MEASURES REGISTRY USER GUIDE
Overview of Individual Diet Measures
Measures of dietary behavior differ in terms of the specific dimensions or facets of behavior they are able to capture. As noted previously, for most research applications with relevance to characterizing dietary behavior, the aim is to capture usual or habitual dietary intake. Quantifying frequency of consumption as well as the amount consumed provides the capacity to link to databases to estimate intake of foods, food groups, nutrients, and other dietary components (though attention must also be paid to issues such as the currency and comprehensiveness of such databases).
Carefully considering measures to be used for all variables and collaborating with statisticians early in the process can help ensure that data are collected, analyzed, and interpreted in a way that leads to the most robust evidence possible for informing obesity prevention.
Addressing research questions may require querying the total diet (possibly including vitamin and mineral supplements) or consumption of specific dietary components, such as fruits, vegetables, sugar-sweetened beverages, fat, fiber, or sugar. Further, there may be interest in characterizing diet quality or dietary patterns more holistically, which requires data on the multiple foods and beverages that make up eating patterns. With respect to specific dietary components, it is salient to consider whether those of interest are consumed regularly by most members of the population of interest or are episodically-consumed. This has implications for assessment in terms of the need to capture multiple days of consumption or a long enough period of time such that consumption days are included. Additionally, depending on the research question, there may be a need for attention to temporal patterns in consumption at the level of meals, days of the week, seasons, and across the lifecycle. Finally, contextual factors, such as where foods or beverages are obtained, with whom meals are consumed, where meals are consumed (i.e., at home versus away from home), and the use of electronic devices while eating may be relevant to a given research question.
In most research relevant to dietary behaviors, self-report measures are used. This is because it is not possible to objectively assess usual intake in free-living individuals. Depending on the measure, self-report data can also capture temporal patterns and contextual factors that are of interest in many studies. We briefly describe available objective measures and their utility before reviewing commonly used self-report measures of intake, along with considerations related to technological innovations and highlights of specific considerations related to children.
Objective Measures of Dietary Intake
The identification of biomarkers and their application in diet assessment (Box 3) are active areas of inquiry.58–61 Recovery biomarkers62 are recognized as objective measures of true intake for energy and a few nutrients, including protein, potassium, and sodium. Energy intake is estimated using the doubly-labeled water (DLW) technique63 whereas the collection of 24-hour urine samples is used for protein,64 potassium, and sodium.65,66 The collection of recovery biomarkers is costly and burdensome for researchers and respondents and not feasible for most studies. Further, recovery biomarkers (and other types of biomarkers) do not provide information on what individuals actually eat and drink, nor contextual factors of salience to understanding how to intervene to shift eating patterns.67 Their usage for research in which the aim is to examine eating patterns, influences on those patterns, and the potential value of interventions for altering them is thus limited. Recovery biomarkers are, however, extremely useful for validation studies62–64 to assess error in self-report measures, and they can also be used to reduce error in self-report data when they are available for a study subsample.
Box 3. Overview of Biomarkers Relevant to the Assessment of Dietary Intake Behavior
- Recovery biomarkers: Biologic products that are directly related to intake and provide unbiased estimates of true intake.62
- Concentration biomarkers: Biologic products reflecting the concentration of a chemical or compound in blood, urine, or tissues after metabolism.62 Indirect measures of intake.
- Predictive biomarkers: Biologic products that have a stronger relation with intake than concentration biomarkers but do not provide unbiased estimates of true intake. For example, predictive biomarkers have been proposed for sugars (urinary sugar)68,69 and whole grains (plasma alkylresorcinols).70 Such biomarkers are postulated to improve estimation of associations between diet and health.
- Metabolomics: Methods used to identify metabolites in biological fluids (e.g., blood, saliva, urine) that are produced through metabolism of foods, as well as toxins and medicines.71 Pinpointing metabolites that vary by dietary pattern has been proposed as a means of advancing understanding of diet and health relationships, as well as discovering novel biomarkers for intake of foods, such as red meat or vegetables.71,72
Several relatively large recovery biomarker-based validation studies have been conducted and the data pooled for analyses,65,66 lending insights into error in commonly used measures such as 24-hour recalls and food frequency questionnaires. The results of such studies are briefly summarized in Section 6.
