Advancing methods to capture and analyze dietary patterns

Abstract

Background: Suboptimal diet quality is linked to poor health outcomes and is associated with many sociodemographic characteristics, including several that are indicators of inequities. Assessments of dietary intake have shifted over time from investigations of single foods and nutrients towards dietary patterns. This change has led to developments in methods to capture and analyze dietary patterns, from short tools that quickly assess the overall diet to novel analytic methods. These methodological advances present opportunities to better understand dietary patterns in Canada and globally. Research objectives: The objectives of this research were to (1) develop a brief dietary screener to assess alignment of dietary intakes with the 2019 Canada’s Food Guide healthy food choices recommendations; (2) develop a scoring system for the screener and assess the construct validity of the screener; (3) identify novel methods used to characterize dietary patterns through a scoping review of the literature; and (4) explore the capacity of probabilistic graphical models to expand our understanding of the joint relationships between multidimensional dietary patterns and intersecting sociodemographic characteristics. Methods and results: The first study in this dissertation (Chapter 4) discusses the process to develop the Canadian Food Intake Screener. This was achieved by mapping the dietary guidance in the 2019 Canada’s Food Guide and reviewing existing tools to develop a draft screener, which was reviewed by Health Canada and external collaborators (n=15). The screener was revised iteratively based on feedback from cognitive testing conducted among adults aged 18-65 years in English (n=17) and French (n=16) and from face and content validity testing conducted with experts (n=16). The screener was well understood overall and testing informed refinement to finalize the Canadian Food Intake Screener, which includes 16 questions to rapidly assess alignment of adults’ intake with the 2019 Canada’s Food Guide healthy food choices recommendations. The second study in the dissertation (Chapter 5) was conducted to develop a scoring system for the screener and evaluate the screener’s construct validity among adults aged 18 to 65 years. Analysis of variance (ANOVA) was used to compare screener scores among subgroups with known differences in diet quality. The correlation between scores on the screener and the Healthy Eating Food Index-2019 (HEFI-2019), which also assesses alignment of intake with the 2019 Canada’s Food Guide healthy food choices recommendations, was assessed. Adults aged 18-65 years (n=154) completed the screener, answered a range of questions about their health and sociodemographic characteristics, and completed up to two 24-hour dietary recalls. The mean screener score was 35 points (SD = 4.7; maximum 65), ranging from 25 (1st percentile) to 45 (99th percentile). Meaningful differences in screener scores were observed in hypothesized directions by gender identity (p = 0.06), perceived income adequacy (p = 0.07), education (p = 0.02), and smoking status (p = 0.003). The correlation between screener and HEFI-2019 scores was 0.53 (SE = 0.12). The screener demonstrated moderate construct validity, indicating that it is appropriate for use for rapid assessment of alignment of adults’ intake with the healthy food choices recommendations when comprehensive dietary assessment is not possible. In Chapter 6, novel methods used to characterize dietary patterns in peer-reviewed literature were summarized using a scoping review. The databases MEDLINE, CINAHL, and Scopus were searched using keywords such as such as machine learning, latent class analysis, and least absolute shrinkage and selection operator (LASSO) to identify novel methods used to describe dietary patterns. Of 5274 records identified, 24 met the inclusion criteria. Twelve of 24 articles were published since 2020. A range of methods was applied to identify dietary patterns, with nine studies using approaches that have applications in machine learning to characterize dietary patterns, and the remaining 15 using other novel methods such as latent class analysis, LASSO, or treelet transform. Future work to guide the application, interpretation, and comparability of these methods is necessary to enable synthesis of the literature to inform policies and programs. The final study (Chapter 7) in this dissertation examined the ability of probabilistic graphical models to explore the joint relationships between dietary patterns and sociodemographic characteristics. While prior research has established relationships between dietary patterns and sociodemographic characteristics, it has rarely considered the multidimensional relationships between dietary components or possible intersecting relationships among sociodemographic characteristics. Mixed graphical models, a network method, enable explorations of these complex joint relationships, which have largely been unexplored in the Canadian context. We conducted a secondary analysis of first 24-hour dietary recalls collected from adults aged 18 years and above who participated in the 2015 Canadian Community Health Survey Nutrition (n=14 097). Mixed graphical models were used to identify joint relationships between amounts consumed in grams of thirty log-transformed food groups and age, sex, education, income, household food security status, geographic region, employment status, and smoking status. Sociodemographic characteristics formed a network, with several pairwise relationships. Several dietary components also formed networks, often patterning by food group. Age and sex were the sociodemographic characteristics most strongly connected to dietary components. This research applied mixed graphical models to provide deeper insights into the internal structures of the dietary patterns of adults in Canada, and how sociodemographic characteristics are jointly related with dietary patterns. Probabilistic graphical models offer promise to complement existing methods to characterize dietary patterns, such as indices. Conclusions: This dissertation makes contributions to dietary patterns research with respect to both collecting data reflective of overall dietary patterns and analytic methods to capture their complexity. The advances from these studies can be applied to inform targeted research and policies promoting public health nutrition in Canada and beyond.