INTERMAP: the dietary data—process and quality control

Abstract

The aim of this report is to describe INTERMAP standardized procedures for assessing dietary intake of 4680 individuals from 17 population samples in China, Japan, UK and USA: Based on a common Protocol and Manuals of Operations, standardized collection by centrally trained certified staff of four 24 h dietary recalls, two timed 24-h urines, two 7-day histories of daily alcohol intake per participant; tape recording of all dietary interviews, and use of multiple methods for ongoing quality control of dietary data collection and processing (local, national, and international); one central laboratory for urine analyses; review, update, expansion of available databases for four countries to produce comparable data on 76 nutrients for all reported foods; use of these databases at international coordinating centres to compute nutrient composition. Chinese participants reported 2257 foods; Japanese, 2931; and UK, 3963. In US, use was made of 17 000 food items in the online automated Nutrition Data System. Average time/recall ranged from 22 min for China to 31 min for UK. Among indicators of dietary data quality, coding error rates (from recoding 10% random samples of recalls) were 2.3% for China, 1.4% for Japan, and UK; an analogous US procedure (re-entry of recalls into computer from tape recordings) also yielded low discrepancy rates. Average scores on assessment of taped dietary interviews were high, 40.4 (Japan) to 45.3 (China) (highest possible score: 48); correlations between urinary and dietary nutrient values—similar for men and women— were, for all 4680 participants, 0.51 for total protein, range across countries 0.40−0.52; 0.55 for potassium, range 0.30−0.58; 0.42 for sodium, range 0.33−0.46. The updated dietary databases are valuable international resources. Dietary quality control procedures yielded data generally indicative of high quality performance in the four countries. These procedures were time consuming. Ongoing recoding of random samples of recalls is deemed essential. Use of tape recorded dietary interviews contributed to quality control, despite feasibility problems, deemed remediable by Protocol modification. For quality assessment, use of correlation data on dietary and urinary nutrient values yielded meaningful findings, including evidence of special difficulties in assessing sodium intake by dietary methods.

Introduction

INTERMAP was undertaken to answer key questions on the relation of multiple nutrients—particularly macronutrients, also micronutrients—to worldwide patterns of adverse blood pressure (BP). Details of the study background, general and specific aims, design and methods, hypotheses are reported elsewhere in this issue of the Journal of Human Hypertension.¹ The underlying concept is that many nutrients have ‘small’ effects on BP that in aggregate have an important impact. Under these conditions, nutritional assessment to detect these effects across individuals requires high precision at every stage, including collection of dietary data, processing, and nutrient calculation. This report focuses on INTERMAP standardized procedures—process and quality control—for assessing dietary intake of individuals from 17 diverse population samples in four countries: China, Japan, UK, and USA.

Methods

Populations were deliberately selected to be diverse, within and across countries. Samples were identified from industrial workers, farmers, residential populations, and community-based subscribers to medical practices.¹ This diversity presented challenges in implementing a common Protocol for collection and processing of high-quality standardized data on intake of all foods, beverages, and dietary supplements by every one of the 4680 participants (men and women aged 40–59 years). To meet these challenges, anticipated from the outset, the INTERMAP Protocol, General Manual of Operations, and Nutrition Manual of Operations, and its four country-specific supplementary Manuals of Operations set down—prior to study onset—comprehensive stipulations on procedures and their quality control.

Key Protocol stipulations

Key procedures required by the common Protocol were:

• Four 24-h dietary recalls, two on consecutive days and then another two on consecutive days approximately 3 weeks later (window 2–6 weeks), and two timed 24-h urine collections corresponding temporally to the first and second pair of 24-h dietary recalls.

• Organization to include an international nutrition coordinating centre, supervising nutritionists at country and site levels, and certified dietary interviewers.

• Country-specific training and certification of all staff by international and national senior researchers, based on standard procedures adapted by each country (forms, manuals, materials, and quality control procedures).

• Complete nutrient databases and software to calculate nutrients.

Collection of dietary data—process and quality control

The 24-h dietary recall method: The 24-h dietary recall method was selected because of its nearly universal applicability across populations with varying literacy skills and its relatively low burden for participants. Single 24-h recalls are limited and generally inadequate for assessing the diet of individuals.^1–8 Multiple recalls enable measurement of intra individual variability in nutrient intake, and use of mean values yields greater reliability for each person. They also reduce inter-individual measurement variability. Consequently, correlation/regression coefficients relating nutrients to outcome variables are less attenuated, for example, for nutrients and BP. Multiple visits, however, may adversely affect participation and increase costs. Use of four 24-h recalls/person was considered to be a reasonable compromise in this situation. Each recall ascertained in depth all foods, nonalcoholic and alcoholic beverages, and dietary supplements consumed over the prior 24-h. Procedures for collection, processing, quality control, and computerization of data on dietary supplements are being reported separately.

At the first and third visits, information was obtained on amount and type of alcoholic beverages ingested each day over the preceding 7 days; these data were in addition to those on alcohol intake from the four 24-h dietary recalls.

To accrue data of optimal validity, and to assess quality of the dietary data on sodium, potassium, and total protein, two timed 24-h urine collections were made, bracketed by the times of the first and second pair of 24-h recalls, respectively. To achieve accuracy in timing, urine collection was begun and was completed at the research centre, before measurement of BP. Several techniques were used to assure completeness of urine collections.^1,9

Based on the international Protocol and Manual of Operations for dietary assessment, country-specific adaptations were made to accommodate differences in food intake habits, nature of the populations, and procedures for coding dietary recall data. Study international and national leadership were responsible jointly for overseeing adherence to the Protocol and Manuals of Operations, approving adaptations, and training and certification of key staff.

