Application of directed acyclic graphs in identification and control of selection bias in causal inference
-
摘要: 在流行病学研究中，选择偏倚会导致研究样本无法代表一般人群，使研究结果偏离真实值，无法推断真实的因果关联。本文通过构建有向无环图（directed acyclic graphs，DAGs），将复杂的因果关系可视化，提供识别选择偏倚的直观方法，并通过冲撞分层偏倚的图形结构来验证不同类型的选择偏倚。在实际研究中，可能同时存在多种偏倚，对冲撞变量进行不恰当的调整会新增冲撞分层偏倚，打开后门路径，引入混杂偏倚，甚至改变原有混杂偏倚的大小与方向。为了得到暴露到结局的无偏估计，研究者可以通过构建DAGs，帮助识别冲撞变量，防止冲撞偏倚的发生。Abstract: In the etiology study of epidemiology, selection bias will lead to the fact that the research sample cannot represent the general population, the association between exposure and outcome among those selected for analysis differs from the association among those eligible, and the true causal association cannot be inferred. Directed acyclic graphs (DAGs) could visualize complex causality, introduce the Collider-stratification bias using simple graphics language, provide a simple and intuitive way to identify Selection bias, different types of selection bias are verified by the graphic structure of the Collider-stratification bias. In practical studies, there may be multiple biases at the same time, improper adjustment of the collider will lead to Collider-stratification bias, open a backdoor path, even change the size and direction of the confounding bias. In order to obtain an unbiased estimate of the exposure to the outcome, it is necessary to identify the collider and avoid the adjustment to prevent the occurrence of Collider-stratification bias by using DAGs.
-
Key words:
- Etiology study /
- Directed acyclic graph /
- Selection bias /
- Collider-stratification bias
-
图 3 各类选择偏倚的有向无环图
Figure 3. Directed acyclic graphs with various types of selection bias
图 5 调整C后的有向无环图
Figure 5. Directed acyclic graph showing E-D association graph after C adjustment
图 6 a、b、c. 肥胖、吸烟、CVD与死亡的DAGs
Figure 6. a, b, c DAGs of obesity, smoking, CVD and death
表 1 不同模型中肥胖的效应
Table 1. Effects of obesity in different models
模型 死亡率比(IRR) 调整吸烟 1.10(95% CI: 1.03~1.18) 未调整吸烟 1.07(95% CI: 1.00~1.15) 
下载: 导出CSV
表 2 CVD分层后肥胖的效应
Table 2. Effects of obesity after CVD stratification
死亡率比(IRR) CVD患者 0.95(95% CI: 0.88~1.03) 无CVD者 1.03(95% CI: 0.91~1.18)
下载: 导出CSV
表 3 CVD患者中吸烟状态对肥胖效应的影响
Table 3. Effect of smoking status on obesity in CVD patients
死亡率比(IRR) 吸烟 0.89(95% CI: 0.81~0.99) 不吸烟 1.20(95% CI: 1.03~1.41)
下载: 导出CSV
-
[1] Janszky I, Ahlbom A, Svensson AC. The Janus face of statistical adjustment: confounders versus colliders[J]. Eur J Epidemiol, 2010, 25(6): 361-363. DOI: 10.1007/s10654-010-9462-4. [2] Pearl J. Causual diagrams for empirical research[J]. Biometrika, 1995, 82(4): 702-710. DOI: 10.2307/233739. [3] 向韧, 戴文杰, 熊元, 等. 有向无环图在因果推断控制混杂因素中的应用[J]. 中华流行病学杂志, 2016, 37(7): 1035-1038. DOI: 10.3760/cma.j.issn.0254-6450.2016.07.025.Xiang R, Dai WJ, Xiong Y, et al. Application of directed acyclic graphs in control of confounding[J]. Chin J Epidemiol, 2016, 37(7): 1035-1038. DOI: 10.3760/cma.j.issn.0254-6450.2016.07.025. [4] Pearl J. An introduction to causal inference[J]. International Journal of Biostatistics, 2010, 6(2): Article 7. DOI: 10.2202/1557-4679.1203. [5] Greenland S, Pearl J, Robins JM. Causal diagrams for epidemiologic research[J]. Epidemiology, 1999, 10(1): 37. DOI: 10.1097/00001648-199901000-00008. [6] Liu W, Brookhart MA, Schneeweiss S, et al. Implications of M bias in epidemiologic studies: a simulation study[J]. American Journal of Epidemiology, 2012, 176(10): 938-948. DOI: 10.1093/aje/kws165. [7] Munafo MR, Tilling K, Taylor AE, et al. Collider scope: when selection bias can substantially influence observed associations[J]. International Journal of Epidemiology, 2017, 47(1): 226-235. DOI: 10.1093/ije/dyx206. [8] Gage SH, Davey SG, Ware JJ, et al. G=E: What GWAS can tell us about the environment[J]. Plos Genetics, 2016, 12(2): e1005765. DOI: 10.1371/journal.pgen.1005765. [9] 詹思延. 