3.1 Methods for measuring food con­sump­tion of indi­viduals

This chapter describes methods commonly used for measuring the food con­sump­tion of indi­viduals. Subsequent chapters discuss the factors associated with the reproducibility and validity of each of these methods (Chapters 5‑7) and the calcu­lation and, subse­quently, evaluation of nutrient intakes (Chapters 4 and 8b).

Two groups of methods are used to measure the food con­sump­tion of indi­viduals. The first group, known as quanti­tative daily con­sump­tion methods, consists of recalls or records designed to measure the quantity of the indi­vidual foods con­sumed over a one day period. By increasing the number of measurement days, quanti­tative esti­mates of the usual intakes of indi­viduals can be obtained, using the same methods. The number, selection, and spacing of the days depend on the food intake, the nutrients of inter­est, the day-to-day vari­a­tion in nutrient intake, and the level of precision required. Determination of the usual intake of indi­vid­uals is partic­ularly critical when relationships between diet and biological parameters or health or chronic disease are assessed. Estimates of usual intakes of indi­vid­uals in a group or popu­lation are also needed to estimate the prevalence of inadequate intakes.

The second group of methods includes the dietary history and the food fre­quency question­naire. Both obtain retrospective inform­ation on the patterns of food use during a longer, less precisely defined time period. Such methods can be used to assess the usual intake of foods or specific classes of foods. With modifi­cation , they can also provide data on usual nutrient intakes of indi­viduals.

Quantitative daily con­sump­tion methods such as 24h recalls and records, when adjusted appropriately statis­tically to assess usual intakes, provide less-biased esti­mates of dietary intake than those such as food fre­quency question­naires and dietary histories that are designed to generate data on usual intakes directly (National Cancer Institute, 2015).

The measurement of food con­sump­tion at the indi­vidual level is costly and time consuming. Hence, such studies should be planned with care (Thompson et al., 2015). Even with the advent of new technologies to measure dietary intakes (Section 3.2), challenges still remain. Consid­eration should be given to the cost-effective collection of additional data from the same indi­viduals at the same time; such additional inform­ation may significantly enhance the inter­pretation of the dietary data. At a minimum, socioeconomic and health-related inform­ation, simple anthropometric measures, possibly a physical activity question­naire, and biological samples for the determination of important biomarkers (Chapter 15) should be collected when time and resources permit (Buzzard and Sievert, 1994).

The accurate assess­ment of the food intake of infants is partic­ularly difficult, espe­cially when infants are receiving both breast milk and complementary foods (Piwoz et al., 1995). WHO (1998) has published guidelines that can be used to evaluate nutrient intakes of breastfed infants receiving complementary foods.

3.1.1 Twenty-four-hour recall method

In the 24h recall method, the respon­dent and, where necessary, their parents or caretakers are asked by the nutritionist, who has been trained in inter­viewing tech­niques, to recall the respon­dent's exact food intake during the previous 24h period or preceding day. Thus the method assesses the actual intake of indi­viduals. However, a single 24h recall is not sufficient to describe an indi­vidual's usual intake of food and nutrients; multiple 24h recalls on the same indi­vidual over several days are required to achieve this objective (Section 3.1.2). Nevertheless, multiple single-day recalls on dif­fer­ent indi­viduals can give a valid measure of the usual intake of a group or popu­lation (Section 3.3.1).

A multiple-pass inter­viewing tech­nique is recommended for the 24h recall methods. The USDA has devel­oped an automated multiple-pass method (AMPM) consisting of five steps to collect dietary data using the 24h recall method; these five steps are shown in Table 3.1.

Table 3.1 USDA's Automated Multiple Pass Method. From (Steinfeldt et al., 2013).

USDA's Automated Multiple Pass Method
Design	AMPM Step	Memory Cues
Unstructured Respondents use their own strategy	Quick List ⇓ Collects recalled foods	Yesterday, midnight to midnight, day of week, activities, snacks and bever­ages, location
Structured Questions for specific types of foods	Forgotten Foods⇓ Probes for categories of forgotten foods	Seven question: Beverages, alcoholic bever­ages; sweets; savory snacks; fruits, veg­etables or cheese; breads or rolls; and one for anything else.
Structured Times and names of eating occasions	Time & Occasion⇓ Collects eating times and meal names	Breakfast, lunch, dinner, snack
Structured Standardized questions to collect details of each food Review foods for each eating occasion & intervals between each occasion	Detail Cycle⇓ Collect food details and amounts.	For breakfast you had a bagel & coffee Anything else? Did you have anything between your 7am breakfast and 10am snack?
Unstructured Respondents use their own strategy	Final Probe⇓ Anything else recalled	Situations where foods may have been easily forgotten: in the car, at meetings, when shopping, cooking, or cleaning up?

The AMPM is used by the US in the National Health and Nutrition Examin­ation Surveys (NHANES). Branded food products have now been added to the NHANES generic data­base to facilitate easier selection of foods con­sumed by the participants.

The first step in the AMPM is the Quick List where respon­dents list all food con­sumed in the previous 24h period. The second step, called Forgotten Foods, includes a series of questions that probes for categories of foods that are commonly forgotten. The third step, Time & Occasion, collects the time each food was eaten and the name of the eating occasion. The fourth step is the Detail Cycle which elicits descrip­tions of foods and amounts eaten, aided by the inter­active use of the USDA Food Model Booklet and measuring guide (USDA 2002+); see Chapter 5 for more details. The Detail Cycle also includes questions that review each eating occasion and each inter­val between eating occasions. The fifth step is a final review question, the Final Probe, which provides the respon­dent a final opportunity to recall any foods that had not been reported previously during the inter­view. The AMPM has been validated by comparing reported energy intake in young children with total energy expenditure assessed using the doubly labeled water method (Johnson et al., 1996) and in an observational study in which actual intakes of energy, protein, carbohydrate, and fat were compared with recalled intakes in men (Conway et al., 2004).

A modifi­cation of the multiple-pass 24h recall consisting of four passes — termed an interactive 24h recall — has been devel­oped to collect information on rural popu­lations in low- and middle-income coun­tries; details are given in the technical monograph by Gibson and Ferguson (2008).

In the first pass a complete list of all the foods and bever­ages con­sumed during the preceding day is obtained.

In the second pass, a detailed descrip­tion of each food and beverage con­sumed, including cooking methods and brand names (if possible) is collected. Standardized probe questions, specific for the popu­lation subgroup and setting, should be used to elicit specific details for each food item. For example, for milk products, probe questions should include the kind of dairy product, brand name (if appropriate), and percentage fat (as butterfat or milk fat). Further examples of probes that can be used to obtain detailed descrip­tions of specified foods are also provided in the monograph.

In the third pass, esti­mates of the amount of each food and beverage item con­sumed are obtained, generally in house­hold measures, and entered either on the data sheet Appendix 3.1 or a computer-based data-entry form. Care must be taken to record whether any fortified foods or bever­ages have been con­sumed. Graduated photo­graphs (Vossenaar et al., 2020), a set of measuring cups, spoons, and rulers, local house­hold utensils (calibrated for use), play dough, or food models of various types (Chapter 5) can be used to assist the respon­dent in assessing portion sizes of food items con­sumed (Gibson and Ferguson 2008; Lazarte et al., 2012). See guidelines on portion-size estimation methods by Vossenaar et al. (2020). Vossenaar et al. (2022) also provide technical guidance on how to collect, compile, and use portion size estimation method conversion factors in a 24h recall. Information on the amount of any mixed dishes con­sumed by the respon­dents, the amount of each ingredient in the mixed dishes, and the total amount of each cooked mixed dish prepared must also be collected at this time. These details are usually recorded on a separate data sheet or computer-based data-entry “recipe form". Vossenaar et al. (2023) have published guidance on the use of standard and non-standard recipes in quanti­tative 24h dietary recall surveys.

In the fourth pass, the recall is reviewed to ensure that all items, including the use of vitamin and mineral supplements, have been recorded correctly. Methods for coding the completed 24h recalls and potential sources of coding errors are discussed in Chapter 5.

Additional modifications that can be used in the interactive 24h recall are listed in Table 3.2

Table 3.2: Interactive 24-h recall modifications suggested for rural popu­lations in developing coun­tries to improve the recall of food items. From Gibson and Ferguson (2008).

Interactive 24-h recall modifications
Provide group training on portion size estimation before the actual recall.
Supply picture charts on the day before the recall for use as a checklist on the day the food is actually con­sumed, and for compar­ison with the recall to reduce memory lapses.
Provide bowls and plates for use on the recall days to help the respon­dents visualize the amount of food con­sumed.
Weigh the portion sizes of salted replicas of the actual foods con­sumed by the respon­dent.

and are discussed in more detail in the technical monograph (Gibson and Ferguson 2008). Whenever possible, recall inter­views in rural settings in low and middle-income coun­tries should be conducted in the respon­dent's home, because the familiar environment encourages partic­ipation, improves the recall of foods con­sumed, and facilitates calibration of local house­hold utensils by the inter­viewer. For a review of potential measurement errors using self-reported 24h recalls in low-income coun­tries and strategies for their prevention, see Gibson et al. (2017). However, major challenges arise when assessing food con­sump­tion at the indi­vidual level in coun­tries or cultures where the majority of the food con­sumed is as shared plate eating. Strategies used to overcome some of these challenges are reviewed by (Burrows et al. 2019).

A major advantage of a 24h recall is that the respon­dent burden is small so that compliance is generally high. In addition, the method is quick and relatively inexpensive, and can be used equally well with both literate and illiterate respon­dents. Nevertheless, any 24h inter­view protocol must be stan­dardized, pretested,and then piloted prior to use. Standardization is partic­ularly important in large-scale national surveys and for compar­isons across coun­tries (Slimani et al., 2000; Vossenaar et al., 2020). Adherence to the inter­view protocol and accuracy of food coding by the inter­viewers should be checked periodically during the survey, and the inter­viewers must be retrained if required to minimize inter­viewer bias (Chapter 5). Detailed sugges­tions on how to conduct the inter­view can be found in Hughes (1986) who stressed that leading questions and judgmental comments should be avoided. An indi­rect approach employing open-ended questions is recommended. This enables respon­dents to freely express their feelings so that answers are not biased. Piloting should be undertaken in an area near the study site, using respon­dents similar to those who will partic­ipate in the actual study.

When 24h recalls are used to characterize the average usual intake of a popu­lation group, the respon­dents should be representative of the popu­lation under study. In addition, the survey should be conducted in such a way that all days of the week are equally repre­sented. In this way, any day-of-the-week effects on food or nutrient intakes will be taken into account (Chapter 6). Seasonality must also be consid­ered as it affects food availability.

Guidelines on how to plan, design, and conduct large-scale 24h recall dietary surveys in low-and middle-income coun­tries have been devel­oped by the Intake Center for Dietary Assessment (Vossenaar et al., 2020; Deitchler et al., 2020), An outline of the main tasks that must be completed prior to the conduct of a large-scale 24h recall survey is presented in Figure 3.1. Intake.org also provide recommended specifications for dietary scales and procedures to test their accuracy and precision (Vossenaar et al., 2020).