Main responsibilities of the national supervising nutritionist were to develop country-specific procedures, forms, coding rules; acquire food composition data; participate in training and certification of interviewers, Site Nutritionists, and coders; provide ongoing support for each centre; and conduct second-level quality control of dietary data collection, supplementary to first-level local quality control by Site Nutritionists (Tables 1 and 2).

Table 1.

Summary of INTERMAP dietary data quality control procedures for China, Japan, and UK

Local level

Site interviewer

1. Takes recall, tape records it (with participant permission)

2. Reviews recall and generates food inquiries

3. Codes recall and generates new food requests

4. Verifies outliers

Site nutritionist

1. Inspects 100% of recalls and provides feedback

2. Recodes one randomly selected recall per interviewer per week (at least 10% of all recalls) and provides feedback

3. Verifies outliers and data entry

4. Reviews one randomly selected taped 24-h recall per interviewer per week for 12 criteria and provides feedback based on score

5. Verifies new food requests and sends them to Country Nutritionist

6. Prepares diskette with completed recalls in batches of 30

7. Generates tracking form

8. Forwards diskette and tracking form to Country Nutritionist

National level

Country nutritionist

1. Logs in batches of 30 recalls

2. Recodes randomly selected 10% and calculates line error rate (if batch fails, it is returned to site and recoded)

3. Computes nutrients and verifies outliers

4. Processes new food requests; transmits them to International Nutrition Coordinator as needed

5. Evaluates randomly selected 10% of tapes scored by Site Nutritionist and provides feedback

6. Revises coding manual as required, including addition of new coding rules and of new codes for new foods

7. Prepares diskette and tracking form, and transmits to International Coordinating Centers

8. Submits monthly quality control report to International Coordinating Centers

International level

International nutrition coordinator

1. Reviews quality control reports and follows up with Country Nutritionists

2. Coordinates new food requests with Nutrition Coordinating Center

3. As member of Steering and Editorial Committee, reviews and acts on proposed modifications in Manuals of Operations

Table 2.

Summary of INTERMAP dietary data quality control procedures for the USA

Local level

Site interviewer

1. Takes recall, tape records it (with participant permission)

2. Responds to standardized computer probes, including computer queries on possible excessive amounts

3. Generates food inquiries to cook to gather additional information

4. Generates new food requests for coding foods not in database

5. Reviews nutrients per food item (kilocalories, protein, fat, sodium, gram weight)

6. Reviews daily nutrient totals for kilocalories and sodium

Site nutritionist

1. Reviews one randomly selected taped 24-h recall per interviewer per week for 12 criteria and provides feedback based on score

2. Re-enters randomly selected taped recalls into NDS (one/week/site)

3. Compares re-entered and original recall and follows up with interviewers

National level

Country nutritionist

1. Evaluates randomly selected taped 24-h recalls for 12 criteria (two tapes/site/week)

2. Re-enters randomly selected taped recalls into NDS (one tape/site/week)

3. Compares re-entered and original recalls and follows up with Site Nutritionists

4. Reviews computerized recalls for food and nutrient outliers

5. Reviews key nutrients from four recalls/person

6. Prepares monthly QC report and sends it to International Nutrition Coordinator

7. Works with International Coordinating Center senior staff in Chicago to prepare data in final form for transmission to Coordinating Center in LondonInternational

International level

International nutrition coordinator

1. Reviews quality control reports and follows up with Country Nutritionist

2. Coordinates new food requests with Nutrition Coordinating Center

3. As member of Steering and Editorial Committee, reviews and acts on proposed modifications in Manuals of Operations

Each centre had to successfully complete a dry run of the study Protocol involving 10 volunteers before it was certified to collect study data. This dry run involved on-site immediate feedback from an international training team.

Final data entry, editing, and analysis have been the responsibility primarily of the INTERMAP Coordinating and Statistical Centre in London, in concert with the International Coordinating Centre in Chicago.

A standard 24-h dietary recall ‘multiple pass’ procedure was implemented in all centres. This procedure was developed initially for LRC and MRFIT studies^10,11 and used in PRC–USA joint studies.^12,13 In this procedure, the participant reported in an unprompted manner times and places of eating, and all foods and beverages ingested on the previous day, without particular attention to quantities and other details. This list was first reviewed for completeness of items. Then the interviewer went back over the list again with the participant, to record details (eg brands of foods, quantities, processing methods, additions in cooking and/or at table, amounts left on plates), and to recheck for areas left unclear or unspecified (eg when there was a big time gap between meals). The participant was queried as to further details, and asked to respond as to any incorrectly interpreted or forgotten information (eg a snack or meal not included). When the participant was uncertain about details, the interviewer attempted to get information either by contacting the cook or asking the participant to do so. To assist the participant in specifying and quantifying foods, various aids were used, including fresh foods of varied standardized portion sizes (eg small, medium, and large apples), food models, pictures, containers of various types and sizes (spoons, glasses, cups, bowls, plates, commercial cans, and packets). These were specific to each country, standardized within country.

In China, individual ingredients in mixed dishes were estimated by the participant, and then the portion of the mixture consumed by the participant was estimated. If foods reported were prepared by someone other than the participant, their ingredients were verified with the cook. Salt used in food preparation was estimated by the participant; the estimated amount was then weighed and recorded.

In Japan, aids included life-like models for specific foods with calibrated lines, calibrated photographs of foods on standard plates, cardboard shapes, a measuring rule, a drawing board with grid lines, real sugar cubes, coffee whitener, and weighing scales. Soy sauce and salt added to foods at table were estimated by the participant using real soy sauce and salt, weighed, and recorded.

In the UK, use was made of food models, standard containers with calibrated lines, calibrated photographs of foods on standard plates, cardboard shapes, a measuring rule, a drawing board with grid lines, photographs of food portions, and selected foods to aid in assessment of amounts eaten (eg dried beans, to estimate a handful; standardized cola cans; small/medium/large apples).¹⁴ Discretionary salt was not measured but included in the definition of the food if salt was used in preparation in the kitchen and/or at table (eg potatoes boiled in salt or salt added at table coded as potatoes boiled in salt). Unusual practices were recorded, for example, tea taken black was coded as no milk added.

