Epidemiological studies on colorectal cancer : Some design issues and analytical considerations

Abstract: Determinants of colon and rectal cancer have been studied in two case-control data sets, one from the north of Sweden, i.e., Norrland, and the other from the county of Östergötland. In the studies carried out, the main interest was focused on occupational aspects, but diet, medication and earlier diseases were also considered, including a detailed interview regarding diet for a sample of the subjects in Östergötland.A main finding was an association between sedentary work and an increased risk of colon cancer whereas a decreased risk was seen for rectal cancer. Forestry and agricultural work was associated with a reduced risk. Railway employees had an increased risk in Norrland and a tendency to increased risk in Östergötland. Among specific exposures, asbestos was associated with an increased risk. Rectal cancer did not share any of these factors except for a decreased risk for forestry work. The main finding regarding rectal cancer was an increased risk for painters who sanded walls before painting. It seems as occupational exposures influence the risk of colon and rectal cancer in different ways.Regarding diet, reduced risks was seen for daily intake of cheese and to some extent for fiber rich diet. In Östergötland, a protective effect was seen for calcium and cereal fiber on colon cancer. Tea was associated with a reduced risk, whereas coffee or alcohol did not show any association. Increased risk was found for smoked food. Fiber had a protective effect on rectal cancer, whereas alcohol appeared as a strong risk factor.The Norrland study was also analyzed regarding drug intake and other diseases than cancer. In particular, increased risks was associated with iron and verapamil medication. No effect was seen regarding acetylsalicylic drugs. Ulcerative colitis appeared as a strong risk factor and other determinants of risk were cholecystectomy and hyperplasia oftl1e prostate. Appendectomy was associated with a decreased risk.Information regarding inactivation of the tumor suppressor gene p53 was available for some cases in the Östergötland study, providing for a new approach by means of a case-control analysis. High intake of alcohol andthe use of hydralazine-containing drugs were associated with an inactivated gene, whereas iron medication and familial colorectal cancer were not. The results indicate that dividing the cases in subgroups based on molecular biology may reveal more specific associations between exposure and subgroups of colorectal cancer.Some variables were further analyzed in order to check if there would be any gain in using time or intensity as an effect variable as compared to a crude exposure variable. Considering design issues and data analysis, it was found that if a crude determinant indicates that there is an association, then a more in-depth analysis by using time or intensity variables might be worthwhile. A simple method for categorizing continuous variables was also demonstrated. Furthermore, a more thorough analysis of the association between dietary fat and physical activity indicated an interaction between low physical activity and high intake of fat.Validity aspects were critically discussed in relation to the studies. Selection and information bias were found unlikely to have occurred in the kind of data at hand and neither differential nor non-differential misclassification should be a serious problem. Confounding, including combinations wiih misclassification, was considered, in relation to the studies, but with the methods used such a phenomena is less likely to have influenced the results.