Esen Buket Öztürk, Ehrenstein Vera, Petersen Irene, Sørensen Henrik Toft, Pedersen Lars
Department of Clinical Epidemiology, Aarhus University Hospital, Aarhus, Denmark.
Research Department of Primary Care and Population Health, University College London, London, UK.
Int J Epidemiol. 2024 Feb 1;53(1). doi: 10.1093/ije/dyad179.
The sibling comparison analysis is used to deal with unmeasured confounding. It has previously been shown that in the presence of non-shared unmeasured confounding, the sibling comparison analysis may introduce substantial bias depending on the sharedness of the unmeasured confounder and the sharedness of the exposure. We aimed to improve the awareness of this challenge of the sibling comparison analysis.
First, we simulated sibling pairs with an exposure, a confounder and an outcome. We simulated sibling pairs with no effect of the exposure on the outcome and with positive confounding. For varying degrees of sharedness of the confounder and the exposure and for varying prevalence of the exposure, we calculated the sibling comparison odds ratio (OR). Second, we provided measures for sharedness of selected treatments based on Danish health data.
The confounded sibling comparison OR was visualized for varying degrees of sharedness of the confounder and the exposure and for varying prevalence of the exposure. The confounded sibling comparison OR was seen to increase with increasing sharedness of the exposure and the confounded sibling comparison OR decreased with an increasing prevalence of exposure. Measures for sharedness of treatments based on Danish health data showed that treatments of chronic diseases have the highest sharedness and treatments of non-chronic diseases have the lowest sharedness.
Researchers should be aware of the challenge regarding non-shared unmeasured confounding in the sibling comparison analysis, before applying the analysis in non-randomized studies. Otherwise, the sibling comparison analysis may lead to substantial bias.
同胞比较分析用于处理未测量的混杂因素。先前已经表明，在存在非共享的未测量混杂因素的情况下，同胞比较分析可能会根据未测量混杂因素的共享程度和暴露的共享程度引入实质性偏差。我们旨在提高对同胞比较分析这一挑战的认识。
首先，我们模拟了具有暴露因素、混杂因素和结局的同胞对。我们模拟了暴露对结局无影响且存在正向混杂的同胞对。对于混杂因素和暴露因素不同程度的共享以及暴露因素的不同患病率，我们计算了同胞比较比值比（OR）。其次，我们基于丹麦健康数据提供了所选治疗方法共享程度的测量指标。
对于混杂因素和暴露因素不同程度的共享以及暴露因素的不同患病率，对混杂的同胞比较OR进行了可视化展示。可以看出，混杂的同胞比较OR随着暴露共享程度的增加而增加，并且混杂的同胞比较OR随着暴露患病率的增加而降低。基于丹麦健康数据的治疗方法共享程度测量指标表明，慢性病治疗方法的共享程度最高，非慢性病治疗方法的共享程度最低。
在非随机研究中应用同胞比较分析之前，研究人员应意识到同胞比较分析中关于非共享未测量混杂因素的挑战。否则，同胞比较分析可能会导致实质性偏差。