A 24h recall has been used in several national nutrition surveys in low‑, middle‑, and high-income coun­tries. Examples include the Cameroon (Engle-Stone et al., 2014), Nigeria (Maziya-Dixon et al., 2004), New Zealand (Hennigar et al., 2018). (MOH, 2011), and the US NHANES III survey (Hennigar et al., 2018). Intake.org has provided technical assistance to support national or large-scale dietary surveys in Ethiopia, Jordan, Kenya, Niger, Nigeria, Senegal, Viet Nam, and Zambia. The AMPR method, used in the US NHANES III, has been adapted for use by both the Canadian and Australian national health surveys.

The US National Cancer Institute have devel­oped an Automated Self-Administered 24h recall (ASA24) based on the US AMPR method which is suitable for large-scale epi­demio­logical studies (Subar et al., 2012). The ASA24 web-based system was tested for use with children and adults in Canada (Kirkpatrick et al., 2017). Overall partic­ipants were receptive to completing ASA24, although the ASA24 inter­face and the steps in completing the recalls were not necessarily intuitive, espe­cially for younger children and elderly adults. The inves­tigators emphasized the importance of piloting protocols using online tools and the potential need for tailored resources to support certain age groups.

In general, recall inter­views can be conducted on children aged > 12y (Deitchler et al., 2020), and on most adults, except for persons with poor memories (e.g., some elderly). Parents are typically relied upon as proxy reporters for capturing dietary intake for children under five years of age. Children aged from 5–12y should be inter­viewed along with their primary caretaker, usually the mother, an approach termed assisted reporting. It may be necessary to inter­view several people if the children are at school, day-care, or play in the homes of friends, to ensure that all foods eaten away from home are reported. Recipe data for school or day-care meals may be required to supplement the food intake data collected in the home. In a study of Canadian preschool children 2–5y in which parents used the online self-admin­istered 24h recall (ASA24), they were able to report the food and bever­ages that their child had con­sumed with reasonable accuracy but the accuracy of portion size esti­mates was low (Wallace et al., 2018).

In the end, the success of the 24h recall, irrespe­ctive of the method applied, depends on the respon­dent's memory, their ability to convey accurate esti­mates of portion sizes con­sumed, the degree of motivation of the respon­dent, and the persistence of the inter­viewer (Acheson et al., 1980).

3.1.2 Repeated 24h recalls

Twenty-four-hour recalls can be repeated to estimate the average food intake of indi­viduals over a longer time period (i.e., usual food intake). The number of 24h recalls required to estimate the usual nutrient intake of indi­viduals depends on the day-to-day vari­a­tion in food intake within one indi­vidual (i.e., within-person vari­a­tion ). In turn, this vari­a­tion is affected by the nutrient under study, the study popu­lation, and seasonal vari­a­tions in intake.

Repeated 24h recalls were recommended as part of a system for measuring food con­sump­tion patterns in the United States to account for within-person variability as early as 1981 (NRC, 1981). In the US NHANES 2011–2014, food intakes on two non-consecutive days were recorded, on the first day through using computer-assisted personal recall inter­views (CAPI), and on the second day via a computer-assisted systematic telephone inter­view (CATI), using the AMPR on both occasions (Amoutzopoulos et al., 2018). The second recall inter­views were not conducted on all respon­dents, but instead were repeated on a randomly selected sub­sample of the popu­lation. Tooze (2020) recommends that the repeated 24h recalls should be performed on non-consecutive days 3–10d apart, and when only a random sub­sample of the popu­lation is used, repeated on at least 50 indi­viduals per stratum. Moreover, when episod­ically con­sumed foods, food groups or nutrients are of partic­ular inter­est, then more replicates per person should be collected rather than increasing the sample size from which two replicates are obtained (Tooze, 2020).

Repeated 24h recalls assisted by inter­viewers were also used for the Dutch National Food Con­sump­tion Survey (DNFCS) (van Rossum et al. 2016) and the French Nutrition and Health Survey (Castetebon et al., 2009). Several research groups in the UK have incorporated new technologies to collect 24h recalls, including the two web-based methods, myfood24 (Carter et al., 2015) and INTAKE24 (Simpson et al., 2017); see Section 3.2.3 for more details.

Estimated food records

For the estimated food record, also referred to as a food diary by some researchers, the respon­dent is asked to record in house­hold measures, at the time of con­sump­tion, all foods and bever­ages eaten (including snacks), for a specified time period. Detailed descrip­tions of all foods and bever­ages (including brand names) and their method of preparation and cooking should also be recorded. For mixed dishes such as Spaghetti Bolognese, the amount of each raw ingredient used in the recipe, the final weight of the mixed dish, and the amount con­sumed by the respon­dent should be recorded, wherever possible. The inform­ation is recorded on a form similar to that shown in Appendix 3-1 , except that house­hold measures are used for food amounts. Usually, the respon­dent, parent, or caretaker completes the food record, although in low or middle-income coun­tries a local field inves­tigator may perform this task (Dufour et al., 1999). Reactivity, defined as a change in behavior due to awareness that behavior is being or will be measured, may be an issue with this method. The respon­dent has the opportunity to modify his or her “usual” diet, potentially in a more socially desirable manner or to simplify the recording task (Thompson et al., 2015) (Chapter 5).

Food portion sizes can be estimated by the respon­dent in a variety of ways. Standard house­hold measuring cups and spoons should be used if possible, supplemented by mea­sure­ments with a ruler (for meat and cake) and counts (for eggs and bread slices). Unfor­tu­nately, errors may arise because the respon­dent may fail to quantify portion sizes correctly. Additional errors may also arise during the conversion of volumes to weights (Chapter 5), although this latter step is usually completed by the inves­tigator. Details on how to convert portion sizes to weight equiv­alents are given in Gibson and Ferguson (2008) and Vossenaar et al. (2022).

The number of days included in an estimated record varies, depending on the study objective. When the objective is to obtain an average intake for a group, then only one day per person is required, provided all days of the week are equally repre­sented in the final sample. However, when esti­mates of usual intakes of each person are required, then the number, selection, and spacing of the days required per person depends on the factors described for the repeated 24h recall (Section 3.1.2). Weekend days should always be proportionately included in the dietary survey period for each person, to account for potential day-of-the-week effects on food and nutrient intakes. This problem is discussed in more detail in Chapter 6.

The European Prospective Investigation of Cancer (EPIC) study in Norfolk, U.K. collected food and nutrient intakes from 2117 men and women using a 7d estimated food diary. EPIC was a large multicenter prospective study aimed at investigating the relationship between nutrition and various life-style factors and the etiology of cancer and other chronic diseases. The study involved 23 regional centers located in ten coun­tries, and involved a total cohort of about 480,000 subjects. The respon­dents of the EPIC study in Norfolk were provided with a “diet diary” — a 45-page colored booklet in which they were asked to record the descrip­tion, preparation, and amounts of foods eaten over seven consecutive days. Food portion sizes were estimated by the respon­dents in terms of house­hold measures, with the help of 17 sets of color photographs of small, medium, and large portions of the dif­fer­ent foods. See Vossenaar et al. (2020) for guidance on the development of food photo­graphs for portion size estimation. Details on the EPIC study in Norfolk Details are given in (Bingham et al., 2001). The Danish National Survey of Diet and Physical Activity (4–75y) also used a seven-day estimated food diary with a paper format (Pedersen et al., 2015).

A four-day estimated food diary, recorded using a paper form, is used by the U.K. National Diet and Nutrition Survey rolling program. This is a continuous cross-sectional survey designed to collect detailed inform­ation on food and nutrient intakes and nutritional status for a range of ages and across the social strata in the UK. The field work for this rolling program began in 2008 and is on-going with about 1000 people surveyed per year (Ziauddeen et al., 2018).

Weighed food record

Weighed food records are more frequently used in the United Kingdom and Europe because weighing scales are often used for food preparation in these regions. In the earlier British National Diet and Nutrition Surveys of adults and children (Ashwell et al., 2006), seven-day weighed food records were used. However, when the rolling program was introduced in 2008, the seven-day weighed food records were replaced by a four-day estimated food diary due to concerns about respon­dent burden.

A weighed food record is the most precise method available for estimating usual food and nutrient intakes of indi­viduals. It is the preferred method when diet counseling or corre­la­tion of intakes with biological parameters are involved. In a weighed record, the subject, parent, or caretaker is instructed to weigh all foods and bever­ages con­sumed by the subject during a specified time period. Details of methods of food preparation, descrip­tion of foods, and brand names (if known) should also be recorded. For mixed dishes such as Spaghetti Bolognese, the weight of the portion con­sumed should be recorded, along with the weights and descrip­tion of all the raw ingredients, including flavors and spices used in the recipe, as well as the final total weight of the mixed dish. The method of recording is similar to that shown for a 24h recall (Table 3.1), with the weight of the food items being recorded under "Amount."

If occasional meals are eaten away from home, respon­dents are generally requested to record descrip­tions of the amounts of food eaten. The nutritionist can then buy and weigh a duplicate portion of each recorded food item, where possible, to assess the prob­able weight con­sumed. Alter­natively, if appropriate, the nutritionist can telephone a restaurant to obtain details of the portion sizes con­sumed.

As with the estimated record, the number, spacing, and selection of days necessary to characterize the usual nutrient intakes of an indi­vidual using the weighed record depend on the within-person vari­a­tion in food intake, which, in turn, depends on the nutrient of inter­est, the study popu­lation, and any seasonal vari­a­tion of intake. Again, week-end days should be proportionately included to account for any weekend effect on the nutrient intake. If a weighed food record method is to be used, respon­dents must be motivated, numerate, and literate. Reactivity may be an issue with this method, and inter­fere with the assessment of usual dietary intake. Respondents may change their usual eating pattern to simplify the weighing process or, alternatively, to impress the inves­tigator (Cameron and van Staveren, 1988; Thompson et al., 2015) (Chapter 7). In addition, respon­dent burden for a weighed food record is higher than for an estimated record or for a 24h recall, so indi­viduals may be less willing to cooperate. Reproducibility, however, is greater in the weighed record than in the estimated record method because the portion sizes are weighed, although significant underreporting (Chapter 5) may still occur.

3.1.3 Dietary history

The dietary history method (Burke, 1947) attempts to estimate the usual food intake and meal pattern of indi­viduals over a relatively long period of time — often a month. This inter­view method was originally designed to be carried out by a nutritionist trained in inter­viewing tech­niques. More recently, computerized versions have been devel­oped which provide stan­dardized methods for data collection and probing, and minimize potential inter­viewer bias in responses (Kohlmeier et al., 1997).

Initially, the dietary history had three components. The first component was an inter­view about the usual overall eating pattern of the subject, both at mealtimes and between meals. Such inform­ation included detailed descrip­tions of foods, their fre­quency of con­sump­tion, and usual portion sizes in common house­hold measures. “What do you usually eat for breakfast?” is a typical question that might have been included in the inter­view.

The second component served as a crosscheck and consisted of a question­naire on the fre­quency of con­sump­tion of specific food items. This part was used to verify and clarify the inform­ation on the kinds and amounts of foods given as the usual intake in the first component. Questions asked related to specific foods, such as: "Do you like or dislike milk." A 24h recall of actual intake may also have been included at this stage.

In the third component, subjects recorded their food intake at home for three days. Portion sizes at this stage were estimated using a variety of tech­niques, including stan­dard measuring cups and spoons, common utensils, commercial plastic food models, photo­graphs, or real foods. Today, the original dietary history method is seldom used in this three-part format, the third component being commonly omitted.

