Centre for Biostatistics, School of Health Sciences, The University of Manchester, Oxford Road, Manchester, M13 9PL, UK.
BMC Med Res Methodol. 2020 Dec 7;20(1):295. doi: 10.1186/s12874-020-01170-0.
Mendelian randomization (MR) has been widely applied to causal inference in medical research. It uses genetic variants as instrumental variables (IVs) to investigate putative causal relationship between an exposure and an outcome. Traditional MR methods have mainly focussed on a two-sample setting in which IV-exposure association study and IV-outcome association study are independent. However, it is not uncommon that participants from the two studies fully overlap (one-sample) or partly overlap (overlapping-sample).
We proposed a Bayesian method that is applicable to all the three sample settings. In essence, we converted a two- or overlapping- sample MR to a one-sample MR where data were partly unmeasured. Assume that all study individuals were drawn from the same population and unmeasured data were missing at random. Then the missing data were treated au pair with the model parameters as unknown quantities, and thus, were imputed iteratively conditioning on the observed data and estimated parameters using Markov chain Monte Carlo. We generalised our model to allow for pleiotropy and multiple exposures and assessed its performance by a number of simulations using four metrics: mean, standard deviation, coverage and power. We also compared our method with classic MR methods.
In our proposed method, higher sample overlapping rate and instrument strength led to more precise estimated causal effects with higher power. Pleiotropy had a notably negative impact on the estimates. Nevertheless, the coverages were high and our model performed well in all the sample settings overall. In comparison with classic MR, our method provided estimates with higher precision. When the true causal effects were non-zero, power of their estimates was consistently higher from our method. The performance of our method was similar to classic MR in terms of coverage.
Our model offers the flexibility of being applicable to any of the sample settings. It is an important addition to the MR literature which has restricted to one- or two- sample scenarios. Given the nature of Bayesian inference, it can be easily extended to more complex MR analysis in medical research.
孟德尔随机化（MR）已广泛应用于医学研究中的因果推断。它使用遗传变异作为工具变量（IVs）来研究暴露与结局之间的潜在因果关系。传统的 MR 方法主要集中在两样本设置中，其中 IV-暴露关联研究和 IV-结局关联研究是独立的。然而，参与者完全重叠（单样本）或部分重叠（重叠样本）的情况并不少见。
我们提出了一种适用于所有三种样本设置的贝叶斯方法。本质上，我们将两样本或重叠样本 MR 转换为部分未测量数据的单样本 MR。假设所有研究个体均来自同一人群，未测量数据是随机缺失的。然后，将缺失数据与模型参数视为未知量进行配对处理，并使用马尔可夫链蒙特卡罗方法对观察数据和估计参数进行迭代条件推断。我们将模型推广到允许多效性和多个暴露的情况，并使用四种指标（平均值、标准差、覆盖率和功效）通过多项模拟来评估其性能。我们还将我们的方法与经典 MR 方法进行了比较。
在我们的方法中，更高的样本重叠率和工具强度导致了更高的功效和更精确的因果效应估计。多效性对估计值有显著的负面影响。尽管如此，覆盖范围很高，我们的模型在所有样本设置中总体表现良好。与经典 MR 相比，我们的方法提供了更精确的估计值。当真实的因果效应不为零时，我们的方法估计值的功效始终更高。在覆盖范围方面，我们的方法与经典 MR 相似。
我们的模型提供了适用于任何样本设置的灵活性。它是 MR 文献的一个重要补充，该文献仅限于单样本或两样本情况。鉴于贝叶斯推断的性质，它可以很容易地扩展到医学研究中更复杂的 MR 分析。