Concentration62 and predictive68,69 biomarkers represent additional classes of indicators increasingly used in the measurement of dietary intake.59,60 However, these do not have the same direct link with intake as recovery biomarkers and do not represent markers of true intake. These biomarkers can be useful, however, in combination with self-report measures in studies requiring the assessment of dietary intake; this is an area of ongoing research and discovery. Further, metabolomics has been recognized as a promising area in nutrition research, with the potential to lead to the discovery of novel biomarkers for intakes of foods, as well as to enhance understanding of how diet influences disease.71,73 Other innovations include the application of spectroscopy to dietary assessment, with the development of techniques to measure skin carotenoid status as a biomarker of intake,74,75 for example. Research with children has suggested high concordance between skin and serum carotenoids, suggesting that skin carotenoids could be used as a marker of fruit and vegetable consumption.74,76 Skin carotenoids have been used in combination with digital photography-based measures of fruit and vegetable intake to assess the influence of an intervention on children’s intake.77,78 Although these innovations are promising in terms of improving dietary assessment in the future, current research continues to rely primarily upon self-report.
Observation represents an additional objective measure of dietary behavior that can be useful for documenting true intake for comparison to self-report. Indeed, observational assessments of children’s eating and diet-related behaviors are often used to indicate objective truth for the evaluation of other dietary assessment methods.79 Researchers either observe participants’ dietary behaviors directly (e.g., videotaping child’s food selection, sitting in a classroom while children eat lunch) or indirectly (e.g., discreetly weighing food containers before and after consumption to calculate precise measurement of food eaten). Importantly, observation can allow measurement of misreporting of foods (unlike the recovery biomarkers described above). However, for situations in which interest is in intake for the purpose of understanding usual diet and eating patterns, the application of observation is limited.
Self-Report Measures of Dietary Intake
Commonly used self-report measures of intake include 24-hour dietary recalls, food records/diaries, food frequency questionnaires, and brief instruments (often referred to as screeners). Each tool has advantages and disadvantages,80 though it should be noted that some of the traditional limitations of tools have been addressed in the last decade or so due to technological innovations in dietary assessment described briefly in this section. Despite such innovations, all self-report measures capture intake with error; the type and extent of error depends on the tool and its characteristics (Section 6). Choosing the best possible measure for the given research application and target population and using appropriate statistical methods can help to reduce this error and its effects on study findings.
The National Cancer Institute’s Dietary Assessment Primer provides a thorough overview of the characteristics of each of the self-report measures discussed. Consult the Primer for details about the measures and recommendations for their appropriate usage and analysis for measuring dietary behavior.
A means of categorizing diet assessment tools is whether they assess short-term or long-term intake. Tools assessing intake over the short-term include 24-hour recalls and records/diaries, whereas those assessing intake over a longer period include food frequency questionnaires and screeners. Recalls and records/diaries are intended to capture intake over a day or a number of days. With food frequency questionnaires and screeners, respondents are prompted to report on their usual intake of a list of foods and drinks, aiming to estimate the frequency of consumption of foods and beverages over a period such as a month or a year, perhaps also with queries regarding typical portion size.
To address challenges in assessing dietary intake among children, multiple methods may be used for data capture. For example, records are sometimes kept to assist with completing recalls, as highlighted later in this section. This is distinct from the use of data from different measures in analysis, which is mentioned in Section 8 and in the Case Studies (Section 9).
With all self-report methods, children may misreport intake because of social desirability biases.81 To consider the potential impact of social desirability on intake in data analyses, a study may include a measure of this bias in conjunction with a measure of intake to characterize and adjust for this source of error.
Method: 24-Hour Dietary Recalls
Example tools: Automated Multiple-Pass Method (AMPM), 82–84 Automated Self-Administered 24-Hour Dietary Assessment Tool (ASA24),52 Nutrition Data System for Research (NDSR),85 various paper-based versions.
The 24-hour dietary recall is aimed at capturing a comprehensive and detailed accounting of all foods, beverages, and in some cases, supplements, consumed on a given day.
