MEASURES REGISTRY USER GUIDE
The following case studies have been designed to illustrate considerations influencing the selection of the most appropriate measure(s) for a given study based on the research aim/question, study design, and other characteristics. The focus in terms of measures is on methods more broadly (e.g., 24-hour recalls, food frequency questionnaires) rather than specific tools falling under the umbrella of each of these methods.
Case Study 1: Examining Influences on Diet Among Population Subgroups
A project team wishes to estimate average intake and main sources of dietary guidance-based food groups among children of varying ages, differentiated by sociodemographic characteristics. This is a surveillance effort with a large, cross-sectional sample designed to enable estimation among various subgroups. The intent is to capture all foods and beverages consumed across settings.
The dietary behavior of interest includes intake of multiple dietary components, including food groups such as fruits, vegetables, and dairy products. Thus, a measure that enables quantification of the total diet with the least bias possible is needed, suggesting a short-term measure (24-hour recall or food record). The focus on capturing various dietary components rules out screeners, which are also not recommended for estimating mean intakes among populations due to bias (the same is true of food frequency questionnaires).
Given the varied ages, proxy reporting will be needed for younger children.
The team has several possible options, each of which has specific strengths and limitations:
- Interviewer-administered recalls are possible, although they are expensive and intensive in terms of coding. Web-based self-administered recalls are another possibility. Proxy-assistance or other types of assistance or training may be needed to successfully implement this method.
- A food record also is possible, but parental assistance would be needed to address issues associated with literacy, numeracy, cognitive abilities, and attention span. Mobile device image-based food records are a possible method for older children.
- In studies combining different approaches, the implications of potential mode (i.e., paper- versus mobile device-based record) effects should be considered.
Because diet is the dependent variable and the team is interested in estimating intake in relation to sociodemographic variables, they will need to weigh the potential for differential biases in reporting (e.g., children with differing body weights or other characteristics may misreport differently). Working with a statistician to identify the most appropriate analytic approaches is recommended.
Given that only mean usual intakes (and not distributions of usual intake, for example, for estimating the proportion with intakes of fruit that meet recommendations) are required, the team will not need repeat measures of the short-term measures to account for day-to-day variation in diet.
Variation: If the team wishes to estimate distributions of usual intake so they can assess proportions of the sample who meet food group recommendations or other characteristics of the distribution beyond the mean, they will need to administer replicate recalls or records for at least a subsample. Two non-consecutive (to avoid autocorrelation or leftover effects) replicates or repeats among a representative subsample are typically adequate, but additional repeats may be required for foods that are more episodically consumed in the target population.
Case Study 2: Examining Diet Quality and Markers of Disease
A project team sets out to elucidate the relationship between diet quality and proximal markers of disease, such as blood glucose levels or markers of inflammation, among adolescents. The study is a prospective cohort, with a large sample. The intent is to capture all foods and beverages consumed across settings to enable the characterization of diet quality, for example, using a diet quality index.
The dietary behavior of interest includes intake of multiple dietary components—in other words, the total diet. Thus, a measure that enables quantification of the total diet is needed. This suggests a leaning toward short-term measures. The focus on capturing the total diet rules out screeners.
Given that the age group is adolescents, self-administration is possible, though time and burden should be considered in terms of the impact on data quality.
Feasible methods include self-administered 24-hour recalls, food records, or food frequency questionnaires. Interviewer-administered recalls would likely be cost prohibitive, and unless the food record involves technology-enabled automated coding, manual coding would require considerable resources. This narrows the choices to a self-administered 24-hour recall, mobile device-based (or otherwise automated) food record, or a food frequency questionnaire. A combination of methods could also be used.
Diet is the independent variable, or an exposure. As in all studies, strategies to arrive at the least-biased possible estimates of amounts of intake. Analytic techniques to mitigate error in such cases include regression calibration, which uses data from a reference measure from a calibration substudy to reduce error in the intake data collected using the main dietary measure for the study. For example, in a study using 24-hour recalls, biomarker data could be collected from a subsample to allow for calibration of the recall data, whereas in a study using food frequency questionnaires, 24-hour recall data may be collected from a subsample to allow for mitigation of bias in the frequency data.