In the USA, the online Nutrient Data System (NDS) (version 2.91), Nutrition Coordinating Centre (NCC), University of Minnesota was used.¹⁵ NDS enabled direct ongoing entry of 24-h recall data by the interviewer, and its systematic prompts served to enhance collection of full information on details of foods and beverages consumed. NDS accepted a variety of units. To help participants describe amounts, paper shapes with dimensions, rulers, graph paper, measuring cups, and a limited number of food models were used. Discretionary salt was not assessed but was included in the definition of the food if salt was used in preparation or at table.

Indicators of performance quality:

In accordance with prior detailed stipulations for quality control procedures, the five indicators of performance quality were: (1) evaluation of audiotapes; (2) number of lines (foods) coded; (3) coding error rates; (4) duration of recall portion of the interview; and (5) total kilocalories per 24-h recall.

Evaluation of audiotapes:

The Protocol specified that each dietary recall be tape recorded, with participant permission. Using 12 prior criteria (privacy of interview, general manner of interviewer, introduction by interviewer, use by interviewer of nondirected questioning, pacing, manner of questioning, objectivity, probing, use of models, documentation, memory aids, review of recall), the Site Nutritionist was to evaluate on a four-point scale at least one randomly selected tape/interviewer/week. The assessment—1. retrain; 2. needs work; 3. acceptable; 4. excellent—was to be fed back promptly to the interviewer. A score of 1 was to lead to prompt retraining. When there was a low score (1 or 2) on any criterion, the Site Nutritionist was to proceed to evaluate two tapes per week from that interviewer until deficiencies were corrected. In addition, the Country Nutritionist was to evaluate a 10% random sample of tapes evaluated by the Site Nutritionist. Divergent assessments were to be discussed with the Site Nutritionist who was to evaluate additional tapes, for review by the Country Nutritionist.

This procedure was carried out in UK and USA. In China also, audiotapes were evaluated by Site Nutritionists. However, it was not feasible for the Country Nutritionist to evaluate the volume of tapes generated during the short time period (3 months) of data collection at each Chinese centre, and provide immediate feedback. Instead, the Country Nutritionist visited each site early during field work and provided on-the-spot feedback.

In Japan, the Site Nutritionist evaluated audiotapes except during a limited peak period in one centre, where the Site Nutritionist observed each interviewer and provided feedback. This procedure was necessary because of the large number of interviewers. The Country Nutritionist evaluated audiotapes according to the Protocol and provided feedback to the Site Nutritionists.

Quality control of coding of reported foods/beverages:

The Protocol provided that at the local level the Site Nutritionist check 100% of recalls, recode one/interviewer/week with immediate feedback to the interviewer, and forward cleared coded recalls to the Country Nutritionist (Table 1). At the country level, coded recalls were logged into batches of 30 (15 pairs), with a 10% random sample (three recalls) selected for recoding by the Country Nutritionist. Per cent line error was to be calculated as: (sum of lines with error/total lines of coding) ×100. If this error rate exceeded 6%, the batch was to be recoded locally and subjected to this quality control procedure again until the batch passed. After coding, range checks on serving sizes and key nutrients were used to flag unusual amounts. When such unusual values could not be documented or explained, the Protocol stipulated participant exclusion.

Foods that could not be coded to existing codes generated a local ‘new food request’ form that was forwarded to the Country Nutritionist. NCC standards were used to determine whether a close match in the country database was suitable for coding this food (eg ± 85 kcal, ± 5 g protein, ± 2.5 g fat, ± 10 g carbohydrate), or whether a new food designation with entry into the database was needed. This decision was made by the Country Nutritionist or the NCC. NCC worked with the Country Nutritionist on ascertainment of nutrient composition of each new food.

Modifications of these procedures were as follows:

In China, in addition to the above procedures, any food not actually prepared by the participant was verified with the cook. Interviewers coded forms on the day of the recall. A 10% random sample was recoded locally by the Site Nutritionist. A 10% random sample was also recoded by the Country Nutritionist.

In Japan, recall data were entered on paper forms and coded with use of an interactive software system that flagged out-of-range amounts of reported foods. Recipes in the software system were utilized to enter standard seasonings and fats used in preparation. Range values for key nutrients for men and women were posted in the coding room, and interviewers and nutritionists checked the recall data when the software identified out-of-range values. Recalls were not recoded, but Site Nutritionists and assistant Site Nutritionists checked all recalls by comparing original handwritten paper forms and printed data from the data entry system. The Country Nutritionist used an independent 10% random sample to compare original handwritten paper forms and printed data while listening to audiotapes. Line error rates were calculated as described above.

In the UK, dietary recalls were collected on paper forms and coded from an INTERMAP codebook derived from FOODBASE food codes;¹⁶ this code-book was updated sequentially to accommodate current dietary practices of participants (see the section Results). Since participants often used volumetric aids to estimate amounts, and FOOD-BASE accepted only gram weights, conversion of volume to grams based on densities had to be done for many foods. Foods not in the database were frequently reported, requiring interruption of coding to ascertain nutrient composition and then update the code book. Consequently, dietary recalls were checked up to four times during quality control and coding checks. As first-phase quality control of coding, with every 10 recalls completed by a coder, the Site Nutritionist on an ongoing basis promptly recoded a randomly selected recall. If coding changes exceeded 6% of entered items, all 10 recalls were recoded. With final completion of a batch of 30 recalls, the Site Nutritionist recoded a random sample of three, with use of the same criteria for acceptance/rejection/recoding.