流行病学-第7版[M]. 北京: 人民卫生出版社, 2012: 144-148.Zhan SY. Epidemiology-seventh edition[M]. Beijing: People's Medical Publishing House, 2012: 144-148. [10] Hernan MA, Hernandez-Diaz S, Robins JM. A structural approach to selection bias[J]. Epidemiology, 2004, 15(5): 615-625. DOI: 10.1097/01.ede.0000135174.63482.43. [11] Berkson J. Limitations of the application of fourfold table analysis to hospital data[J]. Int J Epidemiol, 2014, 43(2): 511-5. DOI: 10.2307/3002000. [12] Hernan MA. Invited commentary: selection bias without colliders[J]. Am J Epidemiol, 2017, 185(11): 1048-1050. DOI: 10.1093/aje/kwx077. [13] Horwitz RI, Feinstein AR. Alternative analytic methods for case-control studies of estrogens and endometrial cancer[J]. N Engl Jo Med, 1978, 299(20): 1089. DOI:10.1056/nejm 197811162992001. [14] Greenland S, Neutra R. An analysis of detection bias and proposed corrections in the study of estrogens and endometrial cancer[J]. J Chronic Dis, 1981, 34(9-10): 433-438. DOI: 10.1016/0021-9681(81)90002-3. [15] Clarice R. Weinberg. Toward a clearer definition of confounding[J]. Am J Epidemiol, 1993, 137(1): 1-8. DOI: 10.1093/oxfordjournals.aje.a116591. [16] Morgenstern JMR. The foundations of confounding in epidemiology[J]. Comput Math Appli, 1987, 14(9): 869-916. DOI: 10.1016/0898-1221(87)90236-7. [17] Greenland S. Quantifying biases in causal models: classical confounding vs collider-stratification bias[J]. Epidemiology, 2003, 14(3): 300-306. DOI: 10.2307/3703850. [18] 谭红专. 病因流行病学研究方法进展[J]. 中华疾病控制杂志, 2017, 21(8): 755-757. DOI: 10.16462/j.cnki.zhjbkz.2017.08.001.Tan HZ. Progress in epidemiological etiology study methods[J]. Chi J Dis Control Prev, 2017, 21(8): 755-757. DOI: 10.16462/j.cnki.zhjbkz.2017.08.001. [19] Tu YK. Directed acyclic graphs and structural equation modelling[M]. Modern Methods for Epidemiology. Springer Netherlands, 2012: 191-203. DOI: 10.1007/978-94-007-3024-311. [20] Whitcomb BW, Schisterman EF, Perkins NJ, et al. Quantification of collider-stratification bias and the birthweight paradox[J]. Paediatr Perinat Epidemiol, 2009, 23(5): 394-402. DOI: 10.1111/j.1365-3016.2009.01053.x. [21] Banack HR, Kaufman JS. The obesity paradox: understanding the effect of obesity on mortality among individuals with cardiovascular disease[J]. Prev Med, 2014, 62(2): 96-102. DOI: 10.1016/j.ypmed.2014.02.003. [22] Banack HR, Kaufman JS. From bad to worse: collider stratification amplifies confounding bias in the "obesity paradox"[J]. European Journal of Epidemiology, 2015, 30(10): 1111-1114. DOI: 10.1007/s10654-015-0069-7. [23] Banack HR, Kaufman JS. Does selection bias explain the obesity paradox among individuals with cardiovascular disease?[J]. Ann Epidemiol, 2015, 25(5): 342-349. DOI: 10.1016/j.annepidem.2015.02.008. [24] VanderWeele TJ, Robins JM. Signed directed acyclic graphs for causal inference[J]. J Royal Stat Soc Series B Stat Methodol, 2010, 72(1): 111-127. DOI: 10.1111/j.1467-9868.2009.00728.x. [25] Shrier I, Platt RW. Reducing bias through directed acyclic graphs[J]. BMC Med Res Methodol, 2008, 8(1): 70-70. DOI: 10.1186/1471-2288-8-70. [26] Ananth CV, Schisterman EF. Confounding, causality and confusion: the role of intermediate variables in interpreting observational studies in obstetrics[J]. Am J Obstet Gynecol, 2017: S0002937817305173. DOI: 10.1016/j.ajog.2017.04.016. [27] Greenland S, Pearl J. Causal diagrams[J]. International encyclopedia of statistical science, 2014, 32(2): 4-32. DOI: 10.1002/9781118445112.stat03732. [28] Maclehose RF, Kaufman JS. The wizard of Odds[J]. Epidemiology, 2012, 23(1): 10-12. DOI: 10.1097/EDE.0b013e31823b5492. -
点击查看大图
图(6) / 表(3)
计量
- 文章访问数: 779
- HTML全文浏览量: 133
- PDF下载量: 129
- 被引次数: 0