Shared and ethnic background site-specific dietary patterns in the Hispanic Community Health Study/Study of Latinos (HCHS/SOL)

Abstract

Dietary patterns (DPs) synthesize multiple related dietary components in one or more combined variables. A drawback of DPs is their limited reproducibility across subpopulations, especially adopting a posteriori DPs, derived using standard multivariate methods [e.g., factor analysis (FA)]. Standard approaches assessing reproducibility of FA-based DPs mostly rely on correlation coefficients/agreement measures between pairs of factors and do not consider any statistical model. Multi-study factor analysis builds upon standard FA model to identify DPs shared across all subpopulations and those specific to some subpopulations. Pattern reproducibility is investigated from a different perspective: a shared DP identified within multi-study factor analysis is “reproducible” since it is common to all subpopulations. Bayesian multi-study factor analysis (BMSFA) has been developed to improve DP retention and identification, two critical issues as the number of subpopulations analyzed increases.Using baseline (2008-2011) 24-hour dietary recalls from the Hispanic Community Health Study/Study of Latinos (n=16,415), we applied the BMSFA on 42 common nutrients to identify shared and subpopulation-specific DPs where subpopulations were defined as the cross-classification of ethnic background and study site (EBS).Overall, 4 shared DPs were identified: Plant-based foods, Processed foods, Dairy products, and Seafood. At the subpopulation level, we identified 12 EBS-specific DPs, one for each EBS category, primarily representing variants of foods from animal sources. Different nuances were expressed by subsets of fairly similar EBS-specific DPs, including an Animal vs. vegetable source, an Animal source only, and a Poultry vs. dairy products overarching DPs. Shared DPs from BMSFA were similar to their counterparts from standard FA and frequentist multi-study factor analysis; EBS-specific DPs from BMSFA were better characterized than those from frequentist multi-study factor analysis.In conclusion, the BMSFA successfully captured sources of both dietary homogeneity and heterogeneity in a large well-characterized study of US Hispanics/Latino adults by ethnic background and site.

Dietary Acculturation Is Associated With Altered Gut Microbiome, Circulating Metabolites, and Cardiovascular Disease Risk in US Hispanics and Latinos: Results From HCHS/SOL

“…Food groups were included according to epidemiological findings linking foods and cardiometabolic diseases 17,18 and data availability and consumption patterns in HCHS/SOL. 19,20 As shown in Table S1, 37 food groups were aggregated into 30 candidate food groups based on food component similarity and consumption level in HCHS/SOL to reduce data sparsity.…”

Dietary Patterns and Cancer Risk: An Overview with Focus on Methods

“…The Bayesian MSFA is adopted in the Hispanic Community Health Study/Study of Latinos (HCHS/SOL) [110], a US multi-site community-based cohort, focusing on health and risk factors of cardiovascular and pulmonary outcomes of Hispanic/Latino adults [27]. The method captures common and subpopulation-specific dietary patterns on 42 nutrients across available combinations of 4 US field sites (Bronx, Chicago, Miami, and San Diego) and 6 Hispanic/Latino ethnic backgrounds (Cuban, Dominican Republic, Mexican, Puerto Rican, Central and South American).…”

“…In recent years, innovation has been observed in methods identifying a posteriori dietary patterns across different studies or known subgroups (e.g., by center or ethnicity) within the same study [24,27]. These dietary patterns have been successfully related to cancer risk [24].…”

Identifying Dietary Consumption Patterns from Survey Data: A Bayesian Nonparametric Latent Class Model

“…Bayesian nonparametric mixture models have been applied in nutritional settings, but implementation accounting for survey design in model estimation has not been fully explored (Fahey and others , 2007; De Vito and others , 2019; Stephenson and others , 2020 a ). Bayesian nonparametric mixture models have been applied to diet survey data, but the survey weights were applied after parameters were estimated from the sampling algorithm (Stephenson and others , 2020 b ; Stephenson and Willett, 2022; De Vito and others , 2022). Kunihama and others (2016) used a Dirichlet Process mixture model to introduce a sampling algorithm that can incorporate survey weights directly into the estimation of a Bayesian nonparametric mixture model.…”

“…Bayesian survey data applications have centered mostly on generating inference for derived population-based estimates (Si et al, 2015; Savitsky and Toth, 2016; Gunawan et al, 2020), but have not been fully explored in regards to model-based clustering. Bayesian nonparametric mixture models that utilized dietary intake data either did not contain complex survey data (Fahey et al, 2007; De Vito et al, 2019; Stephenson et al, 2020a), or applied sampling weights posthoc after model parameter estimation was complete (Stephenson et al, 2020b; De Vito et al, 2022). Kunihama et al (2016) is one of the few that introduced a sampling algorithm that incorporates survey weights directly into the estimation, but did not take into account sampling variability present in nationally-representative surveys.…”