In the USA, collection and processing of the dietary data were made using the NDS.¹⁵ With this interactive system for online coding, the dietary interviewer entered each recall item directly into the computer. Computer prompts elicited responses that enhanced completeness and specificity of items in the recall. Out-of-range quantities were flagged for prompt quality control. Since the NDS is a fully automated interactive system, quality control procedures in the USA (Table 2) differed from the foregoing in the following ways: During the interview and input of the dietary recall, standardized computer displayed probes and queries communicated with the interviewer in relation to items entered. Immediate responses to them served to enhance data quality on amounts and specifications of foods/beverages entered. As the NDS provided immediate nutrient calculations, following the interview printouts of key nutrients per food item were reviewed (kilocalories, protein, fat, sodium, gram weight) as were daily nutrient totals for kilocalories and sodium. Exceptionally high or low intakes were checked and verified, or the recall was designated inaccurate; if no reasonable explanation could be ascertained, the recall was designated as unsatisfactory and the participant was excluded, per Protocol stipulation. As with other countries, interviews were tape recorded (with rare exceptions due to participant demurring), and randomly selected tapes per interviewer were reviewed by the Site Nutritionist for quality. In addition, randomly selected recalls were re-entered into the NDS from the information on tape (one/week/site). Food items, amounts, and key nutrients were compared between the re-entered and the original recall. While this process was analogous to the recoding conducted in the other three countries, error rates are not comparable.

Country-specific databases on nutrient composition of reported foods/beverages:

For each of the four countries, INTERMAP utilized a separate database on nutrient composition of all reported foods/beverages. NCC served as technical consultant and expert to complete these, working with the four Country Nutritionists and international leadership. Nationally available databases had to be updated and expanded to achieve optimal validity, completeness, and comparability of nutrient data. Missing data were imputed using standard procedures.^17–22

For China, initial nutrient data were from the 1991/1992 revisions of national food tables²³ as adapted by the PRC–USA Collaborative Study,^12,13 with additional data from manufacturers and recipe calculations. Nutrients were imputed from raw food values for cooked foods based on standard yield factors available for most foods.

For Japan, initial nutrient data were those in the 4th edition of Standard Tables of Food Composition in Japan, and its additional databases.^24–29 These standard food tables gave nutrient variables for raw foods. Based on these, INTERMAP nutritionists imputed nutrient composition for cooked meats, fish, and vegetables. Missing values were also imputed.

For the UK, initial data (in FOODBASE) on nutrient composition of reported foods/beverages were from the 5th edition (1991) of McCance and Widdowson’s national food tables, including published supplements to the tables up to 1998, additional fatty acid data (mostly from chemical analyses for FOODBASE), and data from manufacturers for any new food that could not be matched to an existing food code.^30–36 The FOODBASE program automated calculation of compositional fats and fats used in food preparation.¹⁶ Data on amino acid composition of all foods were added by NCC.

For the USA, use was made of the extensive NCC database on nutrient composition of foods, encompassing data on 104 nutrients.³⁷

Use of 24-h urinary excretion of sodium, potassium, and urea for quality assessment of dietary data:

Each INTERMAP participant collected two timed 24-h urine specimens, after Protocol-specified detailed briefing and instruction. The first collection began at the research centre at the first visit, and ended there the following day at the second visit. The second urine collection occurred 3–6 weeks later, in like fashion, related to visits three and four. Thus, each 24-h specimen coincided with a pair of 24-h dietary recalls. The INTERMAP Protocol specified criteria for assessment of completeness of 24-h urine collections, and provided for rejection of incomplete collections. These criteria are not all-encompassing for assurance of complete collections; there are no data to document completeness of collection unequivocally. Nevertheless, for 24-h dietary intake of sodium, potassium, and total protein, 24-h urinary excretions of Na, K, and urea serve as relatively objective measures.^5,38 They were therefore defined in the INTERMAP Protocol as criteria for quality assessment of 24-h recall data. Computations to estimate dietary total protein from urinary urea were:

• 24-h urinary urea (g/day) × 0.46667 = 24-h urinary urea nitrogen (N) (g/day),

• 24-h urinary urea N (g/day) × 1.21787 = 24-h urinary total N (g/day)³⁹

• 24-h urinary total N (g/day) × 6.25 = 24-h estimated dietary total protein intake (g/day).⁴⁰

Alternative computations were:

• body weight (kg) × 0.031 N = 24-h urinary non-urea (g/day),

• 24-h urinary non-urea N (g/day) + 24-h urinary urea N (g/day) = 24-h urinary total N (g/day),

• 24-h urinary total N (g/day) × 6.25 = 24-h estimated dietary total protein intake (g/day).⁴¹

Statistical processing and analysis of dietary data at the INTERMAP International Coordinating Centres:

The London International Coordinating Centre has ongoing responsibility for final entry, review, editing, and analyses of data, including conversion of 24-h dietary recall data into nutrients, storage and security of all data, and generation of descriptive and analytic statistics on INTERMAP scientific aims. In this effort, it is aided by senior colleagues at the Chicago International Coordinating Centre.

Results

As expected, INTERMAP documented substantial diversity in macro- and micronutrient intake—intraindividual (day-to-day), interindividual, and interpopulation.^42–46 Here the focus is on INTERMAP experience with the process of acquiring these data under stipulated research conditions encompassing high-level standardization, comparability, and quality control, as described above in the section Methods.

The Four dietary databases on nutrient composition of all foods and beverages consumed by INTERMAP participants from China, Japan, UK, and USA

As noted in the Methods section, for each of the four countries, the starting point in the creation of the INTERMAP dietary database was an already existent national database. For China, Japan, and the UK, compilation of the INTERMAP database was an extensive process. It encompassed national/international joint review and update of available data to assure optimal quality (eg, in regard to amino acid content of all UK foods); imputation of missing nutrient values for many foods; addition of nutrients not present in original tables (eg, omega-3, omega-6, trans fatty acids); incorporation of a full set of nutrient values for new foods not in original tables; extension of databases to include nutrient composition of cooked foods; conversions to achieve complete or virtually complete comparability of nutrient data (eg, for carbohydrates, fatty acids) across countries.