Regarding usual intake, a single recall among a group can provide an estimate of mean usual intake, but for estimates related to the distribution of usual intakes (e.g., prevalence below/above a threshold), it is necessary to collect repeat recalls from at least a subsample to enable accounting for day-to-day variation in intake. For episodically consumed dietary components (i.e., components such as dark-green vegetables that are consumed irregularly by most persons in the target group), it may be necessary to collect additional replicate recalls to achieve a sufficient number of recalls that include that food or nutrient. For non-episodically consumed components (i.e., components such as refined grains or added sugars that are consumed regularly by most persons in the target group), this is not the case.
Traditionally, 24-hour recalls have been administered by an interviewer. Multiple-pass methods are used to improve accuracy1 and may be implemented using computerized systems. For example, the Automated Multiple-Pass Method (AMPM), developed by the U.S. Department of Agriculture (USDA), is a computerized method for interviewer-administered recalls that employs five steps to enhance complete recording as well as reduce burden for respondents.82–84 The steps include a quick list, which is a “mind dump” of all foods and drinks consumed the prior day; a probe for forgotten foods; a time and occasion pass that organizes the foods and drinks according to eating occasion; a detail cycle that probes for details, including how the food or beverage was prepared, the amount consumed, and anything added; and a final probe for anything else consumed but not yet reported. For the reporting of portion sizes, aids are usually used. These may include common household items, such as measuring cups and spoons and pictures or food models. In relation to true intake determined by observation, data collected using AMPM have been found to be relatively accurate for energy, protein, carbohydrate, and fat among men regardless of body mass index.86 AMPM is the method used to collect dietary intake data within What We Eat in America, the dietary component of NHANES.2 Other systems, such as the University of Minnesota’s Nutrition Data System for Research (NDSR),85 also employ multiple-pass methods to enhance the completeness of recalls. AMPM and NDSR offer the advantage of automated coding based on details provided for each food and beverage, eliminating the need for manual coding. Recalls can also be collected by interviewers using paper or computers or mobile devices, with multiple passes recommended. In this case, manual coding of each item reported is needed to enable linkage to a food composition and other relevant databases. This is labor- and time-intensive and thus, costly.
In addition to details on foods and beverages, 24-hour recalls can provide insights into patterning of food intake, such as the consumption of meals and between-meal snacks, as well as the distribution of food intake across the day. Probes can be included to capture contextual factors, such as where meals were eaten (e.g., home, school, fast food restaurant) and with whom, the source of the major ingredients for each item consumed, and the use of electronic devices during the consumption of meals and snacks. The 24-hour recall methodology may also integrate the reporting of vitamin and mineral supplement intake. This may be a separate module that follows the food and beverage recall and prompts respondents to report supplements taken the previous day, along with pertinent details such as the brand, dosage, and amount taken.
Sources of error in 24-hour recall data include imperfect short-term memory, inaccuracies in portion size estimation, and social desirability biases that may contribute to misreporting of some foods and beverages. Recalls have been shown to capture dietary intake with less bias than do food frequency questionnaires (Section 6). Thus, they are recommended for various applications, including those in which the research question relies on quantitative estimation of intakes among a population or subpopulation.
Barriers to the administration of recalls, particularly in large-scale research initiatives, have traditionally related to the significant cost for trained interviewers, as well as coders in circumstances in which coding is not automated. However, technological innovation in dietary intake has led to the development of web-based self-administered recall tools that eliminate the need for trained interviewers and coders. For example, in the United States, the Automated Self-Administered 24-hour Dietary Assessment Tool (ASA24)52 has been developed and is freely available to researchers, enabling the collection of recall data in a range of studies. ASA24 adapts the AMPM, with modifications made to facilitate self-administration (e.g., respondents first report meal occasions and related details, such as time and location, move on to report the foods and beverages consumed at each meal, and then complete a detail cycle, followed by forgotten foods). ASA24 provides the opportunity to collect data across studies using the same tool, lending to comparability of data with the potential for building a stronger evidence base that can be synthesized to arrive at important conclusions regarding diet and other factors as well as the potential for interventions to affect diet. With similar tools adapted (ASA24-Canada and ASA24-Australia) or developed (e.g., MyFood24 in the UK53) elsewhere, the potential for standardized data can extend beyond borders, facilitating pooling of data or cross-country comparisons.
