Food/Food Group-Based Analyses and Measurement Errors in Nutritional Epidemiology

Abstract

Foods and food groups represent a major category of nutritional epidemiologic analyses besides nutrients. They provide a means to explore diet-disease associations when a nutrient-based hypothesis has not been formulated or when the development of disease is attributable to complex interactions and synergistic effects among dietary components, uncommonly studied nutrients without accurate food composition information, or even non-nutrient factors such as glycemic index and beneficial bacteria. More importantly, results from food/food group-based analyses are recognized to be easy to interpret and readily translatable into dietary recommendations. However, despite the importance of foods and food groups, nutritional epidemiology methods were developed and validated primarily for nutrient-based analyses. For example, evaluations of the validity of foods and food groups measured with semiquantitative food frequency questionnaires (FFQs), a dietary assessment method commonly used in large population-based studies, have been less frequently conducted than the evaluations of FFQ-measured nutrients. Measurement error structure and statistical correction approaches have been extensively evaluated for nutrients, while those for foods are understudied. Challenges that partially account for this imbalance in method development include an unstandardized data processing of food and food group intakes and the fact that their distributions are highly skewed. Moreover, the values of food/food group-based analyses are subject to further exploration. In this dissertation, I present three studies, including a validation of FFQ in measuring foods and food groups, an analysis of the association between red meats as a food group and type 2 diabetes (T2D) corrected for exposure measurement errors, and an analysis examining the association of a food-based dietary index with the risk of heart failure. The first study evaluates the validity and reproducibility of an FFQ for measuring intakes of 149 foods and 25 food groups among 736 participants of the Women’s Lifestyle Validation Study (WLVS) and 649 participants of the Men’s Lifestyle Validation Study (MLVS). This is the largest validation of FFQ-measured foods/food groups among women and men in the US. It is also the first comprehensive validation of all individual foods, as well as the derived food groups, on an FFQ widely used in the US population since 1993. Compared to the validation of FFQ-measured foods conducted over 30 years ago, the current study incorporates methodological advances, including the use of rank-based correlation that accommodates the skewed distributions of food and food groups and the application of an arcsin-based estimator, allowing estimations of the confidence intervals without making normality assumptions. This study shows that the FFQ used in the Nurses’ Health Study (NHS), NHS II, and Health Professionals Follow-Up Study (HPFS) has reasonably high reproducibility and validity in measuring food and food group intakes among both women and men. FFQ is eminently applicable in nutritional epidemiolocal studies aiming to measure long-term dietary intakes of foods and food groups for chronic disease risk assessments. The second study assesses the associations of FFQ-measured total red meat, processed red meat, and unprocessed red meat with the risk of T2D among 216,695 US women and men from the NHS, NHS II, and HPFS. At the time when this dissertation was developed, the relationship between red meat consumption and T2D was questioned due to the lack of evidence from short-term randomized controlled trials and the challenges to the quality of observational studies. My analysis of 22,761 incident T2D cases and dietary intakes assessed repeatedly over 30 years of follow-up provides a precise estimation of the relationship between red meats and T2D. With the availability of two large calibration studies, I calibrated self-reported red meat intakes with weighed diet records using a regression calibration approach adapted for repeated dietary assessments for the first time. I also evaluated the effects of substituting other protein sources for red meats, examined latency periods and potential reverse causation, and compared the impacts of different approaches for describing dietary exposure. This study shows that total red meat, processed red meat, and unprocessed red meat are each associated with a higher risk of T2D. Substituting nuts and legumes, and dairy foods for all types of red meat is associated with lower risks of T2D. This study supports current dietary recommendations for limiting consumption of red meat intake and emphasizes the importance of alternative sources of protein for T2D prevention. In the third study, I examined the associations of the Prime Diet Quality Score (PDQS) and the Alternative Healthy Eating Index (AHEI) with the risk of heart failure (HF) and its two major subtypes, HF with preserved ejection fraction (HFpEF) and HF with reduced ejection fraction (HFrEF), among men from the HPFS. The PDQS is a food-based dietary index calculated with intakes of 21 foods and food groups. In contrast, the AHEI, which is the dietary index most predictive of multiple cardiometabolic diseases, is computed with both foods and nutrients. While the two dietary indices are both developed to characterize overall dietary patterns and capture the healthfulness of diet, the food-based scoring procedure of PDQS reduces the burden of dietary assessment for both clinicians and patients and, therefore, is more applicable in clinical settings. This study shows that a high-quality diet is associated with a substantially lower risk of HFrEF among male health professionals. The abbreviated food-based PDQS and the more complex AHEI have similar associations with HF.