For China, the original database encompassed 1455 foods. The INTERMAP effort included addition of nutrient composition for cooked versions of 681 of these foods, since the original tables gave data only for raw versions. Further, 121 new foods were incorporated into the database, to make a total of 2257 foods (Table 3).

Table 3.

INTERMAP dietary databases on nutrient composition of Chinese, Japanese, UK, and US foods

Variable	China	Japan	UK	USA
Number of population samples and participants	3,839	4,1145	2,501	8,2195
Number of food codes in INTERMAP dietary database for each country	2257a	2931b	3963c	17 000
Nonduplicated number of food codes used to code INTERMAP 24-he	839	1145	1438	Comparable data not availabled

^aIncludes 681 foodsFcounted twiceFwith nutrient data for both raw and cooked versions (see text).

^bIncludes 804 foodsFcounted twiceFwith nutrient data for both raw and cooked versions (see text).

^cIncludes foods counted twice based on absence or presence of added salt in their preparation (see text).

^dThe automated Nutrition Data System (NDS) ‘builds’ foods/recipes as the participant reports ingredients, preparation methods, etc; for example, a mixed dish such as macaroni and cheese is not listed in the NDS file, rather its particular ingredients and their amounts (type macaroni, cheese, milk, fat, etc) as reported individually by a participant are identified and their nutrient contents are summated to produce the data on nutrient information for this dish as eaten by that person reporting its intake. The same procedure is repeated separately for each person reporting this dish, and is used for building nutrient composition of other mixed dishes/recipes.

^eIncludes some recipes coded to two or more separate ingredients.

For Japan, there were 1782 foods in the primary database, including 617 foods added before INTERMAP field work began. Based on foods reported by participants, 1149 new foods were added as separate codes (88 foods, identified initially as possible ‘new foods’, were finally coded to already existent similar foods). Also, nutrient composition was incorporated for cooked versions of 804 foods where original tables gave data only for raw versions. Thus, the final total of food codes in the Japanese database is 2931 (Table 3).

For the UK, the original FOODBASE file counted 3750 food codes; 213 new foods were added based on participant 24-h recalls, to give a database of 3963 food codes (Table 3). Also, during field work, 9735 new rules were incorporated into the INTERMAP UK coding manual; these specified that many initially identified possible ‘new foods’ be assigned to already existent codes for similar foods.

For the USA, the 17 000 foods/beverages/ingredients in the updated NDS version 2.91 enabled ‘building’ of nutrient composition for all foods reported by participants (see Table 3 footnotes). For 766 foods, the NDS process identified them as similar to foods/recipes already in the database.

The extensive work on the databases resulted in their encompassing 76 nutrient values for all foods from the four countries (62 of these fully comparable, 14 generally and practically comparable across the four countries) (Table 4). Thus, extensive data are available on macro- and micronutrient composition of all foods and beverages reported by the 4680 INTERMAP participants. For the seven B vitamins (thiamin—B₁, riboflavin—B₂, niacin—B₃, pantothenic acid, pyridoxine/piridoxyl/pyridexamine—B₆, folate, cobalamin—B₁₂), data are available for UK and US foods, and are pending for Chinese and Japanese foods.

Table 4.

Nutrients comparable across four INTERMAP country-specific dietary databases on nutrient composition of foods

Energya (kcal)	SFA 20:0, arachidic acida (g)	Argininea (g)
Total fata (g)	SFA 22:0, behenic acida (g)	Histidinea (g)
Total available carbohydratea (g)	MFA 14:1, myristoleic acida (g)	Alaninea (g)
Total proteina (g)	MFA 16:1, palmitoleic acida (g)	Aspartic acida (g)
Animal proteina (g)	MFA 18:1, oleic acida (g)	Glutamic acida (g)
Vegetable proteina (g)	MFA 20:1, gadoleic acida (g)	Glycinea (g)
Cholesterol (CHOL)a (mg)	MFA 22:1, erucic acida (g)	Prolinea (g)
Saturated fatty acids (SFA)b (g)	PFA 18:2, linoleic acida (g)	Serinea (g)
Monounsaturated F. Ac. (MFA)b (g)	PFA 18:3, linolenic acidb (g)
Polyunsaturated F. Ac. (PFA)b (g)	PFA 20:4, arachidonic acidb (g)	Caffeinea (mg)
PFA/SFAb	PFA 20:5, eicosapentaenoic acid (EPA)a (g)
Keys dietary lipid scoreb	PFA 22:5, docosapentaenoic acid (DPA)b (g)	Total vitamin A activitya (IU)
1.35 (2 SFA-PFA)+1.5 CHOL1/2	PFA 22:6, docosahexaenoic acid (DHA)a (g)	Total vitamin A activity
Hegsted dietary lipid scoreb	Trans F. Ac. 16:1 (transhexadecenoic acid)a (g)	Retinol equivalentsb (μg)
2.105 SFA-1.016 PFAa+0.067 CHOL	Trans F. Ac. 18:1 (transoctadecenoic acid, elaidic acid)a (g)	Betacarotene Equivalentsa (μg)
Omega-3 F. Ac.b (g)		Retinola (μg))
Omega-6 F. Ac.b (g)	Trans F. Ac. 18:2 (transoctadecadienoic acid, linolelaidic acid)a (g)	Total vitamin E activityb
Trans Fatty Acidsa (g)		Alpha tocopherol Equivalents (mg)
Starcha (g)		Vitamin C, ascorbic acida (mg)
Estimated total sugarsa,c (g)	Tryptophana (g)
Alcohola (g)	Threoninea (g)	Calciuma (mg)
Total dietary Fibrea (g)	Isoleucinea (g)	Phosphorusa (mg)
SFA 6:0, caproic acida (g)	Leucinea (g)	Magnesiuma (mg)
SFA 8:0, caprylic acida (g)	Lysinea (g)	Irona (mg)
SFA 10:0, capric acida (g)	Methioninea (g)	Seleniuma (/ig)
SFA 12:0, lauric acida (g)	Cystinea (g)
SFA 14:0, myristic acida (g)	Phenylalininea (g)	Sodiuma (mg, mmol)
SFA 16:0, palmitic acidb (g)	Tyrosinea (g)	Potassiuma (mg, mmol)
SFA 18:0, stearic acida (g)	Valinea (g)	Na/Ka (mmol/mmol)

^aFully comparable across four countries.