ASA24 was initially modelled on a recall system developed for children.52,87 However, evaluative efforts with technology-based recalls, such as ASA24, among children are thus far limited88–90 and have not clearly identified the age at which children can complete a self-administered recall independently. Prior research has suggested that children may be able to independently complete recalls beginning at the age of 10 years91 and in NHANES, independent reporting is used beginning at the age of 12 years.2 Programs such as ASA24 may help to motivate children to report their intake due to their interactive nature; however, the multiple steps in completing recalls may lead to boredom or fatigue regardless of the use of technology. Evaluation of ASA24 among adults through an observational feeding study has shown that it performs well in relation to true intake and interviewer-administered recalls.92 It has also been shown to be feasible for use in large-scale community-based data collection among adults.93 Nonetheless, given the unique considerations in assessing diet in children, further evaluation is needed.
Efforts to enhance methods of administering 24-hour recalls using technology are being complemented by those to improve portion size estimation among children.38 Given that misreporting of portion size can be an important contributor to error, these are likewise important advances and an area for ongoing research.
In summary, 24-hour recalls can be useful for collecting dietary intake among children, but careful attention should be paid to administration to collect the most accurate data possible (Box 4).
Box 4. Using 24-Hour Recalls with Children
- For interviewer-administered recalls, interviewers must be trained, experienced, and follow protocols. Data collectors should be familiar with common foods and beverages consumed by children and the details needed to code them.
- For technology-based self-administered recalls, the tool should be intuitive for and appealing to children. Testing with the target group is critical to ensure feasibility.
- When using self-administered recalls, training prior to the collection of study recalls may be useful. For example, completing a practice recall in the presence of a researcher can help a child to learn the flow of the program. Study staff should be very familiar with the tool and able to provide necessary supports.
- For young children who are unable to report their own intake, proxy reporting is necessary. The involvement of children is important as they become more independent and parents or other caregivers are not present for all eating occasions. For older children, the involvement of a proxy may be viewed as an intrusion.94 NHANES uses proxy reporting for children younger than age 6 years, proxy-assisted recalls for children ages 6 to 11 years, and self-report for those ages 12 years and older.2
- Young children have limited concepts of time as well as limited ability to recall prior food consumption. Recency appears to affect children’s ability to accurately report. As a result, collecting data for the past 24 hours starting from the time at which the recall begins may be preferable to reporting for the prior day, midnight to midnight.95 ASA24, for example, provides an option for reporting for the past 24 hours.
- The use of neutral probing may help to avoid social desirability bias. A measure of social desirability bias may be used to assess this source of error.
- Probing for details that children are unlikely to know (e.g., what oil was used to prepare the food?) may lead to lower accuracy. In interviewer-administered recalls, interviewers should be trained to probe for only the information necessary to code the item. However, encouraging recollection of the context for food consumption (e.g., meal occasion, location) may be helpful in reducing errors in recall.
- Portion size estimation is very difficult for children (as with adults). Prior training may be helpful, but this remains a challenging area within dietary assessment.37
- Boredom and fatigue are possible with the multiple steps involved in completing recalls, as well as in studies involving the completion of multiple recalls.31 In studies involving multiple recalls, training effects are also possible, with declining quality of data as the number of recalls increases.
- In some situations, it may be possible to obtain complementary information from other sources, such as school food services for foods and beverages served on the recalled day. The use of food and beverage descriptions, portion sizes, and USDA standard recipes can allow for more precise coding and reduce probing for food details and portion sizes.
Method: Food Records/Diaries
Example tools: Technology Assisted Dietary Assessment (TADA),54-55 various paper-based forms
Similar to 24-hour recalls, food records or diaries (referred to as records subsequently, for simplicity) are intended to capture a detailed account of all foods, beverages, and possibly, supplements consumed on one or more days. Records are often kept for a period of one, three, or seven days. The distinction between recalls and records is that with a recall, the respondent reports (i.e., recalls, relying on memory) what was consumed yesterday (or over the past 24 hours) whereas with a record, the respondent keeps track of (i.e., records in real time) what he or she consumes. Typically, respondents are given a recording form along with instructions prompting for specific details, such as how each item consumed was prepared. The completion of food records requires literacy and numeracy. Portion size can be estimated using household measures, pictures, or other aids, or respondents may be requested to weigh all items using scales or volume measures. In some studies, research staff review the completed records with participants to fill in missing details. Multi-day weighed records are sometimes referred to as a gold standard for the evaluation of other dietary measures. However, given that all self-report methods are affected by error,65,66,96 it is inaccurate to refer to any such methods as gold standards.