The time periods covered by the dietary history method vary. The maximum time period that can be used has not been definitely established. When shorter time frames (i.e., one month) are used, reproducibility and validity are apparently higher than for longer periods (see Chapter 7). Measurements of food intake over one-year periods are prob­ably unrealistic unless seasonal vari­a­tions in food intakes are taken into account.

Dutch inves­tigators used a three-part dietary history method covering one month to record usual food con­sump­tion on weekdays, Saturdays, and Sundays separately (van Staveren et al., 1985). This approach takes into account the potential effect of weekends on nutrient intake. The portion size of foods most frequently con­sumed in this study were weighed by a dietitian in the home. A weighted daily average intake was then calculated from the data, using the following formula:

\[\small ((5 × \mbox{Weekday}) + \mbox{Saturday} + \mbox{Sunday}) / 7 \]

A modified version of this dietary history method was adopted in the Survey in Europe on Nutrition in the Elderly: A Concerted Action (SENECA). This multicenter survey was designed to examine cross-cultural vari­a­tions in the nutrition, lifestyle, health, and perform­ance of elderly Europeans (Euronut - SENECA, 1991). The method involved the completion of a three-day estimated record, followed by an inter­view during which respon­dents were questioned about their usual dietary intake over the past month. Portions of the most commonly eaten foods were weighed by the inter­viewer (van Staveren et al., 1996).

The recording of a dietary history can be very labor intensive, with inter­views taking up to 2h per respon­dent (Slattery et al., 2000). Several inves­tigators have reported that the dietary history tends to overestimate nutrient intakes, when compared with results from weighed records. Nes et al. (1991), for example, used the dietary history devel­oped for the SENECA study and showed that the method generated consistently higher intakes of energy and nutrients than three-day weighed records. Livingstone and Robson (2000) reported similar findi­ngs in a study of children and adolescents, but claimed that the results obtained from the dietary history were more representative of habitual intake than those obtained from seven-day weighed records. In general, because dietary histories, unlike food fre­quency question­naires, do not limit the variability in the responses, they overcome many of the limitations of a food fre­quency question­naire. However, the absence of a stan­dardized format for the dietary history method limits compar­ison of its usefulness across studies (Thompson et al., 2015).

3.1.4 Food fre­quency question­naire

The food fre­quency question­naire, sometimes referred to as a diet history question­naire, aims to assess the usual fre­quency with which food items or food groups are con­sumed during a specified time period. It was originally designed to provide descriptive qualitative inform­ation about usual food-con­sump­tion patterns. With the addition of portion-size esti­mates and the introduction of improved computerized self-admin­istered question­naires, the method has become semi-quanti­tative, allowing the derivation of energy and selected nutrient intakes (Willett et al., 1985; Block et al., 1986).

In its simplest form, the question­naire consists of a list of foods and an associated set of frequency-of-use response categories ( Appendix 3-2 ). The list of foods may focus on specific groups of foods, partic­ular foods, or foods con­sumed periodically in association with special events or seasons. Alter­natively, the food list may be extensive to enable esti­mates of total food intake and dietary diversity to be made. The frequency-of-use response categories may be daily, weekly, monthly, or yearly, depending on the study objective.

Specific combinations of foods can be used as predictors for intakes of certain nutrients or non-nutrients, provided that the dietary components are concen­trated in a relatively small number of foods or specific food groups. Examples include the fre­quency of con­sump­tion of fresh fruits and fruit juices as predictors of vitamin C intake (Tsugane et al., 1998), green leafy veg­etables and carrots (O'Neill et al., 2001), or fruit and veg­etable intake as predictors of carotenoid intakes (Whitton et al., 2017), whole grain cereals, legumes, nuts, fruits, and veg­etables as predictors of dietary fiber intakes (Merchant et al., 2003), and dairy products as predictors of calcium intakes (Barr et al., 2001; Horiuchi et al., 2019). The method can also be used to assess the intake of fats and cholesterol (Feunekes et al., 1993; Eng and Moy, 2011; Riordan et al., 2018), artificial sweeteners (Dewinter et al., 2016), certain contaminants present in specific foods (MacIntosh et al., 1997; Filippini et al., 2018), alcohol (Bazal et al., 2019), and condiments (Leyvraz et al., 2018).

The food fre­quency question­naires should feature simple, well-defined foods and food categories. Open-ended questions should be avoided as preformatted lists of food categories act as a memory prompt. The method may use a stan­dardized inter­view, a self-admin­istered machine-readable printed question­naire, or a computer-admin­istered question­naire. Most question­naires usually list 80 to 120 indi­vidual items and take from 15–30 minutes to complete (see abbreviated example given in Appendix 3.2). Hence, the food fre­quency question­naire imposes less burden on respon­dents than most of the other dietary assess­ment methods. The results are easy to collect and process and are generally taken to represent usual intakes over an extended period of time, and hence are not affected by day-to-day variability. As food fre­quency question­naires are designed to provide retrospective inform­ation about diet, they are often used in retrospective case-control studies (Schink et al., 2019). However, their validity and feasibility for estimating food intakes in the remote past has not always been clearly established (van Staveren et al., 1986; Dwyer and Coleman, 1997; Ambrosini et al., 2003; Barrett et al., 2019).

Food scores can be calculated from qualitative food fre­quency data and the fre­quency of con­sump­tion of certain food groups (Gil et al., 2015). The 2015–2020 US Dietary Guidelines for Americans or an equivalent stan­dard such as the WHO Healthy Diet, list the optimum number of servings of the major food groups per person per day, and serve as a basis for the scores. The scores can then be examined in relation to psycho­social influences (e.g., level of education, income), as well as vital statistics, season, geographic distri­bution, and biomarkers of dietary patterns (Playdon et al., 2017).

WHO has issued guidelines on a Healthy Diet (WHO, 2020). A (Healthy Eating Index) (HEI‑2015) has been devel­oped by the USDA, that measures alignment of US diets with the 2015‑2020 Dietary Guidelines for Americans. The HEI‑2015 has 13 dietary components in total, of which nine are “adequacy” components (those recommended for inclusion in a healthy diet) and four “moderation” components that should be con­sumed sparingly. These dietary components are listed in Table 3.3.

Table 3.3 HEI–2015 Components & Scoring Standards. From HEI (2020).
¹Intakes between the minimum and maximum standards are scored proportionately.
²Includes 100% fruit juice.
³Includes all forms except juice.
⁴Includes legumes (beans and peas). ⁵Includes all milk products, such as fluid milk, yogurt, and cheese, and fortified soy bever­ages.
⁶Includes legumes (beans and peas).
^6,7Includes seafood, nuts, seeds, soy products (other than bever­ages), and legumes (beans and peas).
⁸Ratio of poly- and monounsaturated fatty acids (PUFAs and MUFAs) to saturated fatty acids (SFAs).

HEI–20151 Components & Scoring Standards
Component	Max. points	Standard for max. score (per 1,000 kcal)	Standard for min. score (0)
Adequacy
Total Fruits2	5	≥ 0.8 cup equiv.	No Fruit
Whole Fruits3	5	≥ 0.4 cup equiv.	No Whole Fruit
Total Vegetables4	5	≥ 1.1 cup equiv.	No Vegetables
Greens and Beans4	5	≥ 0.2 cup equiv.	No Dark Green Veg. or Legumes
Whole Grains	10	≥ 1.5 oz equiv.	No Whole Grains
Dairy5	10	≥ 1.3 cup equiv.	No Dairy
Total Protein Foods6	5	≥ 2.5 oz equiv.	No Protein Foods
Seafood and Plant Proteins6,7	5	≥ 0.8 oz equiv.	No Seafood or Plant Proteins
Fatty Acids8	10	(PUFAs + MUFAs) /SFAs ≥ 2.5	(PUFAs + MUFAs) /SFAs ≤ 1.2
Moderation
Refined Grains	10	≤ 1.8 oz equiv.	≥ 4.3 oz equiv.
Sodium	10	≤ 1.1 gram	≥ 2.0 grams
Added Sugars	10	≤ 6.5% of energy	≥ 26% of energy
Saturated Fats	10	≤ 8% of energy	≥ 16% of energy

This table also shows the scoring stan­dard for each of the dietary components. The scores for each of the 13 com­ponents derived from the indi­vidual's intake, are added to give an HEI Index score (maximum = 100) for the indi­vidual.

Kant et al. ( 2000) used food fre­quency inform­ation to calculate a “Recommended Food Score” (RFS) to evaluate the con­sump­tion of foods consid­ered to be consistent with the U.S dietary guidelines existing at that time. In a prospective study of diet quality and mortality in women from the United States, they showed that the RFS was inversely associated with all‑cause mortality. Since that time, the RFS has been used in Australia as a diet quality index for preschoolers (Burrows et al., 2014), and in Korea to investigate links with physical perform­ance among the elderly (Jeong et al., 2019), and depression, anxiety, and quality of life (Lee et al., 2019).

Many recent users of food fre­quency question­naires have quantified portion sizes of food items of inter­est, often using photo­graphs (Nelson et al., 1994; Amougou et al., 2016). Portion sizes can be ranked as small, medium, and large, preferably based on age and sex-specific portion size data generated from country-specific national nutrition surveys (Willet et al., 1985; Block et al., 1986; Bohlscheid-Thomas et al., 1997). Note that inclusion of inform­ation on portion sizes produces semi-quanti­tative food fre­quency data (Appendix 3-3 ). This can be converted to data on energy and nutrient intakes by multiplying the fractional portion size of each food con­sumed per day by its energy and nutrient content, obtained from appropriate food composition data. The results are then summed to obtain an estimate of an indi­vidual's total daily energy and nutrient intake.

Block et al. (1986) derived a food list for a food fre­quency question­naire with portion sizes from the NHANES II results. Food items selected were based on the fre­quency of con­sump­tion of certain specific food items and which contributed significantly to the total popu­lation intake of energy and each of 17 nutrients. Serving sizes were estimated from observed portion size distri­butions in the NHANES II data. Medium serving sizes for each food were specified in the food fre­quency question­naire, and the respon­dent indi­cated whether his or her usual serving size was small, medium, or large, as shown in Appendix 3.3. A specialized food composition data­base was devel­oped for use with this food fre­quency question­naire. A very similar approach has been used to design semiquanti­tative multi-ethnic food fre­quency question­naires (Deurenberg-Yap et al., 2000; Beukers et al., 2015).

An improvement to the Block food fre­quency question­naire termed the Diet History Question­naire (DHQ) has been devel­oped. The DHQ consists of 124 food items, portion sizes, and dietary supplement questions and takes one hour to complete. Results of a comparative validity study showed that the perform­ance of the DHQ is better than that of the food fre­quency question­naires of both Block and Willett (Subar et al., 2001).

The semi-quanti­tative food fre­quency question­naire has become a widely used tool in dietary assess­ment. Country-specific semiquanti­tative food fre­quency question­naires containing between 130 and 300 food items were used in the EPIC study to estimate indi­vidual usual food intakes (Margetts and Pietinen, 1997). In the EPIC study in Norfolk, U.K. (Bingham et al., 2001) respon­dents (n = 23,003) estimated how frequently foods were eaten over the past year, from nine possible frequency-of-use response categories from a list of 130 foods. Reduced versions containing only 60 food items that require only 17 min to admin­ister by an inter­viewer are available; even the full 98‑item Block question­naire requires only 30‑35 min of inter­viewer time (Block et al., 1990). In some coun­tries, a semi-quanti­tative food fre­quency question­naire has been used in national nutrition surveys (e.g., 1995 Australia National Dietary Survey) (Williams, 2005).