^bGenerally comparable, with small differences, across four countries.

^cEstimated total sugars: computed as total available carbohydrate minus starch.

Indicators of quality of dietary data collection

For collection of the four 24-h dietary recalls/person, there were in China 17 interviewers (for three population samples, 839 people); in Japan, 36 interviewers (four samples, 1145 people); in the UK, 19 interviewers (two samples, 501 people), in the USA, 32 interviewers (eight samples, 2195 people) —that is, on average in China 5.7 interviewers/sample, one interviewer/49.4 persons; Japan, 9.0/sample, 1/31.8 persons; UK, 9.5/sample, 1/26.4 persons; USA, 4.0/sample, 1/68.6 persons.

Data are available for several variables potentially indicative of quality of dietary data collection: number of foods per recall; time to do recall; error rate in coding recalls; scores on 12-item review (by Site and by Country Nutritionist) of recall audiotapes; energy intake per day and per kg body weight; correlations of nutrient intake values measured by both 24-h dietary recall and 24-h urine collection (total protein, sodium (Na), potassium (K), Na/K) (Tables 5–7).

Table 5.

Descriptive statistics, INTERMAP indicators of quality of dietary data collection, by country

Variable	China			Japan			UK			USA
	Mean	s.d.	Range	Mean	s.d.	Range	Mean	s.d.	Range	Mean	s.d.	Range
Number of foods per recall	15.7	6.0	1–50	37.5	9.4	3–89	24.0	5.7	9–43	31.7	10.8	4–91
Time to do recall (min)	21.6	5.6	5–46	30.8	6.9	14–61	31.2	6.8	16–62	28.3	7.5	12–67
Batch coding error ratea (%)	2.3	0.2	1.0–3.9	1.4	1.4	0–6.0	1.4	1.4	0–5.2	Not applicableb
Scores on 12-item local review of recall audiotapec	45.3d	2.2	39–48	40.4e	2.5	34–45	42.5f	1.9	37–45	42.7e, 41.5g	3.3e, 5.2g	32–48e, 23–48g
Low scores (1 or 2) on 12-item local review of recall audiotapec (%)	0.3d	NA	0–3	11.2e	NA	0–42	3.0f	NA	0–22	8.8e, 2.0g	12.5e, 5.3g	0–75e, 0–42g
Total energy intake per recall per day (kcal)	2037	577	747–136	2038	449	902–4485	2168	631	980–4539	2244	699	684–5881
Energy intake/weight (kcal/kg)	35.0	9.8	11.6–74.9	33.7	7.2	14.6–73.6	28.3	8.5	10.5–67.3	28.0	8.7	7.0–79.9

^aBatches of 30 completely coded recalls (15 pairs) were sent by Site Nutritionists to Country Nutritionists, who recoded a randomly selected three from each batch; error rate is: (number of changes in food codes/total number of food codes) 100; these error rates are for finally approved batches, that is, they do not include rates for rejected batches with >6% error rates, returned to site for recoding (see text).

^bIn the USA, there was no coding of dietary data from hard copy, hence no recoding with estimation of error rates, given use of the automated NDS for online entry of dietary items directly into the computer during the dietary interview (see text, and Table 6 for related data).

^cTapes were scored 1, 2, 3, or 4 for each of 12 items (criteria) of quality assessment; thus, for each tape the lowest (worst) possible score is 12 and the highest (best) possible score is 48 (12 ×4) (see text); for USA, data as of 24 May 1998 with over 90% of field survey work completed.

^dThe Site Nutritionists at the three Chinese local centres engaged in audiotape review; the Country Nutritionist was unable because of brevity (3 months) and overlap of field surveys at the three rural sites (see text); data are from local review.

^eData from national (Country Nutritionist) review of audiotapes.

^fData from joint local (Site Nutritionists) and national (Country Nutritionist) review of audiotapes.

^gData from local (Site Nutritionists) review of audiotapes.

NA is not available; only overall calculations were made, that is, all low scores on all tapes divided by (all tapes evaluated ×12).

Table 7.

Partial product moment (Pearson) correlation coefficients between INTERMAP dietary and urinary measurements of protein, sodium, potassium, and Na/K adjusted for sample and gender

Variables	Chinaa	Japana	UKa	USAa	All menb	All womenb	Alla
Number of participants	839	1145	501	2195	2359	2321	4680
Dietary protein and urine urea	0.395	0.524	0.480	0.521	0.501	0.511	0.508
Dietary protein and EPIc	0.404	0.520	0.469	0.514	0.496	0.506	0.503
Dietary Na and urine Na	0.332	0.450	0.355	0.462	0.417	0.409	0.415
Dietary K and urine K	0.303	0.565	0.507	0.580	0.519	0.580	0.553
Dietary Na/K and urine Na/K	0.281	0.530	0.477	0.512	0.385	0.449	0.419

^aAdjusted for sample and gender.

^bAdjusted for sample only.

^cEstimated protein intake (g/day), calculated as: (urea nitrogen+0.031 (body weight in kg)) × 6.25.

