Identifying dietary patterns using a normal mixture model: application to the EPIC study

Determining classes of food items for health requirements and nutrition guidelines using Gaussian mixture models

Introduction

The identification of classes of nutritionally similar food items is important for creating food exchange lists to meet health requirements and for informing nutrition guidelines and campaigns. Cluster analysis methods can assign food items into classes based on the similarity in their nutrient contents. Finite mixture models use probabilistic classification with the advantage of taking into account the uncertainty of class thresholds.

Methods

This paper uses univariate Gaussian mixture models to determine the probabilistic classification of food items in the South African Food Composition Database (SAFCDB) based on nutrient content.

Results

Classifying food items by animal protein, fatty acid, available carbohydrate, total fibre, sodium, iron, vitamin A, thiamin and riboflavin contents produced data-driven classes with differing means and estimates of variability and could be clearly ranked on a low to high nutrient contents scale. Classifying food items by their sodium content resulted in five classes with the class means ranging from 1.57 to 706.27 mg per 100 g. Four classes were identified based on available carbohydrate content with the highest carbohydrate class having a mean content of 59.15 g per 100 g. Food items clustered into two classes when examining their fatty acid content. Foods with a high iron content had a mean of 1.46 mg per 100 g and was one of three classes identified for iron. Classes containing nutrient-rich food items that exhibited extreme nutrient values were also identified for several vitamins and minerals.

Discussion

The overlap between classes was evident and supports the use of probabilistic classification methods. Food items in each of the identified classes were comparable to allowed food lists developed for therapeutic diets. This data-driven ranking of nutritionally similar classes could be considered for diet planning for medical conditions and individuals with dietary restrictions.

A review of statistical methods for dietary pattern analysis

Background

Dietary pattern analysis is a promising approach to understanding the complex relationship between diet and health. While many statistical methods exist, the literature predominantly focuses on classical methods such as dietary quality scores, principal component analysis, factor analysis, clustering analysis, and reduced rank regression. There are some emerging methods that have rarely or never been reviewed or discussed adequately.

Methods

This paper presents a landscape review of the existing statistical methods used to derive dietary patterns, especially the finite mixture model, treelet transform, data mining, least absolute shrinkage and selection operator and compositional data analysis, in terms of their underlying concepts, advantages and disadvantages, and available software and packages for implementation.

Results

While all statistical methods for dietary pattern analysis have unique features and serve distinct purposes, emerging methods warrant more attention. However, future research is needed to evaluate these emerging methods’ performance in terms of reproducibility, validity, and ability to predict different outcomes.

Conclusion

Selection of the most appropriate method mainly depends on the research questions. As an evolving subject, there is always scope for deriving dietary patterns through new analytic methodologies.

Dietary patterns and oesophageal cancer: a multi-country latent class analysis

BACKGROUND

The considerable differences in food consumption across countries pose major challenges to the research on diet and cancer, due to the difficulty to generalise and reproduce the dietary patterns identified in a specific population.

METHODS

We analysed data from a multicentric case-control study on oesophageal squamous cell carcinoma (ESCC) carried out between 1992 and 2009 in three Italian areas and in the Canton of Vaud, Switzerland, which included 505 cases and 1259 hospital controls. Dietary patterns were derived applying LCA on 24 food groups, controlling for country membership, and non-alcoholic energy intake. A multiple logistic regression model was used to derive odds ratio (ORs) and corresponding 95% CIs for ESCC according to the dietary patterns identified, correcting for classification error.

RESULTS AND CONCLUSION

We identified three dietary patterns. The 'Prudent' pattern was distinguished by a diet rich in fruits and vegetables. The 'Western' pattern was characterised by low consumption of these food groups and higher intakes of sugar. The 'Lower consumers-combination pattern' exhibited a diet poor in most of the nutrients, preferences for fish, potatoes, meat and a few specific types of vegetables. Differences between Italy and Switzerland emerged for pattern sizes and for specific single food preferences. Compared to the 'Prudent' pattern, the 'Western' and the 'Lower consumers-combination' patterns were associated with an increased risk of ESCC (OR=3.04, 95% CI=2.12-4.38 and OR=2.81, 95% CI=1.65-4.76).

Empirically Derived Dietary Patterns Using Robust Profile Clustering in the Hispanic Community Health Study/Study of Latinos

BACKGROUND

Latent class models (LCMs) have been used in exploring dietary behaviors over a wide set of foods and beverages in a given population, but are prone to overgeneralize these habits in the presence of variation by subpopulations.

