Department of Biostatistics, The University of Texas MD Anderson Cancer Center, Houston, Texas, United States of America.
Department of Data Science, The University of Mississippi Medical Center, Jackson, Mississippi, United States of America.
PLoS One. 2024 Mar 8;19(3):e0293510. doi: 10.1371/journal.pone.0293510. eCollection 2024.
Mendelian randomization (MR) is an epidemiological framework using genetic variants as instrumental variables (IVs) to examine the causal effect of exposures on outcomes. Statistical methods based on unidirectional MR (UMR) are widely used to estimate the causal effects of exposures on outcomes in observational studies. To estimate the bidirectional causal effects between two phenotypes, investigators have naively applied UMR methods separately in each direction. However, bidirectional causal effects between two phenotypes create a feedback loop that biases the estimation when UMR methods are naively applied. To overcome this limitation, we proposed two novel approaches to estimate bidirectional causal effects using MR: BiRatio and BiLIML, which are extensions of the standard ratio, and limited information maximum likelihood (LIML) methods, respectively. We compared the performance of the two proposed methods with the naive application of UMR methods through extensive simulations of several scenarios involving varying numbers of strong and weak IVs. Our simulation results showed that when multiple strong IVs are used, the proposed methods provided accurate bidirectional causal effect estimation in terms of median absolute bias and relative median absolute bias. Furthermore, compared to the BiRatio method, the BiLIML method provided a more accurate estimation of causal effects when weak IVs were used. Therefore, based on our simulations, we concluded that the BiLIML should be used for bidirectional causal effect estimation. We applied the proposed methods to investigate the potential bidirectional relationship between obesity and diabetes using the data from the Multi-Ethnic Study of Atherosclerosis cohort. We used body mass index (BMI) and fasting glucose (FG) as measures of obesity and type 2 diabetes, respectively. Our results from the BiLIML method revealed the bidirectional causal relationship between BMI and FG in across all racial populations. Specifically, in the White/Caucasian population, a 1 kg/m2 increase in BMI increased FG by 0.70 mg/dL (95% confidence interval [CI]: 0.3517-1.0489; p = 8.43×10-5), and 1 mg/dL increase in FG increased BMI by 0.10 kg/m2 (95% CI: 0.0441-0.1640; p = 6.79×10-4). Our study provides novel findings and quantifies the effect sizes of the bidirectional causal relationship between BMI and FG. However, further studies are needed to understand the biological and functional mechanisms underlying the bidirectional pathway.
孟德尔随机化（MR）是一种使用遗传变异作为工具变量（IVs）来检验暴露对结局的因果效应的流行病学框架。基于单向 MR（UMR）的统计方法广泛用于在观察性研究中估计暴露对结局的因果效应。为了估计两个表型之间的双向因果效应，研究人员天真地分别在每个方向应用 UMR 方法。然而，两个表型之间的双向因果效应会产生反馈回路，当天真地应用 UMR 方法时，会导致估计偏倚。为了克服这一限制，我们提出了两种使用 MR 估计双向因果效应的新方法：BiRatio 和 BiLIML，它们分别是标准比值和有限信息最大似然（LIML）方法的扩展。我们通过对涉及不同数量强和弱 IVs 的几种情况的广泛模拟比较了这两种新方法与天真地应用 UMR 方法的性能。我们的模拟结果表明，当使用多个强 IVs 时，所提出的方法在中位数绝对偏差和相对中位数绝对偏差方面提供了准确的双向因果效应估计。此外，与 BiRatio 方法相比，当使用弱 IVs 时，BiLIML 方法提供了更准确的因果效应估计。因此，基于我们的模拟，我们得出结论，BiLIML 应该用于双向因果效应估计。我们应用所提出的方法来研究动脉粥样硬化多民族研究队列中肥胖和糖尿病之间潜在的双向关系。我们分别使用体重指数（BMI）和空腹血糖（FG）作为肥胖和 2 型糖尿病的衡量标准。我们从 BiLIML 方法的结果中揭示了所有种族人群中 BMI 和 FG 之间的双向因果关系。具体来说，在白种人/高加索人群中，BMI 每增加 1kg/m2，FG 增加 0.70mg/dL（95%置信区间[CI]：0.3517-1.0489；p=8.43×10-5），FG 每增加 1mg/dL，BMI 增加 0.10kg/m2（95%CI：0.0441-0.1640；p=6.79×10-4）。我们的研究提供了新的发现，并量化了 BMI 和 FG 之间双向因果关系的效应大小。然而，需要进一步的研究来了解双向通路背后的生物学和功能机制。