More recently in Australia, an on-line suite of validated semi-quanti­tative food fre­quency question­naires, the Australian Eating Surveys (AES) have been devel­oped. The AES is a 120‑item food fre­quency question­naire with fifteen supplementary questions, including food and sedentary behaviors and supplements, with the aim of capturing the usual dietary intakes of children, adolescents, and adults over the previous six months. The AES takes only 15‑20 min to complete on-line and can generate a personalized dietary feedback in real time as an incentive to encourage partic­ipation and enhance the response rate (Collins et al., 2014).

In many low-income coun­tries no national food con­sump­tion surveys have been conducted so the key inform­ation needed to develop a food fre­quency question­naire is often not available. Such inform­ation includes a listing of foods commonly con­sumed in the study popu­lation, details on the way the foods are typically prepared or con­sumed, and the usual range of portion sizes con­sumed. In response to this need, Hotz and Abdelrahman (2019) have devel­oped some simple methods to obtain food listing and portion size distri­bution esti­mates for use in semi-quanti­tative food fre­quency question­naires.

Table 3.4 Ref = reference quintile. Multivariate model: age at baseline, total calorie intake, smoking history, cancer (yes/no) hypertension diagnosis (yes/no), use of antidepressants, elevated cholesterol, physical activity level, body mass index, cardiovascular disease, multivitamin use, intake of alcohol, total calorie intake, profession, missing indicator for SCF, and number of dietary assess­ments during 1986–2002. From Yuan et al. (2019).

Odds Ratios (95% CIs) for poor Subjective Cognitive Function (SCF), compared with good function, associated with total veg­etable, fruit, and fruit juice intakes
Quintile of intake (n = 27,842 men)	Q1	Q2	Q3	Q4	Q5
Total veg­etable intake Median, servings/d	1.7	2.5	3.2	4.1	5.7
Odds Ratio 95% Confidence Interval	Ref	0.92 (0.79, 1.08)	0.85 (0.73, 0.99)	0.71 (0.60, 0.83)	0.62 (0.52, 0.74)
Total fruit intake Median, servings/d	0.5	1.1	1.5	2.0	3.1
Odds Ratio 95% Confidence Interval	Ref	1.00 (0.85, 1.18)	0.96 (0.81, 1.13)	0.81 (0.68, 0.96)	0.79 (0.66, 0.94)
Total fruit juice intake Median, servings/d	0.1	0.4	0.7	1.0	1.5
Odds Ratio 95% Confidence Interval	Ref	0.78 (0.67, 0.91)	0.81 (0.70, 0.94)	0.66 (0.57, 0.77)	0.65 (0.55, 0.76)

Food fre­quency question­naires are often used by epidemiologists studying associations between dietary habits and disease. For example Yuan et al. (2019) in their study of the relationship between long-term intake of veg­etables and fruits and subjective cognitive function in US men, used this approach (Table 3.4). See also: (Willett, 1994; Harris et al., 2018; Zhong et al., 2019). In such studies, the food fre­quency question­naires must be semi-quanti­tative, with the ability to rank subjects on the basis of their intakes, so that subjects with low intakes can be separated from those with high intakes. This permits the calcu­lation of the odds ratio or relative risk of disease in relation to intake of certain foods, food groups, or nutrients (Masson et al., 2003).

Box 3.1 summarizes the five methods discussed above that can be used to assess the food con­sump­tion of indi­viduals and their uses and limitations. For a more detailed review of the strengths and weaknesses of each of these methods for use in low- and middle-income coun­tries, see (FAO, 2018).

3.2 Technical improvements in food con­sump­tion mea­sure­ments

The increasing evidence of the relationship between diet and chronic disease has led to a number of technical advances in mea­sure­ments of food con­sump­tion for indi­viduals. They include the use of the telephones in surveys, digital photo­graphs of food portions displayed on a computer/tablet or captured with a cell-phone camera, as well as web-based tools, mobile apps, and image-based assess­ment. These innovative technologies aim to improve the speed, accuracy, and partic­ipation, and reduce reporting burden, bias, and the cost of collecting and analyzing dietary intake data during large scale epidemio­logical studies as well as national surveys. A summary of the strengths and limitations of these innovative technologies designed to improve dietary assess­ment methods is shown in Table 3.5.

Table 3.5 Strengths and limitations of innovative technologies to improve dietary assess­ment methods. From: Dietary Assessment: A resource guide. FAO (2018).

All innovative technologies used in dietary assess­ment
Strengths	Limitations
Costs for data collection can be lower (less need for person-to-person interaction)	Larger up-front investments (i.e. purchase of mobile phones, cameras, computers, software development, etc.)
Convenient for users, good acceptability, may improve compliance	Risk of losing devices
Do not rely on respon­dent's memory	Risk of technical problems (i.e. low battery, loss of Internet connection) could impede data collection
Record of qualitative information (e.g. date and time of recording)	A backup method is required to collect information, if technical problems occur
Significantly cuts down data processing time
Personal Digital Assistant (PDA)
Strengths	Limitations
Is portable and can be easily carried by study participants	Face-to-face training of the participants is required
Facilitates real-time data collection, entry and coding	Low level of dietary data details because of pre-coded food listings
It is possible to set an alarm within the PDA to alert participants to record their food intake	Increases the respon­dent burden compared with pen and paper records, due to the extensive list of foods (depending on each PDA)
Can be programmed to allow partic- ipants strict access to the dietary intake software only	Reports of difficulty using the search function and inability to find certain foods
Image-assisted methods i.e. digital photographs
Strengths	Limitations
Easy to use	Participants may forget to take some images
Suitable for low literacy popu­lations (if the technologies are easy-to-use, e.g. digital cameras)	Not all information can be captured with a single photo­graph/image
Quality of digital cameras keeps improving and pictures with higher resolutions can help improve the accuracy of analysis	Difficulties in estimating portion size of food con­sumed from common mixed dishes
Suitable for subjects with memory impairments and for children	Lack of details about cooking methods
Lower under-reporting compared with some tradi­tional methods	Probably more limited accuracy for coun­tries with a wide range of mixed dishes (e.g. Asia) Still needs a written record of foods obscured in photos, and when details of ingredients are required
Image-assisted methods i.e. mobile-based technologies
Strengths	Limitations
Possible higher quality control of data because of shorter delays and real-time responses	It is costly and time-consuming to develop the application as an interface, and the software for automated portion size estimation
Possibility of sending reminders	Certain types of foods, such as mixed dishes (soups, stews, casseroles, etc.) can be difficult to analyse with automated image analysis
Internet access would allow respon­dents to send instant photos, thus minimizing system­atic mistakes	Requires certain level of literacy Requires network/Internet access (for real-time data collection)
Interactive computer and web-based technologies
Strengths	Limitations
Efficiency in terms of recording information and data processing (i.e. reduced costs and interviewer workload)	Some imaging algorithms presently fail to identify foods correctly and to accurately estimate the quantity of food in the computerized images
Increased levels of quality control	Need adaptation of the software to local settings
Include interactive visual and audible aids	Require high levels of literacy and computer skills from the participants
Suitable for large, geo­graph­ically dispersed samples; can include different coun­tries/languages (web-based)	Possibility of collecting less food details (i.e. ingredients, methods of preparation, etc.)
Data processing can be finalized at any time and location (web-based) Can provide personalized dietary feedback (web-based) Possibility of sending interactive reminders (web-based)	Require Internet access (web-based)

A robust assess­ment of the perform­ance, cost, and response rates should be undertaken, however, prior to the adoption of any new-technology-based method.

3.2.1 Telephone

A telephone survey that is well-designed and carefully-admin­istered appears to be a promising method of obtaining dietary inform­ation. In partic­ular, telephoned 24h recalls are being increasingly used. The USDA has conducted several studies to examine the feasibility of using telephone follow-up surveys instead of mail follow-up for 24h recalls. Results have been promising: response rates for telephone follow-up were much greater than those for mail follow-up. Casey et al. (1999) carried out a validation study in which the results of 24h recalls conducted over the telephone were compared with in-person recalls collected in the 1994-1996 USDA Continuing Survey of Food Intakes by Individuals; strong corre­la­tions were reported. As a result, since 2002 NHANES have used the telephone to admin­ister the second recall day 3–10d later using the AMPR method (Hennigar et al., 2018).

Several other large-scale telephone dietary surveys have been used in the United States, some of which have used a food fre­quency instrument rather than a 24h recall. Lyu et al. (1998) successfully showed that agreement between telephone and face-to-face inter­views of a semi-quanti­tative food fre­quency question­naire in Hawaii made up of 115 food items was good and unaffected by age, gender, ethnicity, or education of the respon­dents. These inves­tigators did recommend mailing photo­graphs of foods in three portion sizes in advance, to help respon­dents estimate amounts eaten more accurately.

Many of the smaller telephone dietary studies have been conducted on adult women (Casey et al. 1999; Tran et al., 2000; Yanek et al., 2000). Very few have been carried out on adolescents and adult men (Bogle et al., 2001). In college students, food intakes by telephoned recalls have been compared against actual intakes determined surrep­titiously in the college cafeteria (Krantzler et al., 1982). More studies are needed, however, among certain life-stage groups, to establish the validity of 24h recalls or food fre­quency question­naires admin­istered over the telephone. Under-reporting of self-reported food intakes may still occur in telephone recalls, as they do with in-person recalls; this was observed, for example, when total energy intakes derived from telephone-admin­istered 24h recalls were compared with total energy expenditure measured by doubly labeled water (Tran et al., 2000) (see Chapter 7).

Telephone surveys do have several advantages. These include their ability to reach a large number of persons at perhaps less than half the cost of face-to-face surveys. As well, with the advent of computer-assisted telephone inter­viewing, the inter­views can be readily stan­dardized, queries can be clarified, and responses can be coded immediately, during an inter­viewing time that is much shorter than a face-to-face inter­view. As a result, the response rate is enhanced.

Potential inter­viewer biases can be eliminated by using computer-assisted telephone inter­viewing, an approach used with the AMPR in the NHANES surveys. However, other sources of bias may occur with telephone surveys; these may arise from non­coverage and non­response. In the United States, although over 87% of the popu­lation owns a telephone, subgroups such as the poor, certain minorities, and the` elderly still have fewer telephones than the general popu­lation. Such dif­fer­ential coverage can introduce bias in national surveys unless alternative compensating strategies are employed. Strategies may include using supplementary face-to-face or mail inter­views for persons without telephones; statis­tical adjustments employing weighting for sex, age, race, and income; selective over-sampling of the non-telephone users; and random-digit dialing to contact those with unlisted numbers. Nonresponse is also a source of potential bias that influences telephone surveys. This is espe­cially a prob­lem among certain subgroups such as the ethnic minorities for whom language barriers may be consid­erable, the elderly who may have hearing difficulties, and persons with less education.

Despite these limitations, telephone dietary surveys are a practical, economical, and valid alternative to the conventional face-to-face methods for large-scale epidemio­logical studies and nutrition surveys and for developing and evaluating community-based nutrition inter­ventions in higher income coun­tries.