OBJECTIVES

This study aimed to highlight unique dietary consumption differences by both study site and ethnic background of Hispanic/Latino populations in the United States, that otherwise might be missed in a traditional LCM of the overall population. This was achieved using a new model-based clustering method, referred to as robust profile clustering (RPC).

METHODS

A total of 11,320 individuals aged 18-74 y from the Hispanic Community Health Study/Study of Latinos (2008-2011) with complete diet data were classified into 9 subpopulations, defined by study site (Bronx, Chicago, Miami, San Diego) and ethnic background. At baseline, dietary intake was ascertained using a food propensity questionnaire. Dietary patterns were derived from 132 food groups using the RPC method to identify patterns of the general Hispanic/Latino population and those specific to an identified subpopulation. Dietary patterns derived from the RPC were compared to those identified from an LCM.

RESULTS

The LCM identified 48 shared consumption behaviors of foods and beverages across the entire cohort, whereas significant consumption differences in subpopulations were identified in the RPC model for these same foods. Several foods were common within study site (e.g., chicken, orange juice, milk), ethnic background (e.g., papayas, plantain, coffee), or both (e.g., rice, tomatoes, seafood). Post hoc testing revealed an improved model fit in the RPC model [Deviance Information Criterion DICRPC = 2.3 × 104, DICLCM  = 9.5 × 106].

CONCLUSIONS

Dietary pattern behaviors of Hispanics/Latinos in the United States tend to align by ethnic background for some foods and by location for other foods. Consideration of both factors is imperative to better understand their contributions to population health and developing targeted nutrition intervention studies.

Identifying the pattern of unhealthy dietary habits among an Iranian population: A latent class analysis

Background: An unhealthy diet is one of the most important risk factors for chronic diseases. The goal of this study was to use the latent class analysis (LCA) modeling to define unhealthy diet habits among an Iranian population. Methods: This cross-sectional study was conducted within the framework of Amol (North of Iran) cohort health study (Phase 1). The participants aged 10 to 90 years. All participants provided written informed consent. Latent class analysis was used to classify the participants of the study. All analyses were conducted by PROC LCA in SAS 9.2 software. Significance level was set at 0.05. Results: The mean age of the participants was 42.58±17.23 years. Four classes of individuals with different diet habits were identified using LCA modeling: class 1: individuals with healthy diet patterns (92.6%); class 2: individuals with slightly unhealthy diet habits (6.3%); class 3: individuals with relatively unhealthy diet habits (0.8%); and class 4: individuals with unhealthy diet habits (0.2%). Being female and alcohol consumption increased the odds of membership in latent classes 2,3, and 4 compared to class 1. Physical activity decreased the odds of membership in classes 3 and 4 compared to class 1. Conclusion: Overall, almost more than 7.4% of all participants had some degree of unhealthy dietary habits, and some variables acted as risk factors for membership in risky classes. Therefore, focusing on these variables may help design and execute effective preventive interventions in groups with unhealthy dietary habits.

The genomics of micronutrient requirements

Healthy nutrition is accepted as a cornerstone of public health strategies for reducing the risk of noncommunicable conditions such as obesity, cardiovascular disease, and related morbidities. However, many research studies continue to focus on single or at most a few factors that may elicit a metabolic effect. These reductionist approaches resulted in: (1) exaggerated claims for nutrition as a cure or prevention of disease; (2) the wide use of empirically based dietary regimens, as if one fits all; and (3) frequent disappointment of consumers, patients, and healthcare providers about the real impact nutrition can make on medicine and health. Multiple factors including environment, host and microbiome genetics, social context, the chemical form of the nutrient, its (bio)availability, and chemical and metabolic interactions among nutrients all interact to result in nutrient requirement and in health outcomes. Advances in laboratory methodologies, especially in analytical and separation techniques, are making the chemical dissection of foods and their availability in physiological tissues possible in an unprecedented manner. These omics technologies have opened opportunities for extending knowledge of micronutrients and of their metabolic and endocrine roles. While these technologies are crucial, more holistic approaches to the analysis of physiology and environment, novel experimental designs, and more sophisticated computational methods are needed to advance our understanding of how nutrition influences health of individuals.

The use of a dietary quality score as a predictor of childhood overweight and obesity

BACKGROUND

The use of dietary quality scores/indices to describe diet quality in children has increased in the past decade. However, to date, few studies have focused on the use of these scores on disease outcomes such as childhood obesity and most are developed from detailed dietary assessments. Therefore, the aims of this study were: firstly to construct a diet quality score (DQS) from a brief dietary assessment tool; secondly to examine the association between diet quality and childhood overweight or obesity; thirdly we also aim to examine the associations between individual DQS components and childhood overweight or obesity.

