量偏倚的孟德尔随机化方法比较及其应用

校正弱工具变量偏倚的孟德尔随机化方法比较及其应用

扫码查看
目的 探究无工具变量可用或需评估弱工具变量偏倚对结果的影响时,为选择合适的两样本孟德尔随机化(two-sample Mendelian randomization,TwoSampleMR)方法提供建议.方法 分别在无多效性、均衡多效性、正向多效性模拟情形下,变换工具变量强度考察弱工具变量对修正权重的逆方差加权模型(inverse variance weighting with modified weights,MW-IVW)、稳健校正轮廓评分(robust adjusted profile score,RAPS)孟德尔随机化方法和基于混合正态分布的孟德尔随机化模型(MR mixture model,MR-Mix)3种方法的影响.正向多效性和弱工具变量同时存在时,模拟不同个数工具变量对MR-Mix的影响.MR-Mix为主分析方法,其余2方法作为敏感性分析,探究BMI、高密度脂蛋白(high-density lipoprotein,HDL)、低密度脂蛋白(low-density lipoprotein,LDL)、三酰甘油(triglyceride,TG)以t及总胆固醇(total cholesterol,TC)与血清尿酸之间的因果关联.结果 无多效性和均衡多效性情形下,MW-IVW表现最佳,MR-Mix表现最差.正向多效性情况下,MR-Mix表现最好,MW-IVW表现最差.BMI(β=0.280,P=0.003)和TG(β=0.370,P<0.001)是血清尿酸升高的危险因素,HDL(β=-0.250,P=0.002)是血清尿酸的保护因素.结论 在无多效性和均衡多效性情形下,MW-IVW有更好的统计学性能;但当正向多效性存在时,MR-Mix有更好的稳健性.BMI和TG为血清尿酸升高的危险因素.
Comparison and application of Mendelian randomization methods for correcting weak instrumental variable bias
Objective To provide suggestions to choose the appropriate two-sample Mendelian randomization methods when no instrumental variables are available or weak instrumental variable bias exist,Methods In the case of no pleiotropy,balanced pleiotropy,and directional pleiotropy,respectively,the impact of weak instrumental variables on each method was investigated by changing the intensity of instrumental variables.The study simulated different number of instrumental variables to access the impact on MR-Mix under the condition that both directional pleiotropic effects and weak instrumental variables existed.MR-Mix served as the primary analytical method,while the other two methods were employed as sensitivity analyses to explore the causal associations between BMI,HDL,LDL,TG,TC,and serum uric acid.Results Under scenarios of no pleiotropy and balanced pleiotropy,MW-IVW performed the best,while MR-Mix performed the worst.In the case of directional pleiotropic,MR-Mix exhibited the best performed,whereas MW-IVW performed the worst.BMI(β=0.280,P=0.003)and TG(β=0.370,P<0.001)were identified as risk factors for elevated serum uric acid.HDL(β=-0.250,P=0.002)was identified as a protective factor.Conclusions Under scenarios of no pleiotropy and balanced pleiotropy,MW-IVW demonstrates better statistical performance.However,in the presence of directional pleiotropy,MR-Mix exhibits superior robustness.BMI and TG are identified as risk factors for elevated serum uric acid.

Mendelian randomizaitonPleiotropyWeak instrumentsBody mass indexLipid traitsSerum urate

杨博然、彭刘庆、高雪、王菊平、王彤

展开 >

山西医科大学公共卫生学院卫生统计学教研室,太原 030001

煤炭环境致病与防治教育部重点实验室,太原 030001

孟德尔随机化 多效性 弱工具变量 体质指数 脂质 血清尿酸

国家自然科学基金国家自然科学基金山西省基础研究计划山西省基础研究计划

818727158210394920210302124186202103021223234

2024

中华疾病控制杂志
中华预防医学会 安徽医科大学

中华疾病控制杂志

CSTPCD北大核心
影响因子:1.862
ISSN:1674-3679
年,卷(期):2024.28(4)
  • 13