Purpose: Much concern has been expressed about the high levels of error that affect self-reported dietary intake assessment, to the point some have questioned if there is value in obtaining such data. Wearable measures relying on the passive capture of images and automation of food identification and portion size estimation based on those images have been posited as potentially less biased methods for characterizing intake. The utility of wearable technologies relies on whether they offer accurate measurements of intake, while accounting for cost, convenience, and associated resource considerations. We undertook a narrative review to examine progress made in advancing wearable methods of dietary assessment. 

Methods:  Relevant literature was identified based on systematic searches of PubMed, Web of Science, and Google Scholar. Synthesis of evidence focused on progress made toward steps in estimating dietary intake using images. 

Results: Automating dietary intake assessment using images from wearable devices entails a) participants appropriately wearing the camera and obtaining quality images; b) identifying images that contain foods c) assessing food purchasing, home availability, and preparation practices to inform nutrient estimation; d) segmenting foods on a plate or other surface; e) identifying foods; f) assessing portion sizes (before and after eating); and g) linking information about food identity, preparation and portion sizes with food composition databases to quantify food groups and nutrients consumed. Error can be introduced at each step. For example, images may be missed due to camera malfunction; images containing food may be misidentified; and specific foods and their portion sizes incorrectly identified. Although estimates of the magnitude of error differ depending on the study design and specific devices tested, they are not trivial. 

Conclusions: The goal of dietary assessment using wearable measures is to produce objective indicators of dietary intake. A narrative review indicated important advances have been made toward this goal and innovation continues toward automation of steps in the process of quantifying dietary intake. However, given current estimates of error and the fact that error is additive across steps, measures of intake currently yielded by device-based measures may be no more accurate than those from self-reported methods.

Purpose: Few cross-sectional studies examining diet and depression have used formal depression diagnoses as the outcome variable. We aimed to determine if overall diet quality was associated with prevalence of DSM-IV diagnosed mood disorders among a unique cohort of Australian adults followed up at three time-points in young to mid-adulthood. A secondary aim was to examine confounding effects using a wide variety of covariate measures.

Methods:  Participants from the Childhood Determinants of Adult Health study were followed up during 2004-06 (n=1,974, 50% male, age:  26-36 years), 2009-11 (n=1,527, 35% male, age:  31-41 years), and 2014-19 (n=1,195, 45% male, age:  36-49 years). A Dietary Guidelines Index (DGI) scores was calculated from food frequency questionnaire data at each time-point. A higher DGI score (range: 0-100) indicated better diet quality. The Composite International Diagnostic Interview was used to determine DSM-IV diagnoses of mood disorder (major depression or dysthymia) during the 12 months prior to each follow-up. Cross-sectional prevalence ratios (PR) and 95% confidence intervals (CI) were estimated using log binomial regression. Analyses were stratified by sex. Covariates included age, BMI, social support index, marital status, parenting status, education, occupation, smoking, physical activity, and usual nightly sleep duration.

Results: A 10-point increase in DGI score was cross-sectionally associated with lower prevalence of mood disorders among females at all time-points. However, the association was only statistically significant at the third follow-up (PR=0.73, 95% CI=0.56-0.94), and was attenuated after covariate adjustment (PR=0.92, 95% CI=0.73-1.17). Among males, better diet quality was associated with lower prevalence of mood disorder at the third follow-up (PR=0.68, 95% CI=0.50-0.92), but was also attenuated after adjustment (PR=0.88, 95% CI=0.65-1.19). Adjustment for social support in the final models attenuated the association for females by 57% (from 18% lower prevalence to 8%), and by 47% for males (from 22% to 12%).

Conclusion: Diet quality was not associated with prevalence of mood disorders after covariate adjustment. Social support was a strong confounder and may be an important variable that is not commonly measured or controlled for in studies examining the relationship between diet and depression.

Purpose: The distribution of energy intake and amount of energy consumed in a 24 hour period, or temporal dietary pattern, was previously developed using data-driven methods and associated with dietary quality. A pattern of three moderate-energy eating events spaced from morning to evening had the highest dietary quality compared with other patterns. This study determined the relationship of temporal dietary patterns with health outcomes.

Methods: The first-day 24-hour dietary recall data from 1,627 non-pregnant adults 20-65 years in the cross-sectional National Health and Nutrition Examination Survey, 2003-2006, determined the amount of energy intake (kcal), time of intake (min), and sequence of intake throughout the 24-hour day. Modified dynamic time warping, coupled with kernel k-means algorithm, clustered participants into four groups representing distinct temporal dietary patterns. Outcomes body mass index, waist circumference, fasting plasma glucose, hemoglobin A1c, triglyceride, HDL-C, total cholesterol, systolic and diastolic blood pressure, categories for obesity, type 2 diabetes, and metabolic syndrome were constructed from measures from the examination. Multivariate regression models evaluated the relationship of temporal dietary patterns and each outcome, controlling for potential confounders, energy misreporting, and adjusting for multiple comparisons and complex survey design (p<0.05/6).

Results: The temporal dietary pattern cluster with similar average energy intake at three main eating occasions from 8:00 to 23:00 including peaks averaging 175 kcal at 9:00, 13:00, and 19:00, had statistically significant and clinically meaningfully lower body mass index (p<0.0001) and waist circumference (p<0.0001) and 75% lower odds of obesity compared to three other clusters representing patterns with much higher average peak energy of 500 kcal at 13:00 (OR: 4.41; 95% CI: 2.48, 7.86), 530 kcal at 18:00 (OR: 5.32; 95% CI: 2.80, 10.14), and 550 kcal at 20:00 (OR: 6.72; 95%CI: 3.91, 11.58).