Number of foods/recall:

Across the 4680 participants, number of foods per recall varied from 1 to 91 (Table 5, row 1). Average number, for participants across the four countries, was lowest for China (15.7 ± 6.0) and highest for Japan (37.5 ± 9.4).

Time to do recall:

Time to do a 24-h recall varied from 5 to 67 min/person (Table 5, row 2). Average time was lowest for China (21.6 ± 5.6 min) and highest for UK (31.2 ± 6.8 min). It was similar for men and women, and shorter on average at repeat visits compared to the first, as expected (data not shown).

Coding error rate:

The Country Nutritionists in China, Japan, and UK identified a random 10% sample of three recalls (from every batch of 30 completed coded recalls forwarded from a field site) and recoded them. The Country Nutritionist defined a coding error as any food (line) in a recall that he/she identified as requiring a code different from that used by the local dietary interviewer. Average error rates per batchF(lines with errors/total lines) × 100—were <3.0% across countries (Table 5, row 3).

Scores on 12-item review of recall audiotapes:

For quality assessment of interviews, recalls were routinely tape recorded. Random samples of these tapes were reviewed by Site and Country Nutritionists and scored 1–4 (unsatisfactory to excellent) on each of 12 specified criteria, that is, lowest possible score/tape 12, highest possible score 48. Average scores were high across countries (Table 5, row 4).

Per cent of low scores was also computed, as (number of 1, 2 scores/12) × 100. On average, low scores—based on Country Nutritionist evaluation—were most frequent in Japan (11.2%), least frequent in UK (3.0%) (Table 5, row 5). For China, only assessments by Site Nutritionists were made, with 0.3% low scores; the comparable statistic for the USA was 2.0%.

Total energy intake—kcal/day and kcal/kg body weight:

Average energy intake (kcal/day) was lower for East Asians than Westerners, highest for US participants (Table 5, row 6). However, energy intake/kg body weight (indicative of energy expenditure) was higher for East Asians than Westerners (Table 5, row 7). Interindividual variation in daily energy intake, as measured by the standard deviation (s.d.) and the coefficient of variation (s.d./mean), was greatest for US participants, least for Japanese.

US re-entry of 24-h recalls into computer, based on tape recordings of recalls:

From onset of US field work through 24 May 1998, with over 90% of the survey effort completed, 8762 24-h dietary recalls were done; 601 (6.9%) were re-entered into the NDS system by Site Nutritionists; 473 (5.4%) were re-entered by the Country Nutritionist (Table 6). In both these re-entry efforts, for almost half the tapes there was at least one discrepant re-entry compared to the original, producing an edit (modification, correction). Discrepancies per 10 re-entered tapes averaged 7.9 for local re-entry, 11.5 for national re-entry.

Table 6.

Local (Site Nutritionist) and national (Country Nutritionist) re-entry of INTERMAP dietary recalls into computer in USA (NDS), by sample^a

Variable	Baltimore	Chicago	Corpus Christib	Corpus Christic	Honolulu	Jackson	Minneapolis	Pittsburgh	All eight samples
Number of recalls completed	1142	1268	996	998	1130	1092	1074	1062	8762
Local (Site Nutritionist) re-entry
Tapes re-entered (no. (%))	97 (8.5)	102 (8.0)	51 (5.1)	59 (5.9)	601 (6.9)	84 (7.7)	51 (4.7)	67 (6.3)
Tapes with any (one or more) discrepanciesd (no. (%))	56 (57.7)	42 (41.2)	6 (11.8)	7 (11.9)	290 (48.3)	59 (70.2)	28 (54.9)	24 (35.8)
Number of discrepanciesd in all re-entered tapes (discrepanciesd per 10 re-entered tapes)	98 (10.1)	57 (5.6)	6 (1.2)	8 (1.4)	477 (7.9)	118 (14.0)	47 (9.2)	27 (4.0)
National (Country Nutritionist) Re-entry
Tapes re-entered (no. (%))	62 (5.4)	55 (4.3)	18 (1.8)	32 (3.2)	84 (7.4)	73 (6.7)	76 (7.1)	73 (6.9)	473 (5.4)
Tapes with any (one or more) discrepanciesd (no. (%))	33 (53.2)	23 (41–8)	3 (16.7)	15 (46.9)	41 (48.8)	57 (78.1)	14 (18.4)	31 (42.5)	217 (45.9)
Number of discrepanciesd in all re-entered tapes (discrepanciesd per 10 re-entered tapes)	88 (14.2)	51 (9.3)	6 (3.3)	30 (9.4)	99 (11.8)	153 (21.0)	38 (5.0)	79 (10.8)	544 (11.5)

^aData as of 24 May 1998, with over 90% of US field survey work completed; from a working report prepared by the US INTERMAP national coordinating centre in Chicago.

^bHispanic-American (Mexican-American) sample.

^cNon-Hispanic white sample.

^dDiscrepancy defined as an entry producing an edit (modification, correction) of diet recall data originally entered into computer by interviewer at local site.

Correlation between dietary and urinary measurement of total protein, sodium, potassium, Na/K:

Results were almost identical for product moment and rank order partial correlations, controlled for sample and (as appropriate) gender; the former are given (Table 7). For total protein intake as assessed by 24-h recall and 24-h urinary urea, the correlation coefficient (r) for all 4680 participants was 0.51; for all men, 0.50; for all women, 0.51 (Table 7, row 2). Range across the four countries was from 0.40 (China) to 0.52 (Japan and USA). Results were almost identical with use of the alternative method for estimation of protein intake (based on urinary urea and body weight—see the section Methods) (Table 7, row 3).

For dietary Na and urinary Na, overall r was 0.42 (Table 7, row 4). This was the lowest overall r of any of the four single variables. The Na r for men was 0.42; for women, 0.41; across the four countries values for this r were 0.33 for China, 0.36 for UK, 0.45 for Japan, and 0.46 for USA.