Similar to recalls, data from records can provide insights into behaviors such as meal patterns and snacking and contextual factors such as where meals were eaten and with whom, and other activities during consumption, such as television watching. Unlike recalls though, data collected using records can be affected by reactivity in that individuals may change their eating behavior in response to tracking or monitoring. For some intervention studies, this form of self-monitoring and its implications for eating patterns may be the desired effect of the use of food records. However, in studies aiming to capture usual intake, reactivity is a source of error. As with recalls, data collected using records can be affected by inaccuracies in portion size estimation (if estimated rather than weighed), and social desirability biases likely contribute to misreporting of some foods and beverages.
Food records have also been affected by technological innovation with the advent of mobile device-based records, which may rely on images taken by the respondent before and after consumption of foods or beverages.97 Records require a significant coding effort. Thus, efforts are underway to automate coding using image-based assessment and image recognition. As capabilities related to image processing and recognition continue to evolve, tools such as the Technology Assisted Dietary Assessment (TADA)54,55 have the potential to shift methods of diet assessment with adolescents, who increasingly have access to mobile devices such as smartphones. Such advances may be useful for improving children’s engagement in dietary assessment, with potential benefits for the accuracy of the data collected. However, ongoing monitoring of the diet using a program on a mobile device may change the dietary patterns that are of interest in a study.
- Some researchers have used record-assisted recalls in an attempt to enhance accuracy of data collected from school-age children.
- In such studies, children are instructed to complete a written food record/diary.
- The record/diary is used to probe for food and beverage details, portion sizes, and forgotten foods and beverages during the subsequent completion of the recall.126–130
- An evaluation of this approach found that recalls may not add significantly to the accuracy of records, and that the use of records as a memory cue should be considered in the context of the additional burden imposed, as well as the potential to elicit reactivity 131
- Further research to better understand the utility of multiple methods for data capture among children is warranted.
As with recalls, various considerations should be accounted for in using food records with children Box 5 to maximize the accuracy of the data.
Box 5. Using Food Records with Children
- Completion of food records requires a minimum level of literacy and the ability to legibly write39,98 what was consumed (or to search or browse for foods and beverages, in the case of technology-based approaches). Children must also have some basic knowledge of foods and how they are prepared, and the ability to quantify intake.
- For young children who are unable to report their own intake, proxy reporting is necessary. The involvement of children is important as they become more independent and parents or other caregivers are not present for all eating occasions.
- The burden associated with the completion of food records may result in compliance issues. As children get older and become more independent, they may be irritated by the need to record their intake at multiple points throughout the day.37
- The accuracy of dietary records may be improved with initial training and follow up review and verification of all details with the child, or proxy. As with recalls, probes should be phrased in a neutral manner to avoid eliciting social desirability or other forms of bias.
- With multiple days of recording, boredom and fatigue are possible.
Method: Food Frequency Questionnaires
Example tools: Diet History Questionnaire (DHQ), EPIC Food Frequency Questionnaire, Harvard Food Frequency Questionnaire, Multiethnic Cohort (MEC) Food Frequency Questionnaire
The aim of food frequency questionnaires is to gather information about the frequency with which different foods and beverages are consumed over some period of time, often the last month or year. Frequency questionnaires may prompt about typical or usual portion size, sometimes using images intended to facilitate accuracy of reporting. Questions regarding how often supplements are taken and usual dosages may also be included. Frequency questionnaires are typically self-administered. Depending on the comprehensiveness of the items included and their representation of the foods consumed by the target group, they may capture total diet or particular aspects of the diet. A questionnaire aimed at capturing total diet can be lengthy, requiring 30-60 minutes to complete. Given linkage to a food composition database, estimated nutrient intakes can be generated, though these have been shown to be affected by significant bias, at least for energy and the nutrients for which recovery biomarkers have been identified.65,66 Thus, estimating mean intake among populations using frequency data is not recommended. Unlike the short-term methods, food frequency questionnaires generally do not provide insights into aspects such as patterning of meals and snacks, where foods and beverages are eaten and sourced, other activities engaged in while eating, and similar constructs that may be salient to obesity-related research. In terms of technology in food frequency questionnaires, this is typically limited to web-based administration.