Conclusion:  Temporal dietary patterns differentiate clusters by body mass index, waist circumference, and odds of obesity among U.S. adults, providing unique evidence of the importance of timing of dietary intake throughout a day to health and supporting previous findings of higher dietary quality among those with similar temporal dietary patterns. Temporal dietary patterns hold promise for the development of future interventions and dietary guidance.

Purpose: Joint and Individual Variance Explained (JIVE)¹ is an alternative to principle component analysis (PCA) that highlights variability shared across different datasets collected on the same individuals. We have applied this method in a novel context: datasets consisting of food frequency questionnaires administered at different ages.  By construction, the components summarizing joint variability encode the aspects of the diet that are predictive of diet observations at later ages.  We hypothesize that this method will produce dietary patterns that are  more predictive of health outcomes than those derived by PCA applied separately to the data collected for each age group.

Methods:

We test the method using food frequency questionnaires administered  at ages 3.5, 7,  and 11  as part of the Auckland Birth cohort study.   466 individuals have questionnaire responses and health outcomes (BMI z-score, diastolic blood pressure and systolic blood pressure) at all time points.  As with principle components, the scores are computed as a linear combination of the original variables, and these loadings can be used to interpret the patterns.  The score can also be partitioned into a contribution from each age group.  We use the age 3.5 score contributions for the first 2 JIVE components to predict health outcomes at ages 7 and 11 using linear regression, using the same health outcome measured at age 3.5 as a covariate.  The standardised regression coefficients are compared to those found when the predictors are taken to be principle component scores of the age 3.5 data. 

Results/findings:  The first age 3.5 JIVE score and age 3.5 PCA score are both significant (p< 0.05) predictors of BMI at age 7 and 11.   However, the standardized regression coefficients for the JIVE scores were 20% and 45% larger (respectively), indicating stronger relationships.  The only significant relationship between diet scores and blood pressure was for the age 7 systolic blood pressure and the 2^nd PCA score. 

Conclusions: By focussing on aspects of the diet that predict consumption at later ages, JIVE produces dietary patterns that have a stronger relationship with BMI.

¹Lock et al (2013), Ann Appl Stat 7:523-542. 

Purpose: Interrelated disordered eating attitudes and behaviors may exist on a single dimension. In this study, we examine the appropriateness of creating a global score from five disordered eating attitudes and behaviors, examine the track the fit over time, and examine its convergent validity.

Methods: Five disordered eating attitudes and behaviors were assessed among 1492 participants in a longitudinal cohort (Project EAT, age 11 to 18 at 1998-1999). The appropriateness of creating a global score was examined by confirmatory factor analysis. To examine whether the individual variables functioned differently in relation to the overall latent construct across time, two models were compared: one requiring indicator-level factor loadings to remain equivalent across three-time points (baseline [EAT-I], five-year follow-up [EAT-II], and 15-year follow-up [EAT-IV]), and the second allowing the factor loadings to vary over time.  Convergent validity of the global score was examined by Pearson correlation with body satisfaction, self-esteem, depressive symptoms, and BMI. The correlation was compared across three time-points (EAT-I, II, and IV). 

Results: The use of five disordered eating attitudes and behaviors in creating a global score was supported by the goodness of fit indices for a single factor structure (standardized loadings: 0.60-0.87, 0.67-0.89, 0.59-0.77 at EAT-I, II, and IV respectively), which were consistent over time. As expected, the global score negatively correlated with body satisfaction, self-esteem, and positively correlated with depressive symptoms and BMI over time (all p < 0.01).

Conclusions: The five disordered eating attitudes and behaviors can be viewed on a single dimension.  The five-point global score of disordered eating attitudes and behaviors is a stable analysis tool to measure the severity of disordered eating attitudes and behaviors in population-based studies.

PURPOSE: To develop the first gold standard tool to measure household cooking environments.

METHODS: The Home-Cooking Environment and Resource Inventory Observation Form (Home-CookERI™ OF) was developed in Australia in 2019 as an 80-item online Qualtrics™ survey. Items included domestic spaces/resources for the storage, disposal, preparation and cooking of food or non-alcoholic beverages. Home-CookERI™ OF was assessed for face and content validity by 13 experts (i.e. dietitians, nutrition researchers, qualified chefs, a food technology teacher and kitchen designer) and 14 lay people. They considered 95% of items to be clear and 99% to be relevant. In 33 different homes, a pair of research dietitians (i.e. rater 1, rater 2) completed Home-CookERI™ OF. Raters searched for each item before recording in Home-CookERI™ OF the presence/absence of each item. To prevent data contamination, home occupants were instructed to only assist with locating non-visible items when asked. Furthermore, the second rater could not see or hear the first rater complete Home-CookERI™ OF. Inter-rater agreement (IRA) was determined using percent agreement [%] and Cohen’s Kappa [κ].

RESULTS: IRA was ≥80% for 74/80 items (93%) and ≥69% for all items. κ was: perfect [κ=1.0] for 23 items (28.75%), near perfect [κ=0.81-0.99] for 14 items (17.5%), substantial [κ =0.61-0.8] for 18 items (22.5%), moderate [κ=0.41-0.6] for 8 items (10%), fair [κ=0.21-0.4] for 6 items (7.5%), slight [κ=0.1-0.2] for 3 items (3.75%), and chance or worse [κ≤ 0] for 8 items (10%). Items with κ≤0.2 were either highly common or uncommon items. Hence, item disagreement in one or a few households substantially impacted κ.

CONCLUSION: Overall, Home-CookERI™ OF has established face/ content validity and is a reliable tool for researchers to measure Australian household cooking environments via direct observation. A self-completed Home-CookERI™ form for home occupants is being developed to have applications in nutrition epidemiology and nutrition interventions.