3.2.2 Portion size estimation aids

Errors in quantifying the portions of food con­sumed are often the largest error in most food con­sump­tion surveys, unless weighed food records are used. As a result, several types of portion-size measurement aids (PSMAs) have been devel­oped in an effort to enhance the accuracy of portion size esti­mates (see Chapter 5 for more details). PSMAs can be divided into two categories: (1) non-photo­graphic aids and (2) photo­graphic aids. Canada was one of the first coun­tries to use non-photo­graphic aids in its National Nutrition Survey in 1973. The aids consisted of a collection of three-dimensional graduated food models of various volumes and surface areas prepared from paper-mache, wood, or hardboard. A range of graduated food models was used to prevent the tendency to generate a “direct” response. The latter phenomenon is observed when plastic food models representing only “average” portion sizes are used (Samuelson, 1970). The Canadian surface-area models were accom­panied by stan­dard thickness indi­cators made of hardboard squares. These were used during the 24h recalls to assist in assessing the overall size and thickness of foods such as cheese, cold meats, cakes and cookies. Use of thickness indi­cators is critical for assessing portion sizes of intact cuts of meat, espe­cially when irregular in shape.

Since 1973, graduated food models (Figure 3.2)

have been used with 24h recalls in many national food con­sump­tion surveys, including those in the United States (NHANES, 2017) and New Zealand (MOH, 1997). In the dietary component of the NHANES nationwide surveys, the tool devel­oped to measure portion sizes in the 24h recalls consists of a food model booklet of 32 life-size two-dimensional drawings of house­hold vessels (glasses, mugs, bowls), abstract shapes (mounds and spreads), and geometrical models (circles, a grid, wedges, and thickness bars), together with a set of measuring cups, spoons, and rulers (USDA 2002+); More details are given in Chapter 5. In low-income coun­tries, PSMAs for 24h recalls may be salted replicas of staple foods, real foods, and local house­hold utensils (e.g., bowls, plates, spoons). These are often used in combination with items to simulate liquid and solid foods (e.g., water or raw rice), manip­ulative models (modelling clay or play dough), tape measures or rulers and graduated food models (Gibson and Ferguson 2008). Guidelines on PSMAs for use with 24h recall methods in low-income coun­tries are available from www.intake.org. The use of less than five dif­fer­ent PSMAs are recommended for surveys in low-income coun­tries. (Vossenaar et al., 2020).

More recently, photo­graphs are being increasingly used as PSMAs. They may include pictures of indi­vidual foods or meals which have been drawn, printed, or displayed digitally, often in color and sometimes shown next to a stan­dard sized object (e.g., a coin) or grid­lines to aid the respon­dent's perception of scale. For surveys using 24h recalls, a series of graduated portion-size photo­graphs (i.e., amount con­sumed on any one occasion) for each food item are commonly displayed during the inter­view, whereas for surveys employing food fre­quency question­naires, either an average serving size for each food (i.e., average amount served in one helping) or a series of serving sizes (small, average, large) are often used. The graduated portion-size photo­graphs are often bound together in a photo­graphic atlas.

Practical guidelines on how to develop a photo­graphic atlas are given by Nelson and Haraldsdottir (1998) and Vossenaar et al. (2020). Factors that must be consid­ered in relation to the format of the photo­graphs include size of the image, number and range of portions sizes depicted, and the inter­val between portion sizes. In the EPIC study, for example, there was a 25% difference between portion sizes to allow a real visual perception of differences in size (Vilela et al., 2018). Nelson et al. (1994) used portion weights from the British Adult Dietary Survey ranging from the 5th to the 95th percentile for a series of eight photo­graphs for each food. Other important factors that should be stan­dardized include the order of presentation of the photo­graphs, labels used, angle at which the photo­graph was taken, background and use of reference objects for scale, color versus black and white, and use of one versus several foods on a plate.

Increasingly digital photos of food portions with multiple images displayed on a computer or a tablet are replacing printed photos to reduce the cost of printing atlases and their transportation to dif­fer­ent inter­view sites. Nichelle et al. (2019) compared the accuracy of three portion sizes (small, medium, large) from printed photo­graphs in the Brazilian GloboDiet atlas with digital photos displayed on a computer screen for 20 foods selected from the Brazilian GloboDiet atlas. The mean error (difference between the estimated and true portions) was not significantly dif­fer­ent between the printed and digital photos, with agreement on using the printed and digital photos being 91% and 90%, respe­ctively.

During the development of the Internet-based, Automated, Self-Admin­istered 24h recall (ASA24), Subar et al. (2012) investigated the best way to provide digital images as portion-size estimation aides. They compared the photo­graphs said to represent portion sizes of foods con­sumed by the respon­dents on the previous day with the actual weights obtained by surrep­titious weighing. The images were also varied by the angle in which the photo­graphs were taken, the type of image, and the number and size of digital images presented in an effort to determine how best to present the digital images to facilitate accuracy. None of the photo­graphs were labeled with portion-size inform­ation. Results indi­cated that accuracy was not affected by the type of image, whether the images were presented simul­taneously vs. sequen­tially, or the size and number of images presented, although there was a tendency for eight rather than four images to be more accurate. Based on these results, aerial photo­graphs are used for portion size estimation in the ASA24.

Alter­natively, real-time images of foods and bever­ages con­sumed by the respon­dent, and captured passively, or actively with a cell-phone camera (Arab et al., 2011), digital camera (Lazarte et al., 2012), or wearable camera (Gemming et al., 2013), can be used as PSMAs during a 24h recall the following day. The active methods require the respon­dents to take pictures in the appropriate way (e.g., 45°angle, arms distance from the plate, with a size reference included for scale) before and after eating all meals, and to add accompanying notes. However, all of these tasks necessitate training the respon­dents. As a result, for low-literacy popu­lations, the passive tech­niques are more suitable, where an indi­vidual wears a camera to capture photo­graphs, thus requiring less involvement by the respon­dent.

In many food con­sump­tion surveys in both lower and higher income coun­tries, a mixture of non-photo­graphic and photo­graphic aids have been used as PSMAs. A system­atic review that included both non-photo­graphic and photo­graphic PSMAs concluded that culturally appropriate digital or hard-copy photo­graphic PSMAs were more accurate than non-photo­graphic food models and house­hold utensils (Amoutzopoulos et al., 2020). The inves­tigators also emphasized when selecting an appropriate tool for estimating portion sizes, both the setting and popu­lation under study must be consid­ered. Most of the studies included in this review were from high income coun­tries. In a study in rural Bolivia, however, the mean energy and nutrient intakes of 45 women (20–52y) obtained using an inter­viewer-admin­istered atlas of graduated photos of local foods combined with digital food photo­graphs taken by the respon­dents on the day prior to the 24h recall, were compared with those derived from a weighed record. No significant mean differences were reported in estimated and actual food amounts between the 24h recall using the food photo album and the digital photo­graphs and weighed food records, except for a few foods (rice, potatoes, eggs, veg­etables). For the latter foods, differences ranged from −5.4% for veg­etables to −6.8% for rice. However, there was a slight under­estimation in values between the two methods for some nutrients (Lazarte et al., 2012).

3.2.3 Web-based systems for dietary assess­ment

Table 3.6 Some of the web-based 24-h recall tools for dietary assess­ment

Web-based 24-h dietary recall tools for dietary assess­ment
Author	Tool (Country)	Notes (All tools use photographs to aid portion size estimation)
(Cade, 2017),	(myfood24), (UK)	Large data­base with generic and branded food items; meal-based approach; simplified AMPM approach,
(Bradley et al., 2016),	(INTAKE24), (UK)	Meal-based approach; based on AMPM approach; Multipass recall for subjects 11–24y.
(Lassale et al., 2015),	(NutriNet Santé) (France)	Web-based 24h diet record; Food portion pictures for 250 foods
(Thompson et al., 2015),	( ASA24), (USA)	Meal-based approach; uses the five steps of the AMPM; questions about food preparation; available in English and Spanish; optional supplement module
(Arab et al., 2015),	DietDay (USA)	Multiple steps similar to the AMPM approach; questions about food preparation; recording of supplements; provides dietary feedback
(Biltoft-Jensen et al., 2014),	WebDASC (Denmark)	Self-administered tool for children 8–11y; animated; guides response to six eating occasions; 1300 foods selected via category browse or search.
(Vereecken et al., 2014).	CANAA-W (Belgium)	A feasibility study on 131 children and 53 parents; to refine the lay out and structure of the tool; and the list of food items included.

On-line dietary assessment tools based on 24h recalls, food fre­quency question­naires, and dietary records are becoming increasingly available. They can be inter­viewer- or self-admin­istered so data can be collected at a time and location that is convenient to the respon­dent. Examples of web-based 24h dietary recall tools are shown in Table 3.6. These include the US National Cancer Institute ASA24 and DietDay (Arab et al., 2011), the UK INTAKE24 (Bradley et al., 2016), and UK myfood24 (Cade, 2017), the French NutriNet Sante (Touvier et al., 2010), WebDAS in Denmark (Biltoft-Jensen, 2014), and the CANAA-W in Belgium (Vereecken et al., 2014). Some of these have been modified for use in other high-income coun­tries such as Canada, Australia and Germany.

The use of automated 24h recall inter­views in web-based systems stan­dardizes the inter­viewing process so there is consistency in the format in relation to the questioning and sequence, food descriptors, and methods used to estimate food portion sizes. Some (e.g., ASA24; INTAKE24; myfood24) use a meal-based approach. Figure 3.3

shows an example of a screenshot from ASA24 for the recall of breakfast. Most are based on a version of the AMPM inter­view format devel­oped by USDA (Moshfegh et al., 2008).

Groupings of foods and recipes, as well as the level of detail used to describe foods varies across systems. In most of the applications, ingredients of mixed dishes are entered separately in either the raw or cooked form, with an option to select a generic mixed dish if the ingredients of the recipe are unknown. Some of these on-line systems use only food portion photo­graphs to help respon­dents report the amount of each food con­sumed (e.g., ASA24; DietDay), whereas others (e.g., INTAKE24) use multiple options for portion size selection. These may include graduated photo­graphs or images of actual foods, guide photo­graphs for selected prepackaged items (e.g., snacks), with or without additional options such as photo­graphs/pictures of house­hold measures (glasses, cups, bowls, teaspoons etc.) Generally, additional automated prompts for potentially forgotten items are programmed and quality controls are implemented at each step of the inter­view procedure, with checks made on outlier values. Standardized algorithms are often used to calculate automatically grams of the food con­sumed, and the amounts of ingredients (either raw or cooked) con­sumed from a mixed dish. Some of the applications, espe­cially for children, record the amounts served and leftovers rather than the amounts con­sumed. Most applications have a separate screen for describing and quantifying dietary supplements. Usually, at the end of the inter­view, a summary is displayed with the opportunity for the respon­dent to correct or edit inform­ation at each stage of the inter­view. Each web-based system is linked to an in-country food composition data­base, some of which include branded food items (e.g., ASA24, myfood24), with automated coding for those reported foods and bever­ages that are in the food composition data­base. Items not in the data­base are coded manually.

When developing web-based systems, the potential technology readiness of the user must be consid­ered. Certain popu­lation groups, including older indi­viduals with lower computer skills and technical experience, may have difficulty using a computerized assess­ment tool. Currently, only a few of these web-based 24h recall systems have undergone rigorous evaluation using either independent biomarkers (i.e., Intake24) (Foster et al., 2019), or measures of true intakes (i.e., ASA24) (Kirkpatrick et al., 2014), although for most when compared with the stan­dard 24h recall method, agreement is generally close (Cade, 2017). More well-designed validation studies of web-based systems employing biomarker measures to investigate misreporting of dietary intakes, by age, educational level, BMI, and gender are needed before they can be applied in epidemio­logical studies involving multi-center sites (Illner et al., 2012).

