Bayesian Mediation Analysis for Neonatal Neurodevelopmental Outcomes in Pregnancy with Opioid Exposure

联系PD/PI：Wen，Xuerong 摘要 中介分析可用于量化母体阿片类药物暴露之间的直接和间接关系， 妊娠、短期不良新生儿结局和儿童长期神经发育结局。有 不同的方法来估计中介模型，贝叶斯方法提供了一些优势相比， 频率论对应者：包括先验信息的可能性，多水平模型时的计算可行性 结构，并在分析较小的样本量的数据更灵活。在进行最初资助的 项目（1 R15 HD 097588），我们观察到妊娠期阿片类药物暴露对多个短期 先天性畸形和儿童的长期神经发育结果。然而，目前尚不清楚是否 观察到的影响是直接或间接的，以及母体阿片类药物使用对新生儿发育的影响是否仍然存在 在调整了中介因素后，有限的数据来源限制了我们的知识，并创造了一个强大的 研究机会。在这个项目中，我们将调查母体使用阿片类药物对多个短期和短期的中介作用。 对儿童的长期不良妊娠结局。将使用两个链接的数据集来检查不同的结果， 提供了较长的随访时间，并验证了妊娠窗的估计。我们的发现来自于 获奖项目表明，孕妇使用处方阿片类药物与增加 先天性畸形和新生儿神经发育结果。要为调解进一步调整 因素，并确定怀孕期间阿片类药物暴露的直接影响。本项目的重点是研究 围产期阿片类药物暴露与儿童不良健康结局的关系我们假设处方 怀孕期间使用阿片类药物直接对儿童的长期发育构成风险， 孕妇的管理改善了儿童的长期健康结果。具体目标1：确定 在母体暴露于处方阿片类药物和新生儿 神经发育结果。这一目标的目的是确定与下列因素有重大关联的调解人： 母体阿片类药物暴露也与儿童长期神经发育结果密切相关。一是 拟合多变量logistic回归模型，以评估短期不良新生儿结局与 校正基线潜在混杂因素后的宫内阿片类药物暴露。二、参数与非参数 将评估长期神经发育结局和短期不良新生儿结局的相关性， 与子宫内阿片类药物暴露显著相关。第三，将根据所有选定的 调解员具体目标2：制定和估计长期新生儿的贝叶斯中介模型 神经发育结果首先，我们将制定贝叶斯中介分析模型。的规格 贝叶斯中介模型需要未知参数的先验分布推导，使用来自 以前的研究（如有），以及响应或中介变量的抽样分布。默认的选择 先验允许与频率论方法进行比较。其次，马尔可夫链蒙特卡罗方法（MCMC）将 用于估计贝叶斯模型。将获得所有感兴趣参数的后验分布，特别是 母亲在怀孕期间接触阿片类药物的影响。后验分布分析允许提取 关注的汇总统计量、后验均值、中位数和可信区间。将评估结果的耐用性， 不同的先验分布选择。 结果：这项工作将通过准确量化妊娠期阿片类药物的安全性， 围产期暴露于处方阿片类药物与儿童长期神经发育和教育状况之间的关系 孩子 项目总结/摘要

Contact PD/PI: Wen, Xuerong ABSTRACT Mediation analysis is useful to quantify the direct and indirect relationship among maternal opioid exposure during pregnancy, short-term adverse neonatal outcomes, and long-term neurodevelopmental outcomes in children. There are different approaches to estimate mediation models, and Bayesian methods offer some advantages compared to their frequentist counterparts: possibility of including prior information, computational feasibility when a multilevel model structure is needed, and more flexibility in analyzing data with smaller sample sizes. In conducting the originally funded project (1R15HD097588), we have observed significant effects for pregnancy opioid exposure on multiple short-term congenital malformations and long-term neurodevelopmental outcomes in children. However, it is unclear whether the observed effects are direct or indirect, and whether the effect of maternal opioid usage on neonatal development remains significant after adjusting for mediation factors. The limited data sources restrain our knowledge and creates a formidable research opportunity. In this project, we will investigate mediation effects of maternal opioid use on multiple short- and long-term adverse pregnancy outcomes for children. Two linked datasets will be used to examine different outcomes, provide a long follow-up time, and validate the estimation of pregnancy window. Our findings from the originally awarded project have shown that prescription opioid use in pregnant women is associated with increased risk of congenital malformations and neonatal neurodevelopmental outcomes. It is necessary to further adjust for the mediation factors and determine the direct effects of opioid exposure during pregnancy. The focus of this project is to study the relationship between perinatal opioid exposure and adverse health outcomes in children. We hypothesize that prescription opioid use during pregnancy directly poses risks on long-term development of children, and that optimized pain management in pregnant women improves long-term health outcomes for children. Specific Aim 1: To identify the mediation factors that are on the causal pathway of maternal exposure to prescription opioids and neonatal neurodevelopmental outcomes. The goal of this aim is to identify the mediators that are significantly associated with maternal opioid exposure and also strongly correlated with child long-term neurodevelopmental outcomes. First, we will fit a multivariable logistic regression model to assess the association between short-term adverse neonatal outcomes and in-utero opioid exposure after adjusting for baseline potential confounding factors. Second, parametric and non-parametric correlations will be assessed for long-term neurodevelopmental outcomes and short-term adverse neonatal outcomes that are significantly related to in-utero opioid exposure. Third, a mediating risk scores will be calculated based on all selected mediators. Specific Aim 2: To formulate and estimate a Bayesian mediation model for the long-term neonatal neurodevelopmental outcome First, we will formulate a Bayesian mediation analysis model. The specification of a Bayesian mediation model requires the prior distribution elicitation for the unknown parameters, using information from previous studies when available, and a sampling distribution for the response or mediation variable. The choice of default priors allows for comparison with frequentist methods. Second, Markov Chain Monte Carlo methods (MCMC) will be applied to estimate Bayesian models. Posterior distributions will be obtained for all the parameters of interest, in particular for the effect of maternal opioid exposure during pregnancy. Posterior distribution analysis allows for the extraction of summary statistics of interest, posterior means, medians, and credible intervals. Robustness of results will be assessed for different prior distribution choices. Outcome: This work will significantly impact the field of opioid safety in pregnancy by accurately quantifying the association between perinatal exposure to prescription opioids with long term neurodevelopment and educational status in children. Project Summary/Abstract