Correspondence re: H. J. Lin et al., Glutathione Transferase Null Genotype, Broccoli, and Lower Prevalence of Colorectal Adenomas. Cancer Epidemiol., Biomark, Prev., 7: 647–652, 1998

We thank Drs. Acquavella and Cullen for their interest in our report and appreciate the opportunity to further examine our results. As noted in Lin et al. (1) and reported earlier in Witte et al. (2), our study found an inverse association between cruciferous vegetables and colorectal adenomatous polyps among all study subjects. To further explore this result, Lin et al. (1) investigated the potential interaction between glutathione transferase M1 (GSTM1) genotype and the intake of these vegetables among 459 cases and 507 controls. The cases and controls were matched on age, sex, date of sigmoidoscopy, and medical center (i.e., where the sigmoidoscopy was performed). In addition, potential confounders included race, body mass index, leisure time physical activity, family history of colorectal cancer, intake of nonsteroidal anti-inflammatory drugs, smoking, and intake of total energy, saturated fat, and fruits and vegetables.

We used a basic stratified approach to initially analyze the data. This is certainly an important step (3) and was used by Acquavella and Cullen (4) to analyze the data published in Lin et al. (1). However, in light of the numerous factors requiring consideration as potential confounders, sample size limitations quickly made stratified analysis impractical. Therefore, we undertook conventional logistic regression of colorectal adenomas on cruciferous vegetables, genotype, matching factors, and potential confounders. That is, we regressed adenomas on the covariates. (Our final model included only the matching variables smoking, and intake of total energy, saturated fat, and all other fruits and vegetables.) Furthermore, to assess the possible interaction between GSTM1 genotype and cruciferous vegetable intake, we included an interaction term in the model. Our primary analysis, therefore, included all of the data at once and evaluated trends and interactions with respect to a single referent category (i.e., GSTM1 null and no intake of broccoli).

In contrast, the analysis by Acquavella and Cullen (4) split the data into two groups based on GSTM1 genotype, and then undertook an analysis within each group. Thus, their analyses each had different referent categories (i.e., no intake of broccoli). In a similar fashion, the two groups of data could be analyzed separately with a logistic model that included the matching factors and potential confounders. Again, because of sample size, this was not a realistic option for the refined analysis of our data. Furthermore, splitting the data based on GSTM1 genotype restricts one from being able to investigate any main genetic effect on colorectal adenomas.

More important, however, is the impact of using different referent categories on the interpretation of results. Acquavella and Cullen (4) show that, for individuals within each GSTM1 genotype group, there is a decrease in colorectal adenoma prevalence with increasing broccoli intake. Table 1 shows their analytic approach, whereby individuals with identical GSTM1 genotypes are analyzed separately, and those with no intake of broccoli are the referent for each subset of data. From this analysis, we see that, within GSTM1 genotypes, there seems to be a decreased prevalence when comparing high-with no-broccoli intake.

A more thorough look at the raw data, however, indicates that the prevalence of colorectal adenomas may differ depending on both broccoli intake and GSTM1 genotype. The ratio of cases/controls for no- and high-broccoli intake are 1.7 (61:36) and 0.9 (64:70), respectively, among those with the nonnull genotype. In contrast, these ratios are 1.2 (44:36) and 0.4 (33:77) among those with the null genotype. Therefore, concluding that there is no difference in prevalence reduction based on genotype may be erroneous. Table 2 shows the analytic approach used in Lin et al. (1). Here, we see that both increasing broccoli intake and the GSTM1-null genotype are inversely associated with the prevalence of colorectal adenomas. Furthermore, comparing subjects with both a high intake of broccoli and the GSTM1-null genotype to the referent group (no-broccoli, nonnull) gives an OR² of 0.25. Thus, the analytic approach and corresponding referent categories drive the difference between our results (1) and those of Acquavella and Cullen (4).

On another point, Acquavella and Cullen (4) state: “The lack of heterogeneity in the dichotomous analysis suggests that the data might be well characterized by an analysis that is not stratified by genotype.” In general, one should be extremely careful about making any sort of conclusion about how to best estimate main effects or interactions based on a dichotomous analysis. Because associations are estimated as being constant within categories, fewer groupings result in stronger assumptions about homogeneity. For example, our Table 1 (used here for illustrative purposes) makes the assumption that the associations are constant within the highest and no categories of broccoli. To help address this, one can use analyses that incorporate more flexible modeling approaches such as quadratic splines, as was done with our initial report on cruciferous vegetables (2, 5).

Finally, selective quoting of Lin et al. (1) might lead to over-interpretation of our results. The sentence partly quoted by Acquavella and Cullen (4) actually reads: “A protective effect of broccoli and other cruciferous vegetables was observed only among subjects with the GSTM1-null genotype, as judged from statistically significant Ps for trend”. The statement corresponds directly to the results from P values obtained when assessing trend with the conventional logistic regression model. A more appropriate conclusion to our findings is found in the last paragraph of Lin et al. (1): “In summary, prevalence of colorectal adenomas was lowest among people with high broccoli intake and the GSTM1-null genotype . If confirmed, the results suggest that people with the GSTM1-null genotype may benefit more from broccoli.”