Reply to D Allotey et al and SC Lucan

Dear Sir:

We appreciate the comments by Allotey et al and Lucan regarding our article. Allotey et al were concerned about dietary measurement. Unfortunately, there is no comprehensive food-frequency questionnaire (FFQ) that has been validated in a multitribal cohort of American Indians (AIs). We agree that more work is needed on validating FFQs in minority populations. However, as noted in our article, the Block 98 FFQ is one of the most widely used food questionnaires, and it has been shown to be both reliable and valid in several multiethnic cohorts (1, 2). The FFQ was administered by trained community members, and average nutrient intakes obtained from the FFQ were similar to those obtained from 24-h dietary recall in 1993–1995. Among participants who completed both the FFQ and 24-h recall, the median percentage of calories from saturated fat was 12% for both instruments (3). Because there are foods that are commonly consumed in AI communities that are not included on standard FFQs (4), we included an AI foods supplemental questionnaire in addition to the Block FFQ. The foods listed on the supplemental questionnaire, developed with the help of the participating tribes, included 9 foods commonly consumed in these communities, including spam, menudo, polzole, guysava, red/green chili, Indian tacos, fry bread, and corn and flour tortillas. We also agree that geography may influence the intake of specific foods. In preliminary analyses we found no evidence of major dietary differences between sites, and we combined data across the sites for our analyses. In addition, our analyses examined the impact of adjustment for site and assessed the interaction of dietary intake and site. As noted in the article, we found no evidence of an interaction between dietary intake and site when assessing diabetes risk.

Allotey et al referenced Serdula et al (5) to support their claim that the use of single food items on FFQs leads to overreporting of intake. In that study, dietary intake differed between participants given a “grouped foods FFQ” (27 high-fat foods grouped into 14 food categories) compared with a “separated FFQ” (29 nongrouped high-fat food items). Specifically, participants who completed the “grouped foods FFQ” were more likely to report never consuming the foods or had lower mean intake than did individuals who completed the “separated FFQ.” However, it is unclear whether the “grouped foods FFQ” underestimated dietary intake or whether the “separated FFQ” overestimated dietary intake. As Serdula et al mentioned in their discussion, they did not conduct an in-depth dietary assessment using 24-h recalls or food diaries as a comparison and could not determine whether the grouped or separated FFQs produced more accurate estimates of intake. The FFQ and AI supplement used in our analyses included 4 individual questions regarding intake of processed meat (breakfast sausage, hot dogs, lunch meat, and spam) and 6 questions about intake of unprocessed red meat (hamburger/cheeseburger, steak/roast beef, pork chops/pork roast/dinner ham, veal/lamb/deer, ribs, liver). We believe it is unlikely that ascertainment of spam intake created meaningful bias.

We agree with Lucan that qualitative differences in types of unprocessed red meat may differ in their relation with diabetes risk. However, the Block FFQ does not ascertain information on source of meat (free range compared with factory), and we could not address this question. In our FFQ, the intake of wild meat, such as deer, was combined into a category with veal/lamb. The consumption of these foods was uncommon, so we were unable to assess the association of wild meat with diabetes risk separately.

Allotey et al were also concerned about the impact of nutrient density models on risk estimates. Multivariate nutrient density models are widely used to describe dietary data in nutritional epidemiology because these models have the advantage of correcting for over- or underreporting of intake (6). However, in sensitivity analyses we used standard multivariate models (servings/d) and residual methods to classify nutrients and foods. The use of these alternative methods did not materially alter risk estimates.

Allotey et al also had concerns about generalizability. Our reported risk estimates (and the partial attenuation of risk estimates after adjustment for baseline BMI) are consistent with the results of most studies that examined the association of processed meat intake and incident diabetes. As shown by Mozaffarian et al (7) and Vergnaud et al (8), intake of processed meats is associated with weight gain and obesity. Although baseline BMI and past-year dietary information were collected at the same time, it is possible that intake of processed meats could have led to higher baseline BMIs, because the FFQ-ascertained information on past-year dietary intake and dietary habits are quite stable in AI communities. As such, the model that adjusted for BMI may underestimate the association of processed meat intake with incident diabetes; for this reason, we reported risk estimates unadjusted and adjusted for BMI in our article’s Table 2. In sensitivity analyses, we also assessed the relation of processed meat intake and weight gain (change in BMI from baseline to follow-up), and participants who reported consuming more processed meat at baseline were more likely to gain weight during the 5 y of follow-up.

We recognize that the components of the metabolic syndrome are major risk factors for diabetes development. However, adjustment for fasting glucose, blood pressure, HDL cholesterol, and triglycerides did not materially alter reported ORs. Similar to adjustment for BMI, adjustment for waist circumference slightly attenuated the results. However, like BMI, adjustment for waist circumference may underestimate the association of processed meat intake and incident diabetes if higher processed meat consumption leads to obesity.

The association of processed meat intake and incident diabetes has been reported in several cohorts across a wide range of ethnicities, BMIs, and other metabolic risk factors. The magnitude of our reported risk estimates are consistent with those found in other cohort studies. For the reasons noted in our article and those mentioned above, it is unlikely that our findings could be explained by baseline BMI, imprecision of dietary estimates, or the effect of other metabolic risk factors on diabetes risk.