Contributors to error in data collected using food frequency questionnaires include imperfect long-term memory and the cognitive tasks associated with averaging frequency of consumption (and possibly amounts typically consumed) over a period of time such as a month or a year. Given that the list of foods and beverages is finite, error can come about if the questionnaire does not appropriately cover items commonly consumed by the target population. Thus, it is important the questionnaire is tailored to the target population. However, this tailoring, resulting in many different food frequency questionnaires for use with different populations, may pose a barrier to comparison across studies and pooling of data.
The tasks involved in completing food frequency questionnaires may not be well suited to children depending on their age and development. Young children have limited concepts of time31 and concepts of memory that are not fully developed.36 As a result, food frequency questionnaires querying about a long period of time are problematic. When these questionnaires are used, shorter time periods with meaningful start and end dates that provide cues to memory may be helpful.94 However, even with shorter time periods, children may not have developed the cognitive skills, including recalling recent intake of foods, averaging frequency of consumption over the reference period, taking into account weekday and weekend differences, and seasonal differences in frequency, to complete frequency questionnaires accurately.99 Children may also not have a good understanding of composite foods used in such questionnaires.94 The length of some frequency questionnaires may result in boredom and fatigue, leading to poor compliance or reporting quality among youth.39 In addition, when portion size is queried, the use of sizes not tailored to children can potentially result in systematic overestimation of intake.39
Example tools: Adolescent Food Habits Checklist, Block Screener for Kids, NCI Fruit/Vegetable/Fat Screener
Screeners are brief instruments that enable the collection of basic information about particular foods or beverages or other dietary behaviors. Screeners may query the frequency of intake of certain foods or beverages and thus may be thought of as short food frequency questionnaires, usually without questions regarding portion sizes. Alternately, screeners may ask about dietary practices, including routine use of items such as butter on bread. Similar to food frequency questionnaires, screeners tend to be self-administered. However, they can be very quick to complete (i.e., less than 15 minutes). As with food frequency questionnaires, information regarding patterning of food consumption and other contextual information are not collected unless queried separately.
Data collected using screeners are affected by similar sources of error as those collected using food frequency questionnaires. They cannot be used to estimate total diet. Screeners may be of particular use for dietary components that are not widely spread through different sources in the food supply, such as sugar-sweetened beverages, whereas they are likely to be less useful in terms of quality of data for components such as fruits and vegetables that can be consumed in many different forms and as part of mixed dishes. As with frequency questionnaires, the use of screeners to estimate mean intakes among populations or subpopulations is not recommended.
One dietary screener, the DSQ developed by NCI,100 can provide quantitative estimates of intake, converted from frequency responses using scoring algorithms derived from NHANES age- and sex-specific portion sizes. This questionnaire is currently being tested for validity among children. However, given their shared characteristics with food frequency questionnaires, the use of screeners with children may not be optimal. They are by definition shorter than frequency questionnaires so may reduce burden and resulting potential loss of interest among children. However, the cognitive challenges in averaging frequency of intake over time persist, as do issues with conceptualizing time and recalling intake.
Method: Measures Querying Related Dietary Behaviors
Example tools: Children’s Eating Behavior Questionnaire, Children’s Eating Attitudes Test (ChEAT)
Questionnaires may be used to identify or assess diet-related behaviors, including responsiveness, enjoyment, preferences, and attitudes toward food; snacking; restrictive behaviors; social pressures and norms surrounding food consumption; and satiety and hunger. These measures vary greatly in length and complexity, and may require parental assistance if the child does not yet have the cognitive skills to comprehend the content, or if the content has not been tailored to a younger audience. The use of age-appropriate measures in conjunction with parental perception may be useful in capturing more involved psychological constructs. Similar to other methods of measurement, questionnaires assessing diet-related behaviors may be influenced by the respondent’s body mass index101 (or that of the proxy reporter) and social desirability biases.