In such studies, food frequency questionnaires traditionally have been used, at least with adults. However, with web-based tools, it is possible to use recalls or records either as the main instrument (which is desirable given that short-term data have been shown to be less biased than frequency data) or as a reference instrument for calibration purposes. Recalls and records also offer the advantage of greater comparability across populations and studies given that there is no finite food list. Food frequency questionnaires need to be tailored to the population of interest and this can limit comparability as well as the potential for pooling and harmonization across studies. Further, food frequency questionnaires may be cognitively difficult for children depending on their age.
Case Study 3: Examining Implications of Modifications to Foods Offered for Sale in Vending Machines Within an Institution
A project team wishes to assess intake of sugar-sweetened beverages and alternatives before and after changes to vending machine policies in an institution, such as a school, university, workplace, or recreation center. This is an intervention study involving swapping out of energy-dense choices within vending machines for more nutrient-dense options, including replacing sodas and energy drinks with water. Given a systems perspective, the intent may be to capture intake across settings to allow the project team to account for trade-off effects. For example, reduced consumption of sugary beverages at school may be offset by increased consumption in other settings.
The dietary behavior of interest could be conceptualized narrowly as intake of snacks and beverages, or broadly as the total diet. This would enable characterization of how the intervention relates to changes (if any) in sugar intake overall or diet quality more holistically. For example, reductions in soda consumption may be offset by increases intake of juice or possibly other foods or beverages.
In addition, intake could be conceptualized either as quantitative estimates, requiring querying amounts consumed, or frequency of consumption of energy-dense snacks and beverages.
Depending on the target population within the institution of interest, investigators will need to consider whether self-reporting is possible. This will affect which measures can be selected. For example, self-administration is not possible for younger children.
If the project team chooses a narrower focus, screeners could be used, which would reduce team and respondent burden but increase bias. This bias is less of an issue for items like sugar-sweetened beverages than for other dietary components (e.g., sugars, fruits and vegetables) that are distributed throughout many contributing food and beverage sources. Screeners may be difficult for children, depending on cognitive abilities to average intake over a long period of time.
If the team chooses a broader focus, a more comprehensive tool, such as 24-hour dietary recalls, food records, or food frequency questionnaire, is needed as such a tool allows interrogation of different aspects of the diet.
In this project, dietary intake is the outcome, and the study design is an intervention. As a result, respondents could potentially report differently after the intervention due to exposure to the intervention itself. However, given the environmental focus of the intervention (as opposed to nutrition education or counseling about reducing intake of energy-dense foods), this is unlikely unless the intervention is accompanied by an intensive marketing campaign. Nonetheless, the project team could complement the intake data with sales data from the vending machines. However, these data would be limited to the single setting within which the vending machines were modified, not to changes in consumption behaviors more broadly.
Case Study 4: Assessing the Effects of a Home-based Obesity Prevention Program on Pre-School Children’s Dietary Behaviors
A project team is interested in examining the effects of a home-based obesity prevention program on preschool children’s meal and snacking behaviors. This is an intervention study, with control and intervention groups. The control group receives minimal exposure to the study team (including nutritionists), whereas the intervention group receives intensive programming related to healthy eating patterns, parenting, and other potential influences on dietary intake among children and the family as a whole.
The diet behaviors of interest could be conceptualized as patterns, in terms of meals and snacks (e.g., frequency of between-meal snacks). In this study, behaviors could be conceptualized in different ways. Depending on the information desired, the team could consider different measures:
- The diet behaviors of interest could be conceptualized as patterns, in terms of meals and snacks (e.g., frequency of between-meal snacks). Such behaviors could be measured by a questionnaire, to be completed by the parent or other proxy reporter, querying these behaviors specifically.
- Alternatively, with more intensive measures such as 24-hour recalls and records, the foods consumed at different meals and snacks as well as contextual factors, such as who the child ate with and whether he or she was using an electronic device (e.g., eating a snack in front of the television) could be considered. With a frequency questionnaire, the foods consumed over a period of time (e.g., before or after the intervention) could be assessed, but contextual factors could not.