METHODS

A secondary analysis of cross sectional data of a sample of 8,568 9-year-old children and their families as part of the Growing Up in Ireland (GUI) study. Subjects were drawn from a probability proportionate to size sampling of primary schools throughout Ireland over the school year 2007-2008. Height and weight were measured by trained researchers using standardised methods and BMI was classified using the International Obesity Taskforce cut-points. The DQS (un-weighted) was developed using a 20-item, parent reported, food frequency questionnaire of foods consumed over the past 24 h. Adjusted odds ratios for overweight and obesity were examined by DQS quintile, using the first quintile (highest diet quality) as the reference category.

RESULTS

The prevalence of normal weight, overweight and obese was 75, 19 and 6% respectively. DQS ranged from -5 to 25, higher scores indicated higher diet quality in the continuous score. In analyses adjusted for gender, parent's education, physical activity and T.V. viewing, child obesity but not overweight was significantly associated with poor diet quality: OR of 1.56 (95% CI 1.02 2.38) in the 5th compared to the 1st DQS quintile. Findings from individual food items were inconsistent.

CONCLUSIONS

The findings suggest that diet quality may be an important factor in childhood obesity. A simple DQS developed from a short dietary assessment tool is significantly associated with childhood obesity.

Capturing changes in dietary patterns among older adults: a latent class analysis of an ageing Irish cohort

OBJECTIVE

Data-driven approaches to dietary patterns are under-utilized; latent class analyses (LCA) are particularly rare. The present study used an LCA to identify subgroups of people with similar dietary patterns, explore changes in dietary patterns over a 10-year period and relate these dynamics to sociodemographic factors and health outcomes.

DESIGN

The 1998 baseline and 2008 follow-up of the Cork and Kerry Diabetes and Heart Disease Study. Diets were assessed with a standard FFQ. LCA, under the assumption of conditional independence, was used to identify mutually exclusive subgroups with different dietary patterns, based on food group consumption.

SETTING

Republic of Ireland.

SUBJECTS

Men and women aged 50-69 years at baseline (n 923) and at 10-year follow-up (n 320).

RESULTS

Three dietary classes emerged: Western, Healthy and Low-Energy. Significant differences in demographic, lifestyle and health outcomes were associated with class membership. Between baseline and follow-up most people remained 'stable' in their dietary class. Most of those who changed class moved to the Healthy class. Higher education was associated with transition to a healthy diet; lower education was associated with stability in an unhealthy pattern. Transition to a healthy diet was associated with higher CVD risk factors at baseline: respondents were significantly more likely to be smokers, centrally obese and to have hypertension (non-significant).

CONCLUSIONS

LCA is useful for exploring dietary patterns transitions. Understanding the predictors of longitudinal stability/transitions in dietary patterns will help target public health initiatives by identifying subgroups most/least likely to change and most/least likely to sustain a change.

Evidence of sample use among new users of statins: implications for pharmacoepidemiology

BACKGROUND

Epidemiologic studies of prescription medications increasingly rely on large administrative health care databases. These data do not capture patients' use of medication samples. This could potentially bias studies of short-term effects where date of initiation may be inaccurate.

OBJECTIVES

To assess the extent of sample use among patients initiating statin therapy.

RESEARCH DESIGN

Retrospective cohort of patients who filled a first prescription for a statin after at least 6 months of statin-free period in 2007-2010. Low-density lipoprotein (LDL) values obtained within the 15 days preceding the first prescription were analyzed using a 2-component Gaussian mixture model to look for evidence of prior treatment.

SUBJECTS

A total of 26,033 statin initiators with at least 1 LDL laboratory result within the 15 days preceding the prescription fill.

MEASURES

Estimators for the proportion of patients filling a new prescription already on treatment.

RESULTS

Among 9256 patients filling a branded statin, LDL distribution was bimodal, consisting of 2 Gaussian distributions: one, which made up 13.4% of the total population, had much lower LDL values (mean=71.8 mg/dL) compared with the second (mean=148.0 mg/dL), suggesting drug use before first dispensed prescription. Among 16,777 patients filling a generic statin, LDL levels were substantially higher with no evidence of bimodality that would suggest prior sample use.

CONCLUSIONS

These results provide indirect evidence that the initial period of branded medication use may often be missed when using pharmacy claims data to define drug initiation. Further research is needed to examine approaches to better identify incident medication use when assessing short-term effects.