In low-income coun­tries where reliable inter­net access may be lacking, espe­cially in rural areas, web-based methods are not feasible for assessing dietary intakes, espe­cially when self-admin­istered, due to the low literacy rates that often exist. To address these concerns in low-income coun­tries, the International Dietary Data Expansion (INDDEX) Project has devel­oped a dietary assess­ment platform called INDDEX24 (Coates et al., 2017). The platform consists of a mobile application for the collection of off-line indi­vidual-level dietary data using the 24h recall method with the AMPM, linked to a web data­base application so that the users can download the required country-specific food composition data. The dietary recall mobile application uses an open source mobile data collection platform (Commcare) so that it is accessible to researchers at no or low cost. The application is designed so that it is straightforward to use and contains a number of quality control features such as controls to prevent entry of implausible values, pre-defined portion-size estimation methods depending on the type of food reported, and real time calcu­lation of the respon­dent's reported energy intake in order to identify possible under- and over-reported intakes (Burrows et al., 2014). The mobile app converts food portions reported in non-stan­dard units into gram-weight equiv­alents by linking to context-specific portion size conversion factors housed within the web data­base (Coates et al., 2017). Both the perfor­mance and cost of a 24h recall using the INDDEX24 mobile app compared with a pen-and-paper 24h recall have been explored. See Rogers et al. (2022) and Adams et al. (2023) for further details.

3.2.4 Image-based methods of dietary assess­ment

These are defined as any method in which images are used for identifying foods, or for portion size estimation. Personal digital assistants (i.e., handheld computers) were used initially for these image-based methods, but have now been replaced by camera-enabled cellular phones or smartphones, digital, disposable, or 35mm cameras, wearable cameras, or digital video cameras. Many of these image-enabled devices rely on users manually taking photo­graphs of foods (i.e., active image capture). Alter­natively, for those methods termed “passive image capture”, pictures are automatically taken, usually at timed inter­vals (e.g., every 10 seconds) by the devices (predominantly wearable digital cameras), so that no inter­vention is required on the part of the user. The majority of tools that use active image capture have been designed for research purposes to enhance the accuracy of self-reported intakes. Fewer passive image capture methods exist and none is fully automated, most relying on specialists to manually estimate portion size (Gemming et al., 2015a).

A few image-based methods are based on a combination of tools. Examples include: SmartPlate (SRI International, 2015), which consists of a camera and sensor-enabled dinner plate, the Diet Data Recorder System (DDSR), a three-part system comprised of a camera-enabled smartphone, laser module, and circuit module (Bell et al., 2017), and the Dietary Intake Monitoring System (DIMS). The latter is a five-part system with a digital camera, weighing scale, infrared thermometer, radio-fre­quency identification reader (RFID), and a user RFID transponder card (Ofei et al., 2014). Most of the image-based methods are portable. Exceptions are the DIMS which is mounted on a small cart and the SmartPlate which is dinner-plate size and hence less convenient to carry compared to the other smaller devices.

Image-based methods can be further subdivided into three groups: image-based food records, image-assisted food records, and image-assisted 24h recalls. Image-based food records are defined as methods in which images of foods captured during eating episodes serve as the entire basis for the dietary assess­ment, whereas in image-assisted food records, the images captured are used to enhance or supplement a tradi­tional text-based food record (written or electronic). In image-assisted 24h recalls, the food images captured during eating episodes are used to aid self-reporting during the 24h recall. For more details of image-based methods, the reader is referred to the following reviews (Illner et al., 2012; Gemming et al., 2015a; Boushey et al., 2017).

Several studies have confirmed that image-based methods of dietary assess­ment are preferred over tradi­tional dietary records, and can enhance self-reported dietary intake by revealing unreported foods and misreporting errors. However, study costs will be increased when respon­dents have to be provided with a suitable device (Gemming et al., 2015a). With the additional dietary inform­ation from the images, there is likely to be an increase in reported energy intake and a decrease in reporting errors, except when the methods are followed incorrectly so the images are of poor quality, or the user forgets to capture images before eating episodes. In addition, details of mixed dishes are difficult to extract from the images, some ingredients may be hidden, and cooking methods omitted. To avoid these prob­lems, respon­dents should be trained to review the images after taking them, to take a second image if necessary, and to record descrip­tions of the mixed dishes, including lists of ingredients and quantities. Coding errors associated with image analysis, however, are likely to be less random and less prob­lematic compared to the system­atic bias observed when food type and portion size are self-reported.

Figure 3.4

depicts the entire system — the Technology Assisted Dietary Assessment (TADA) system — that commences with the respon­dent capturing an image with the mobile food record. In this method, respon­dents use the camera of a mobile devise to capture images of their intake of foods and bever­ages which they send to a secure server, where trained analysts estimate the reported intakes of energy and nutrients. Figure 3.5 compares the mean total energy expenditure (TEE based on doubly labeled water (DLW), with the reported energy intake (rEI) using images from a mobile food record (mFR) recorded over 7.5d by 45 community dwelling men and women, 21–65y (Boushey et al., 2017). In this study, mean energy intake for TEE and rEI were 2932 vs. 2353kcal/day with a significant mean difference of 580kcal/d, p < 0.0001). A moderate statis­tically significant corre­la­tion was found between rEI and TEE (Spearman corre­la­tion coefficients 0.58; p < 0.0001), with reporting accuracy that was consistent across all energy intake levels. The mFR was well received by the respon­dent and usability was rated as easy by Boushey et al. (2017, For details of the development of the mobile food record (mFR), see Six et al. (2010).

Unfor­tu­nately, there are few validation studies of image-based methods among the elderly, children, and adolescents, and the feasibility of their use in large studies (n > 100) has not yet been demonstrated. Burrows et al. (2019) provide a review of validation studies based on image-based methods using doubly labeled water.

3.2.5 Wearable devices

Under development are wearable devices worn around the neck, wrist, ear, or attached to clothing and designed to objectively measure diet without relying on user-reported food intake (Fontana et al., 2015). Some (e.g., WearSens, worn around the neck) lack the ability to detect specific types of foods, and thus quantify nutrient intakes, while others (e.g., GoBe, worn as a wristband) only measure calories, fat, protein, and carbohydrate. Several chewing and bite-related wearable sensors are also being devel­oped which aim to dif­fer­entiate between food intake and non-food intake, with some aiming to estimate energy intake (e.g., BitBite) (Fontana et al., 2015). Discomfort has been reported by some users of these wearable devices which suggests that their format needs to be modified (Boushey et al., 2017). Privacy issues also need to be addressed when using wearable devices (Gemming et al., 2015a).

3.3 Selecting an appropriate method

The method of choice for assessing food or nutrient intakes depends primarily on the objectives of the study. No method is devoid of random or system­atic errors (Chapter 5), or prevents alterations in the food habits of the respon­dents. However, statis­tical techniques are now available which are designed to mitigate the impact of measurement errors on study results. See Freedman et al. (2011), Kirkpatrick et al. (2018), and Chapter 5. The US National Cancer Institute has devel­oped a  Dietary Assessment Primer (2020), a web resource to aid researchers choose the best available dietary assess­ment approach to achieve their research objective. Readers are advised to consult the primer before selecting a self-report dietary instrument.

Box 3.2 provides guidance on the most appropriate methods for assessing food or nutrient intakes in relation to four possible levels of objectives.

Note that the number and selection of replicate 24h recalls, or weighed or estimated one day food records required to obtain level two, level three, or level four data, depends on the day-to-day vari­a­tion within one indi­vidual (i.e., within-person vari­a­tion ) of the nutrient of inter­est (Chapter 6). This vari­a­tion depends on the nutrient, the study popu­lation, and the seasonal vari­a­tions of intake. Generally, for nutrients found in high concen­trations in only a few foods, such as vitamins A and D and cholesterol, the number of replicates needed is greater than for those found in a wide range of foods (e.g., protein). Noncon­secutive days should be selected for the replicates when possible, to enhance the statis­tical power of the inform­ation: day-to-day corre­la­tions between intakes often occur when food intake data are collected over consecutive days. The length of time needed between the observation days also depends on the nutrient (IOM, 2000; a 3–10d inter­val after the previous recall is recommended (Tooze, 2020).

Additional factors that should be consid­ered when choosing a method for assessing the food con­sump­tion of indi­viduals are the characteristics of the indi­viduals within the study popu­lation, the respon­dent burden of the method, and the available resources. For instance, certain methods are unsuitable for elderly subjects with poor memories, for busy mothers with young children, or for illiterate indi­viduals. Other methods require highly trained personnel and specialized laboratory and com­put­ing facilities, which may not be available. Generally, the more accurate methods are associated with higher costs, greater respon­dent burden, and lower response rates. Unfor­tu­nately, compromises often have to be made between the collection of precise data on usual nutrient intakes of indi­viduals and a high response rate.

3.3.1 Determining the mean nutrient intake of a group: level one

Level one is the easiest objective to achieve and can be met by measuring the food intake of each subject in the group using a single 24h recall or a one day food record, provided the indi­viduals are representative of the study popu­lation and all the days of the week are proportionately repre­sented in the final sample. Data on mean usual nutrient intakes of a group can be used for inter­national compar­isons across coun­tries of the relationship of nutrient intakes to health and disease. However, this method based on a single intake day per subject should never be used for reporting the distri­bution of intakes (i.e., as percentiles of intake) (Tooze, 2020). To calculate n, the number of subjects required in the group, an estimate of the between-person vari­a­tion for the nutrient of inter­est is needed. This is usually obtained from the literature but may be determined during a pilot study.

As an example, assume the expected mean iron intake obtained from the literature is 10mg/d, with an anticipated stan­dard deviation (s) of 3mg/d. Also assume that we want to be 95% confident that the true mean lies between 9.2 to 10.8mg/d (i.e., the confi­dence inter­val has limits which are 0.8mg/d on either side of the mean). A 95% confi­dence inter­val is calculated approximately as the mean ±2 × e, where e is the stan­dard error of the mean — a measure of the precision of the estimated mean. Hence , the required e = 0.8/2 = 0.4. We can use the following formula to calculate n, the desired group size. \[\small n = s^{2}_{b} /e^{2 } \] where s2
b is the between-person variance of the nutrient of inter­est, and e is the desired stan­dard error — a measure of precision required for the estimate of the mean intake of the nutrient of inter­est. Hence \[\small n = 3^{2 } / 0.4^{2 } = 56.25 \] and so 56 subjects are required. Alter­natively, if we wanted to be 99% confident that the true mean lies between 9.2 and 10.8mg/d, then more subjects must be studied. A 99% confi­dence inter­val is calculated approximately as the mean ±3 × e. Hence the required e = 0.8/3 = 0.27 Therefore

\[\small n = 3^{2 } / 0.27^{2 } = 123 \] and 123 subjects are required.

Clearly, the size of the group (n) necessary to characterize the group mean usual nutrient intake depends on the degree of precision required: more indi­viduals must be studied to achieve higher precision. This calcu­lation of the required group size should be repeated for each of the nutrients of inter­est, and the largest n (i.e., the worst case) should be used if possible.

