Nurminen M
Department of Epidemiology and Biostatics, Finnish Institute of Occupational Health, Helsinki, Finland.
World Health Stat Q. 1995;48(2):78-84.
Ecological studies require a methodological theory distinct from that used in individual-level epidemiological studies. This article discusses the special problems that need to be considered when planning ecological studies or using the results of such studies. Ecological studies are much more sensitive to bias from model mis-specification than are results from individual-level studies. For example, deviations from linearity in the underlying individual-level regressions can lead to inability to control for confounding in ecological studies, even if no misclassification is present. Conditions for confounding differ in individual-level and ecological analyses. For ecological analyses of means, for example, a covariate will not be a confounder if its mean value in a study region is not associated with either (i) the mean exposure level across regions, or (ii) the mean outcome (disease rate) across regions. On the other hand, effect modification across areas can induce ecological bias even when the number of areas is very large and there is no confounding. In contrast to individual-level studies, independent and nondifferential misclassification of a dichotomous exposure usually leads to bias away from the null hypothesis in aggregate data studies. Failure to standardize disease, exposure and covariate data for other confounders (not included in the regression model) can lead to bias. It should be borne in mind that there is no method available to identify or measure ecological bias. While this conclusion may sound like a general criticism of ecological studies, it is not. It does, however, serve as a reminder of the problems that need to be considered when one designs, analyses, or critically evaluates ecological studies.
生态学研究需要一种与个体水平流行病学研究中所使用的方法理论不同的方法理论。本文讨论了在规划生态学研究或使用此类研究结果时需要考虑的特殊问题。与个体水平研究的结果相比，生态学研究对模型错误设定所导致的偏差更为敏感。例如，即使不存在错误分类，潜在个体水平回归中的线性偏差也可能导致生态学研究中无法控制混杂因素。个体水平和生态学分析中混杂因素的条件有所不同。例如，对于均值的生态学分析，如果一个协变量在研究区域内的均值与（i）各区域的平均暴露水平，或（ii）各区域的平均结局（疾病率）均无关联，那么该协变量就不是一个混杂因素。另一方面，即使区域数量非常大且不存在混杂因素，跨区域的效应修饰也可能导致生态学偏差。与个体水平研究不同，在汇总数据研究中，二分类暴露的独立且无差异错误分类通常会导致偏离零假设的偏差。未能对其他混杂因素（未包含在回归模型中）的疾病、暴露和协变量数据进行标准化可能会导致偏差。应该记住，目前没有可用于识别或测量生态学偏差的方法。虽然这一结论听起来像是对生态学研究的一般性批评，但并非如此。然而，它确实提醒人们在设计、分析或严格评估生态学研究时需要考虑这些问题。