- Finally, home observation could be used, with trained interviewers documenting meal and snack behaviors during times at which they are present in the home. However, this would be intensive and burdensome.
Given that the target population is preschool children (and their parents and families), any questionnaires of behaviors more generally (e.g., frequency of snacking) or intake more specifically would need to be completed by proxy-reporters, with necessary considerations regarding error in reporting.
With this intervention design, parents may report their children’s diets differently as a result of being exposed to the intervention than they did at baseline. Thus, complementary measures, such as in-home observation, may be beneficial to provide a means of corroborating (or not) changes in meal and snacking patterns from pre to post.
Case Study 5: Assessing Differences in Diet Quality Among Subgroups with Different Rates of Obesity
A project team would like to assess whether diet quality differs among adolescent children from different racial and ethnic subgroups with varying rates of obesity. The study makes use of NHANES data.
In this case, the dietary measure is predetermined because the team is using NHANES dietary data, which are collected using 24-hour recalls. However, the case highlights that the team will need to take various considerations that must be taken into account when comparing the dietary intake of groups differentiated by factors potentially related to body weight. Because body weight is strongly associated with misreporting, any such analyses must be undertaken and interpreted cautiously.
Case Study 6: Evaluating the Effects of Calorie Labeling within a Given Institution on Energy Intake
A project team wishes to examine whether adolescents change their energy intake when menu calorie labeling is instituted, for example, within a cafeteria. This case is intended to illustrate the caveats related to measuring calorie intake using self-report data.
Considerations and Measures Selection
Given the known biases in self-report data, it is generally recommended that they not be used to arrive at absolute estimates of energy. However, in this case, intercept surveys could be used before and after the introduction of calorie labeling to assess changes in purchases. The team also could examine sales data to make these comparisons.
Further, the team could use dietary intake data to assess implications for diet quality more broadly. For example, if calorie labeling was effective in stimulating changes, did it encourage healthier choices overall, such as the consumption of more fruits and vegetables or higher diet quality, rather than merely encouraging lower-calorie choices?
Case Study 7: Assessing Children’s Food Preferences in Relation to Advertising
A project team sets out to assess the preferences of preschool children for fruits, vegetables, and snacks after they were exposed to advertising of different types of foods, with and without cartoon characters and celebrities.
The study is a cross-sectional experimental design with control and comparison groups. Recruitment and study collection occur in an early care and education center. Children are invited to play a video game on an iPad that features or does not feature cartoon character or celebrity endorsements related to food. The diet behavior of interest, food preferences, is the dependent variable.
Considerations and Measures Selection
Given the age of the children, a questionnaire regarding food preferences is not feasible, though they could be prompted to choose between photos of different types of foods, indicating the one they would choose out of dyads differentiated by energy- and nutrient-density.
Alternatively, the children could be offered different snacks (fruit, vegetables, and snacks, some of which were highlighted by the cartoon characters or celebrities). The team could weigh the snacks served as well as leftovers (i.e., plate waste) so that intake can be quantified as an indicator of preferences.
Associations between exposure to cartoon characters and celebrities and food preferences could be examined by comparing preferences or intake between the experimental and control groups. With observed food intake, issues regarding differential biases in reporting of intake are not an issue.
Case Study 8: Assessing the Impact of a Body Image-Based Program on Adolescents’ Dietary Behaviors and Intake
The primary aim of this study is to assess changes in diet-related behaviors and intake before and after the implementation of an intervention to improve body image. This case is included to illustrate the potential impact of social desirability on youth’s reporting of dietary intake, and why this should be considered in analysis of results.
Considerations and Measures Selection
Because older children and adolescents have greater concerns surrounding body image as compared to younger children, it is important to consider the role of social desirability in their reporting of diet-related behaviors and intake. This may be especially salient when they are queried about body image and weight-related constructs. To account for this source of bias, researchers may assess social desirability using established measures,35,47–49 and stratify diet-related variables by social desirability categories. If youth with high social desirability significantly differ from youth with low social desirability on reporting of diet-related behaviors and intake, then this must be considered in the analysis of the intervention’s effectiveness.