If the study objective is to demonstrate a significant difference in the mean intakes of two groups, or a significant change in the mean intakes, based on unpaired or paired data, then alternative formulae must be applied; details are given in (Gibson and Ferguson, 2008).

3.3.2 Calculating the popu­lation percentage “at risk": level two

To determine the percentage of the popu­lation “at risk” of inadequate nutrient intakes, an estimate of the distri­bution of usual intakes of the indi­viduals is required. This, in turn, requires that the food con­sump­tion of indi­viduals be measured over more than one day. Hence, repeated 24h recalls, or replicate weighed or estimated one day food records are the methods of choice, again ensuring that all days of the week are proportionately repre­sented in the final sample. Often, it is not feasible to carry out repeated observations on all the indi­viduals, as in the case of a national dietary survey, and the recalls or records are repeated on a sub­sample of the indi­viduals only.

To achieve a level two objective, at least two independent mea­sure­ments of food intake should be obtained on at least a representative random sub­sample of indi­viduals in the survey. The U.S. Food and Nutrition Board (IOM, 2000) recommends that the replicate mea­sure­ments should be independent and made on non-consecutive days 3–10d after the first recall or record. However, if the data can be collected only on consecutive days, then three daily mea­sure­ments should be used. The random sub­sample should consist of at least 50 indi­viduals per demo­graphic group. Where possible, the demo­graphic groups to be sampled should be defined according to the sex and life-stage groups used for the chosen Nutrient Reference Values (Deitchler et al., 2020),

Note that it is more important to have a minimum number of replicate observations in the sub­sample than a minimum proportion of replicate observations. Once the required number of replicate observations have been obtained, methods can be applied to correct for the measurement error associated with both day-to-day variability in intakes for a single person (termed within-person vari­a­tion ) and the vari­a­tion in intakes between indi­viduals (termed between-person vari­a­tion ) (see Chapter 6).

Four statis­tical methods are now available for estimating the distri­bution of usual intake. For more details, the reader is referred to (Tooze, 2020). Of the methods, the first was outlined by the National Research Council and later refined by Iowa State University (ISU) (Nusser et al., 1996) with software devel­oped to use the ISU method. Note, this method does allow for a random sub-sample of repeated 24h recalls or food records, and has the capacity to adjust for season, day of the week, and/or sequence effects, but is not recommended for general use with episod­ically con­sumed foods, food groups, or nutrients (Tooze, 2020). The adjustment process provides esti­mates of the usual nutrient intakes for each specified age and gender-specific subgroup. An example comparing the adjusted distri­butions of usual zinc intakes (using the refined NRC approach of (Nusser et al., 1996) with the observed zinc intakes for New Zealand adult females aged 19–50y is shown in Figure 3.6.

The adjustment process used yields a distri­bution with reduced variability (sometimes referred to as “shrinking”) because the within person vari­a­tion has been removed, while preserving the shape of the original observed distri­bution (Gibson et al., 2003).

Later, the National Cancer Institute (NCI) (Tooze, 2020) devel­oped a method which also adjusts for the same covariates (season, day of the week, and/or sequence), but also has the capability of incorporating covariates to identify esti­mates for sub-popu­lations in the survey. In addition, the NCI method, unlike the ISU, can be used with episod­ically con­sumed foods, food groups, or nutrients.

The European Food Con­sump­tion Validation Project have devel­oped the Multiple Source Method (Haubrock et al., 2011) (MSM) that can also be used for episod­ically con­sumed foods, food groups, or nutrients, although caution must be used when employing the program with models containing covariates. The Dutch National Food Con­sump­tion Survey 2007-2010 have devel­oped a program called the Statistical Program to Assess Dietary Exposure (SPADE) , but this program currently requires 24h recalls on at least two-days for all respon­dents in the sample and is not recommended for use with episod­ically con­sumed foods, food groups, or nutrients.

All the methods described briefly above use varying approaches to yield an adjusted distri­bution of “usual" nutrient intakes that can then be used to predict the proportion of the popu­lation at risk of nutrient inadequacy using either the full probability approach, or the Estimated Average Requirement (EAR) cutpoint method; details are given in Chapter 8b. However, the choice of which “usual intake method” method to use depends on whether foods, food groups, or nutrients are episod­ically con­sumed, whether the probability of con­sump­tion and the con­sump­tion-day amount are correlated or not, and whether replicates of the 24h recalls or food records are available on all the popu­lation groups to be surveyed or only on a sub-sample (Souverein et al., 2011).

Table 3.7: “True” and observed prevalence esti­mates and the number of repeated mea­sure­ments needed to reduce the observed prevalence to within 5% of the “true” prevalence. Data for women aged 45–54y from NHANES II (1976–1980). From: Sempos CT, Looker AC, Johnson CL, Woteki CE. (1991). The importance of within-person variability in estimating prevalence. In: Macdonald I (ed.), Monitoring Dietary Intakes, pp. 99–109. Springer-Verlag, Berlin..

Variable	Prevalence (%) "True" Observed		No. of repeated mea­sure­ments needed
Cholesterol > 300mg	15	37	39
Calcium > 800mg	12	21	9

Figure 3.6 shows that, in the example, adjusting the distri­bution significantly using the ISU method reduces the proportion of indi­viduals consid­ered to have intakes below the EAR for zinc. Within-person vari­a­tion can also have a significant effect on esti­mates of the prevalence of abnormally high nutrient intakes. Table 3.7 shows data from NHANES II. The large differences between the observed prevalence and the calculated “true” prevalence represent the effect of removing the within-person vari­a­tion by calcu­lation. In this case, very large numbers of repeated mea­sure­ments on each indi­vidual are required to reduce the observed prevalence of abnormally high intakes of cholesterol and calcium to within 5% of the true prevalence (Sempos et al., 1999). Comparable data are essential for national food policy development and food fortification planning. Food patterns associated with inadequate nutrient intakes can also be identified using this approach, enabling food assistance programs to be designed and improvements in nutrition education made.

More recently, a new method has been devel­oped based on single-day dietary data which can be used to estimate popu­lation distri­butions of usual intake of nearly-daily con­sumed foods and nutrients, provided a suitable external within-person to between-person variance is available (Luo et al., 2019). Nevertheless, researchers are urged to collect replicate data where possible.

3.3.3 Ranking indi­viduals by food or nutrient intake: level three

When the study objective is at level three and involves ranking indi­viduals within a group, often for the purpose of linking dietary intakes with risk of chronic disease, the preferred approach is to obtain multiple observations on each indi­vidual. The number of days required to achieve the level three objective can be calculated from the ratio of the within- to the between-person vari­a­tion in nutrient intakes (often termed the “variance ratio"); for more details see Chapter 6 . Sometimes an estimate of the variance ratio can be obtained from the literature, again preferably from an earlier study on a comparable group. Alter­natively, a pilot study may be necessary to obtain this inform­ation.

Several authors have devel­oped equations for calculating the number of replicate days required to meet level three objectives (Black et al., 1983; Basiotis et al., 1987; Nelson et al., 1989). Black et al. (1983) suggest using the following formula for the number of days (n) of diet records needed:

\[\small n = (r^{2} /(1 − r^{2})) × (s^{2}_{w}/s^{2}_{b})\]

In this equation, r is the unobservable corre­la­tion between the observed and true mean intakes of indi­viduals over the period of observation, and s2
w and s2
b are the observed within‑ and between-person variances, respe­ctively. This equation should be used in association with Table 3.8 which shows the proportion of indi­viduals correctly and incorrectly classified in the extreme fractions for dif­fer­ent values of the corrlation coefficient between the observed and true intakes (r). The value of r chosen will depend on the degree of misclas­sification that the inves­tigator is prepared to accept (Table 3.8).

Table 3.8: Proportion of indi­vid­uals correctly and incorrectly classified in the extreme fractions for dif­fer­ent values of the corre­la­tion coefficient (r) . (a), correctly classified in the extreme thirds, fourths, or fifths of the distri­bution of intakes; (b), misclassified into the opposite extreme fraction. From Nelson et al. (1989).

		Correctly and incorrectly classified into extreme fraction
r		Thirds	Fourths	Fifths
0.75	a b	0.69 0.049	0.63 0.013	0.59 0.004
0.80	a b	0.72 0.033	0.68 0.006	0.65 0.002
0.85	a b	0.76 0.018	0.72 0.002	0.69 <0.001
0.90	a b	0.80 0.006	0.77 <0.001	0.75 <0.001
0.95	a b	0.86 <0.001	0.84 <0.001	0.83 <0.001

As an example, assume that the inves­tigator requires that when the indi­viduals are divided into terciles, fewer than 5% (< 0.05) of the indi­viduals are grossly misclassified into the opposite tercile. This will require an r value of 0.75 (Table 3.8). Assuming s2
w / s2
b = 1.7, then the number of days (n) \[ \small n = (r^{2}/(1–r^{2}))× 1.7\] \[ \small n = 0.75^{2}/(1–0.75^{2}) × 1.7\] \[ \small n = \mbox{3 days}\]

The number of days needed to generate a given r increases as the chosen r increases. If the size of the within-person vari­a­tion (s2
w ) in nutrient intake is small compared with the size of the between-person (s2
b) vari­a­tion , then fewer replicate days are needed to meet level three objectives.

The U.S. subcommittee on criteria for dietary evaluation (NRC, 1986) recommended using independent days for replicating the mea­sure­ments of one day nutrient intakes to reduce any effect of autocorre­la­tion between intakes on adjacent days.

An alternative approach to achieving level three objectives is to use a semi-quanti­tative food fre­quency question­naire. This approach is often used in epidemio­logical investigations to study associations between intakes and risk of disease and does not require a measurement of absolute nutrient intakes. Although this approach is much simpler, involving only a single inter­view with each subject, it is difficult to quantify the errors involved and to separate the effects of within- and between-person variance.

3.3.3 Determining usual intakes of nutrients of indi­viduals: level four

Reliable esti­mates of usual food or nutrient intakes of indi­viduals that can be used with confi­dence to meet a level four objective, involving corre­la­tion or regression analysis with indi­vidual biochemical measures, are the most difficult to obtain. Large numbers of measurement days for each indi­vidual are required using 24h recalls or estimated or weighed food records.

An estimate of the within-person vari­a­tion for each nutrient of inter­est should be obtained from the literature, preferably from an earlier study on a comparable group or a pilot study, as noted earlier. This estimate may be expressed as the variance, s2
w ; stan­dard deviation, s_w; or as the coefficient of vari­a­tion (CV_w) expressed as a percentage: \[\small CV_{w}= s_{w}/ (\mbox{mean intake}) × \mbox{100%}\]

This estimate can be used in the following equation to determine the number of days required per indi­vidual to estimate an indi­vidual's nutrient intake to within 20% of their true mean 95% of the time (Beaton et al., 1979).

\[\small n = (Z_{α}CV_{w} / D_{0})^{2}\]

where n = the number of days needed per indi­vidual, Z_α = the normal deviate for the percentage of times the measured value should be within a specified limit (i.e., 1.96 in the example below), CV_w = the within-person coefficient of vari­a­tion (as a percentage), and D₀ = the specified limit (as a percentage of long-term true usual intake) (i.e., 20% in the example given below).

