Year: 1997
Type: letter
Abstract: Several ( 1 – 8 ), but not all ( 9 – 12 ), epidemiologic studies indicate that consumption of fat-rich diets increases risk of lung cancer. In 1993, we reported ( 13 ) a pronounced association between dietary fat and lung cancer risk. In a large, population-based, case-control study of nonsmoking Missouri women, risk of lung cancer increased across increasing levels of saturated fat intake. The odds ratio (OR) was more than sixfold greater for the highest quintile of consumption compared with the lowest quintile. We have since found that the ORs vary substantially according to the method of energy adjustment. In this correspondence, we report ORs based on two regression models other than the one used in the original analysis. The significantly elevated risk associated with saturated fat intake was not eliminated under these two models; however, the magnitude of most of the risk estimates and the consistency of a trend were clearly reduced using the alternative methods of energy adjustment ( Table 1 ). For the reasons noted below, we believe our original estimates of risk were inflated and those resulting from the other methods are more valid. The high correlation between saturated fat and total calories makes it difficult to assess the effect of saturated fat independent of its association with energy. In our original analysis ( 13 ), we entered both saturated fat and total calories as categorical variables in a standard multivariate model, since the nutrient residual method did not appear to offer any advantage. As others ( 14 – 16 ) have noted, the linear nutrient residual model is equivalent to the linear standard multivariate model. Furthermore, residuals do not convey an intuitive sense of nutrient intake, such as grams or calories of saturated fat consumed. At the time, we did not appreciate that the equivalence of the two methods ended when the nutrient intake data were categorized, a standard practice in epidemiologic studies. As Brown et al. ( 17 ) and others ( 18 ) noted, when estimating the effect of increasing intake from dietary fat while keeping total energy intake constant (substitution effect), the standard multivariate model exaggerates the true variation in fat intake when data are modeled as quantile-categorical variables. Because the variability of the nutrient residuals provides a better estimate of the true variation in the nutrient of interest when total energy intake is fixed, the quantiles of the residual distribution provide a more valid description of the actual ORs to be expected when changing dietary composition. The multivariate nutrient density approach has the added advantage of being readily interpretable, representing the effect of changing the percentage of fat (or saturated fat) in the diet while keeping total energy intake constant. Despite years of discussion about energy adjustment methods, we noted that only two observational studies of dietary fat and lung cancer ( 12 , 13 ) used any method of energy adjustment. In part, this circumstance reflects the fact that both the rationale and the appropriate method of energy adjustment remain controversial ( 19 , 20 ). In summary, computer simulations ( 17 ) and two analytic studies of dietary fat and cancer risk ( 13 , 21 ) now provide evidence that risk estimates and subsequent conclusions can be profoundly influenced by the method of energy adjustment.
Language: English
Topic: Nutritional Studies and Diet
Field: Medicine
Domain: Health Sciences
Open Access status: bronze
OpenAlex ID: W2119927742