Smith K R, Slattery M L, French T K
Department of Family and Consumer Studies, University of Utah, Salt Lake City 84112.
J Clin Epidemiol. 1991;44(7):715-23. doi: 10.1016/0895-4356(91)90031-4.
The relationship between colon cancer risk and the relative contributions of fat and caloric intake are assessed. A lack of consensus exists regarding the role of each of these dietary factors in the development of colon cancer. This lack of agreement originates from the high correlations between the nutrients, as well as the manner in which researchers treat these dietary variables in their analyses. Four proposed methods are evaluated which attempt to address the collinearity problem in nutritional epidemiology: (1) exclude one or more collinear variables, (2) use the proportion of calories consumed attributable to each dietary component, (3) use a regression-adjustment approach to purge the collinearity correlated nutrients, and (4) ridge regression. Diagnostic tests are reported which assess the degree of collinearity on data collected for a case-control study of colon cancer conducted in Utah between 1979 and 1983. Using logistic regression analyses, we apply each of these methods to case-control data. We find that the risks associated with fat and caloric consumption are extremely sensitive to a priori analytic decisions made by epidemiologist about the underlying collinearity problem.
评估结肠癌风险与脂肪和热量摄入的相对贡献之间的关系。关于这些饮食因素中每一种在结肠癌发展中的作用，目前尚无共识。这种不一致源于营养素之间的高度相关性，以及研究人员在分析中处理这些饮食变量的方式。对四种提出的方法进行了评估，这些方法试图解决营养流行病学中的共线性问题：（1）排除一个或多个共线变量；（2）使用归因于每种饮食成分的热量消耗比例；（3）使用回归调整方法去除共线相关营养素；（4）岭回归。报告了诊断测试，这些测试评估了1979年至1983年在犹他州进行的一项结肠癌病例对照研究收集的数据的共线性程度。使用逻辑回归分析，我们将这些方法中的每一种应用于病例对照数据。我们发现，与脂肪和热量消耗相关的风险对流行病学家关于潜在共线性问题的先验分析决策极为敏感。