The following example calculates the number of days required to estimate a Malawian woman's zinc intake using 24h recalls to within 20% of the true mean, 95% of the time. In this example, the CV_w (i.e., 34%) for zinc intakes on Malawian women via 24h recalls is taken from the literature (IZiNCG, 2004). Thus if Z_α = 1.96 and CV_w = 34%. then:

\[\small n = (1.96 × 34\% / 20\%)^{2} = 11 d \]

If a pilot study is undertaken in which replicate 24h recalls are conducted, then the actual CV_w for each nutrient of inter­est can be calculated. In this way, the estimate of the number of days required to measure the usual intake of each of the nutrients of inter­est in an indi­vidual, with a required degree of precision, can be defined. In general, consid­erably more days are required to obtain reliable esti­mates of intakes of indi­viduals to meet the level four objective, compared with level three (i.e., relative ranking of subjects into groups) (Palaniappan et al., 2003).

Sometimes, dietary histories or semi-quanti­tative food fre­quency question­naires are used to obtain this level four data on usual nutrient intakes for corre­la­tion with biomarkers (Jacques et al., 1993). Some inves­tigators emphasize, however, that the accuracy of a semi-quanti­tative food fre­quency question­naire is only equivalent to two to three repeat 24h recalls (Sempos et al., 1999).

In some experimentally controlled studies such as balance studies, inform­ation on the actual nutrient intakes of an indi­vidual over a finite time period are required. For such data, weighed food records (Section3. 1.4), completed for the duration of the study period, are the recommended method. Nutrient intakes can then be calculated using food composition data. Alter­natively, duplicate meals can be collected throughout the period for later analysis; details are given in Chapter 4.

3.3.4 Combining dietary instruments

Recognition of the limitation of even two 24h recalls to adequately measure, at the indi­vidual level, usual intake of foods or nutrients that are not con­sumed daily (i.e., episod­ically con­sumed foods) has led to the use of more than one dietary instrument in a given study. In this way, the advantages of each method can be exploited and their weaknesses minimized. Consequently, in large scale epidemio­logical studies, a common approach is the admin­istration of a food fre­quency question­naire to all respon­dents, and 24h recalls or food records in a sub­sample. The food fre­quency question­naire provides inform­ation on foods that are con­sumed less frequently (often, “in the past year") not given by the 24h recalls while the 24h recalls yield richer detail by attempting to correctly quantify portion size for each eating occasion (Subar et al., 2006). Alter­natively, a food fre­quency question­naire could be used as a supplement to one or more quanti­tative 24h recalls admin­istered to the full sample, providing richer detail and less system­atic bias. This approach appears to have the most added value when the research question of inter­est is to estimate diet-health relationships, espe­cially when the estimation of usual dietary intake at the indi­vidual level involves a food, food group, or nutrient that is con­sumed episod­ically (Tooze, 2020).

Statistical methods have been devel­oped to combine the dietary data from repeated 24h recalls and food fre­quency question­naires. Conceptually, these statis­tical methods presume that the usual food intake of an indi­vidual equals the probability of consuming a food on a given day (queried as fre­quency of usual intake over a specified time period), multiplied by the average amount of intake of that food on a typical con­sump­tion day. Repeated 24h recalls from the same indi­vidual yield inform­ation on con­sump­tion probability and amount, while the food fre­quency question­naire provides inform­ation on intake fre­quency of rarely con­sumed foods and usual intakes of nutrients when portion size information is available (Conrad and Nothlings, 2017). The reported food frequencies and nutrient intakes (if available) can be used as covariates in both steps of the NCI method (Tooze, 2020) and with MSM (Haubrock et al., 2011) to enhance the estimation of usual intakes from the 24h recall data, espe­cially from foods that are not con­sumed every day. Note, however, that in low income coun­tries where context-specific food fre­quency question­naires are often not available or inappropriate in the study setting, it may be preferable to increase the number of repeat 24h recalls per respon­dent rather than embarking on the development, validation, and collection of dietary data using a food fre­quency question­naire in addition to using the 24h recall method. For more details, the reader is advised to consult (Tooze, 2020). at: Intake.org

3.3.5 Surveys of indi­vidual food consumption at the global level

FAO/WHO Global Individual Food consumption data Tool (FAO/WHO GIFT) is a global data­base of surveys of food consumption at the indi­vidual-level which are collected at the national and sub­national level throughout the world. Only data collected from 1980 by quanti­tative methods such as 24h recalls, food records (weighed or estimated), 12hr recalls, and direct food weighing are considered. Portion sizes of all foods and beverages con­sumed by each survey participant must be assessed, including water if possible. For low‑ or middle-income coun­tries, surveys comprising at least 100 subjects (with no evidence of strong selection bias) are included, whereas for high-income coun­tries, surveys that are nationally repre­sen­tative are an additional criterion. The dietary data generated from FAO/WHO GIFT can be used to inform agricul­tural, nutrition, food safety and environmental policies and programs. A summary of 218 dietary surveys performed in low‑and middle-income coun­tries from 1980 to 2019 and included in the FAO/WHO GIFT inventory is available in de Quadros et al. (2022).

FAO/WHO GIFT apply several criteria to validate the dietary data­bases included. For example, for all foods and drinks con­sumed by each survey participant on each survey day, a complete descrip­tion, the amount con­sumed, and energy and nutrient values for each item must be included in the datasets. Additional compulsory variables are the age and sex of each subject, geo­graph­ical location (country, and region(s) if available), type of area (e.g., rural, urban), and the number of survey days recorded per subject or for a subset of subjects. Inclusion of variables based on anthropometric data such as weight and height and the physiological status of both women (e.g., pregnant, lactation) and infants (i.e., breast­feeding status) are also recommended.

Recipes should be disaggregated whenever possible to provide data on the quantity of each separate ingredient, cooking method used, total amount prepared, number of people served from the recipe, and the quantity con­sumed by the survey participant (or served and leftover). Only with this information can each ingredient be attributed to their appropriate food group for the calculation of the FAO/WHO GIFT indicators and summary statistics.

The food composition values used are provided by the data providers. However, all indi­vidual quanti­tative dietary datasets shared through FAO/WHO GIFT are coded with the FoodEx2 system, a comprehensive and flexible food classification and descrip­tion system. Eligible dietary data­bases are mapped manually by the inves­tigator with FoodEx2 codes for food groups (n=24) and food subgroups based on the descrip­tions provided by FAO/WHO GIFT. FoodEx2 was first devel­oped by the European Food Safety Authority (EFSA) and was later scaled up to the global level in collaboration with FAO and the World Health Organi­zation (WHO). The use of this common food classification and descrip­tion system among dietary surveys from dif­fer­ent coun­tries contributes to the global harmoni­zation of dietary data. Prior to the analysis and format­ting by FAO/WHO GIFT, all eligible datasets are screened for potential errors, missing values and outliers. For further details see FAO/WHO GIFT (Methods)

Provided the eligible dietary datasets have been coded using FoodEx2, the FAO/WHO GIFT platform can compute ready-to-use indicators and summary statistics based on indi­vidual food consumption data in the areas of food consumption, food safety and nutrition. For example, of the indicators, one reflecting dietary diversity at the popu­lation level (a key component of diet quality) entitled the Minimum Dietary Diversity for Women (MDD-W) is computed. This is a food group-based indicator that estimates the proportion of non-pregnant women of repro­ductive age (15‑49y) who con­sumed at least five out of ten defined food groups over the previous 24h.

The summary statistics generated by FAO/WHO GIFT comprise the estimated usual intakes of selected nutrients for pre-defined popu­lation groups by sex and age (except children less than aged 12 months) provided the datasets contain multiple non-consecutive days of 24h recalls / food records for at least a subset of 50 indi­viduals. The statis­tical program chosen to adjust the dietary data for day-to-day vari­a­tion is the Statistical Program to Assess Dietary Exposure (SPADE) by Dekkers et al. (2014). In addition, information on Nutrient Reference Values (NRVs) set by FAO/WHO, the European Food Safety Authority (EFSA) and the Institute of Medicine (IOM) are provided on the FAO/WHO GIFT platform to facilitate comparison of the estimated usual intakes of selected nutrients with NRVs. For more details of the statis­tical program, see (SPADE) and the instruction manual (SPADE - 4100). For NRVs and their applications, see Chapters 8a and 8b, respectively.

The Global Dietary Database (GDD) aims to identify, compile, and standardize indi­vidual-level data on dietary factors related to maternal-child health and chronic diseases. Currently the GDD comprises surveys across 188 coun­tries conducted between 1980 and 2018 and provides empirical evidence on dietary intakes both across and within coun­tries worldwide. Priority is given to nationally or sub-nationally representative dietary surveys based on 24-h recalls, food fre­quency question­naires or short standardized question­naires (e.g., Demo­graphic Health Surveys (DHS). Household-level surveys are included if indi­vidual-level surveys are not available in a country and converted to indi­vidual-level intakes within each household using Adult Male Equivalents (AME), also known as Adult Consumption Equivalents. See: Weisell & Dop (2012). These authors account for the household composition and differing energy intakes by age and sex of household members. See Coates et al. (2017) for discussion of the validity of the application of the AME method for household-level data.

The FoodEx2 categorization system is used to standardize the descrip­tion and classification of foods into food groups. Standardization also includes categorising nutrients and their units; quality assess­ment; aggregation by demo­graphic strata and energy adjustment. For more details on data extraction and standardization, see: (GDD)

Mean intakes of 54 dietary factors by country, year, age, sex, education, urbanicity, and pregnancy / lactation status within nations can now be estimated using the GDD prediction model in 188 coun­tries / territories in Asia, Asia-Pacific high-income coun­tries, Oceania, Former Soviet Union, Latin America and Caribbean, Middle East and North Africa, South Asia, Sub-Saharan Africa, and the Western high-income coun­tries in Australasia, Europe and North America (Miller et al., 2021). The dietary factors selected and defined based on evidence for relationships with maternal-child health or chronic diseases include 14 foods, 7 beverages, 12 macro­nutrients, and 18 micro­nutrients. See (GDD) for more details on data extraction and the standardization used for estimating dietary intakes. Several indicators of global dietary patterns (e.g., Alternative Healthy Eating Index (AHEI); Dietary Approaches to Stop Hypertension (DASH), and the Mediterranean Diet Score (MED)) among children and adults have also been compiled from the GDD sets and compared globally, regionally, and nationally (Miller et al. ,2022).

Institute for Health Metrics and Evaluation (IHME) initiative data­base aims to provide rigorous and comparable measurement of the world’s most important health problems and evaluates the strategies used to prevent them. IHME uses FAO Food Balance Sheet estimates, national product sales, household surveys, and data based on 24h recalls (considered the gold standard). Datasets created by IHME are stored in the IHME data catalogue known as the Global Health Data Exchange and can be freely downloaded from the IHME website.

IHME provides freely available modeled data by country, age, sex and year, based on primary data collected in 204 coun­tries and 87 indicators, including 15 dietary indicators (9 foods and 6 nutrients). The dietary indicators are included in GBD 2017, a worldwide observational epidemiological study that tracks the progress within and between coun­tries of the changing health challenges (Lim et al., 2013).

Acknowledgements

The assistance of Nutrition International in the preparation of this web-page is gratefully acknowledged. RSG would like to thank past zinc collaborators, particularly my former graduate students, and is grateful to Michael Jory for the HTML design and his tireless work in directing the trans­ition to this HTML version.