Statistical Approaches for Estimating Sex-Specific Effects in Endocrine Disruptors Research.
Statistical Approaches for Estimating Sex-Specific Effects in Endocrine Disruptors Research.
复制标题
DOI：
10.1289/ehp334
复制
发表时间：
2017-06-23
影响因子：
10.4
通讯作者：
Engel SM
中科院分区：
文献类型：
--
作者：
Buckley JP;Doherty BT;Keil AP;Engel SM
当预期感兴趣的生物机制根据性别而有不同的运作时（内分泌干扰物研究中经常出现这种情况），研究人员通常会估计性别特异性关联。人们很少关注混杂因素与结果之间潜在的性别异质性。当协变量与结果的关系因性别而异时，用于估计性别特异性内分泌干扰物效应的常用统计方法可能会产生不同的估计。我们讨论了基本假设并评估了两种用于估计特定性别效应、分层和乘积项的传统方法的性能，并引入了一种简单的建模替代方案：增强乘积项方法。我们描述了关于混杂关系的性别异质性的假设对三种方法的兴趣暴露的性别特异性影响的估计的影响：分层、传统产品术语和增强产品术语。通过模拟和应用示例，我们展示了每种方法在一系列场景下的属性。在模拟中，当混杂因素与结果存在性别依赖性关联时，使用传统乘积项方法估计的特定性别暴露效应会出现偏差。通过分层和增强乘积项方法进行的特定性别估计是公正的，但不太精确。在应用示例中，这三种方法产生了相似的估计，但根据统计显着性得出了一些有意义的结论差异。研究人员在选择分析方法来估计内分泌干扰物对健康的性别特异性影响时，应考虑混杂因素的性别异质性。在存在性别依赖性混杂的情况下，当有先验知识可用于拟合简化模型或当研究人员寻求效应测量修改的自动测试时，我们的增强乘积项方法可能比分层更有优势。 https://doi.org/10.1289/EHP334
When a biologic mechanism of interest is anticipated to operate differentially according to sex, as is often the case in endocrine disruptors research, investigators routinely estimate sex-specific associations. Less attention has been given to potential sexual heterogeneity of confounder associations with outcomes. When relationships of covariates with outcomes differ according to sex, commonly applied statistical approaches for estimating sex-specific endocrine disruptor effects may produce divergent estimates. We discuss underlying assumptions and evaluate the performance of two traditional approaches for estimating sex-specific effects, stratification and product terms, and introduce a simple modeling alternative: an augmented product term approach. We describe the impact of assumptions regarding sexual heterogeneity of confounder relationships on estimates of sex-specific effects of the exposure of interest for three approaches: stratification, traditional product terms, and augmented product terms. Using simulated and applied examples, we demonstrate properties of each approach under a range of scenarios. In simulations, sex-specific exposure effects estimated using the traditional product term approach were biased when confounders had sex-dependent associations with the outcome. Sex-specific estimates from stratification and the augmented product term approach were unbiased but less precise. In the applied example, the three approaches yielded similar estimates, but resulted in some meaningful differences in conclusions based on statistical significance. Investigators should consider sexual heterogeneity of confounder associations when choosing an analytic approach to estimate sex-specific effects of endocrine disruptors on health. In the presence of sex-dependent confounding, our augmented product term approach may be advantageous over stratification when there is prior knowledge available to fit reduced models or when investigators seek an automated test for effect measure modification. https://doi.org/10.1289/EHP334