For dietary K and urinary K, overall r was 0.55; 0.52 for men; 0.58 for women; range across the four countries, 0.30 (China) to 0.58 (USA) (Table 7, row 5).

For dietary and urinary Na/K, overall r was 0.42; for men, 0.38; for women, 0.45; range across the four countries, 0.28 (China) to 0.53 (Japan) (Table 7, row 6).

Discussion

All prior hypotheses and exploratory analyses in the INTERMAP Protocol deal with relationships of dietary variables of individual participants to their BP. The underlying concept is that single nutrients have ‘small’—but additive—effects on BP of individuals, at least in cross-sectional observational studies and short-term trials.^{6–9,39,47–49} Hence the goal of dietary assessment had to be to approximate as closely as possible true intake of each individual participant for each of 4 days, and then use 4-day mean values for each person in analyses, thereby lessening the role of intraindividual variation in producing coefficients for nutrient–BP relations less than true values (the regression–dilution problem).^1–8

Multiple procedures, stipulated in the Protocol and Manuals of Operations, were implemented to accomplish the goal of collecting dietary data of high-order validity for each participant. All these sought to minimize known sources of error; for this purpose they drew upon experience of preceding studies and the literature on cognitive behaviour.^10,11,50 Their implementation was ongoing at the local, national, and international level throughout the study. They involved training and certification of staff, monitoring dietary data collection and processing and their quality, and institution of timely specific steps to rectify detected inadequacies. They also entailed systematic procedures to control and enhance quality in computerization of dietary data, their conversion to nutrient data, and edits of these data.

For monitoring dietary data collection and processing, INTERMAP used measurement data on several variables. While some of these were generally of limited value in detecting shortcomings of a single 24-h recall, they were meaningful in the aggregate. Thus, for any one recall, it could be valid that only a few foods were eaten, it took a short time to do the recall, and energy intake was low, or—alternatively—these could be signs of inadequate data collection, with the distinction between these possibilities not ascertainable. However, in the aggregate on an ongoing basis, across interviewers, samples, countries, sizable differences in findings for such variables can be signs of problems. At least across the four countries, there were no such differences interpretable as evidence of lower quality in dietary data collection in one country compared to others. Similarly, data on batch coding error rates and scores on the 12-item reviews of audiotapes of dietary recalls were consonant with the judgement that data collection and processing were generally of high quality.

Considered together, these INTERMAP dietary quality control procedures were extensive and time consuming. Some were new to studies of this kind, especially review of randomly selected tape recordings to assess quality of 24-h recall interviews, and in the USA use of randomly selected tape recorded interviews to re-enter 24-h recall information into the computer. Hence it seems valuable here to comment on feasibility and usefulness of the key procedures as applied in the field situations: An advantage of prompt and ongoing local evaluation of tape recordings by Site Nutritionists was that it gave rapid feedback to the dietary interviewers. It thereby served as a practical tool for maintenance and enhancement of quality of data collection. Access to tapes was also useful in review of recall hard copies and coding, serving as objective basis for identifying, adjudicating, and correcting errors. However, feasibility issues and disadvantages of the tape recording Protocol became apparent early in field work. Each tape took about 45 min to review and score. With use at a centre of many interviewers, time spent by Site Nutritionists on review of tapes (one/interviewer/week) became overwhelming. Also, there was a bias problem, that is, a disincentive to give low scores because these triggered review of more recorded recalls—note the considerably lower per cent of poor scores with Site (2.0%) compared to Country (8.8%) Nutritionist review of US tapes. Further, intensity of field work influenced feasibility of full implementation of this quality control procedure. Thus, it was a very different situation with 10 participants/day surveyed at Chinese sites compared to 2–3/day for most US samples. Tape costs and storage were additional difficulties. With sample size 260 and four recalls/person, over 1000 tapes/site were required to record and archive every interview, a need that could not be met. Instead, tapes were reused after completion of quality control procedures.

In our judgement, taping recalls serves well as a teaching tool and source for what was actually said in the interview. Thus, tapes were used in Japan to check written records with participants’ descriptions. Likewise, they were relied on for adjudicating discrepancies in coding. Prompt review of a sample of tapes is useful to give rapid feedback to interviewers. Without early feedback, there is no opportunity to correct interviewing errors before many interviews have taken place. However, this must be balanced with other considerations described above. We did not attempt to test reliability of tape scoring by the same and different nutritionists, or to assess objectivity in scoring standards. To do so may require a scripted interview. This would interfere with spontaneity of the interview, deemed important to establish rapport with the participant; it would also limit flexibility in probing.

For dietary recalls coded manually, duplicate coding of at least a random sample is considered essential.^5,10,11 INTERMAP used such repeat coding, with the stipulation that any batch of 30 recalls had to be recoded locally with identification by the Country Nutritionist of a line error rate greater than 6% in any one or more of three recalls (10%) randomly selected for recoding. Our judgement is that this approach served well, including as an incentive to dietary interviewers and Site Nutritionists to do high-quality coding. For the three countries doing manual coding, final error rates (exclusive of batches returned for recoding) averaged 1.4–2.3%. Further, 100% range checks on serving sizes and on key nutrients were useful in identifying and assessing possible errors not detected by duplicate coding of 10% of recalls.

In the USA, there were multiple checks aiming to achieve high-quality dietary data with use of the NDS automated online method for the 24-h recall. These encompassed programmed checks transmitted by the computer to the interviewer as the recall proceeded and reported foods/beverages were entered. They also entailed review of food item amounts, nutrients per food, and key nutrient totals for each recall. Re-entry at site and country level of random samples of taped recalls, analogous to recoding of recalls in the three other countries, served to identify problems in a timely way. Structured evaluation of the recall interview by listening to and grading random samples of tapes also was a useful procedure for quality assessment, maintenance, and enhancement. However, these multiple checks added considerable work to the dietary methodology. Listening to tapes was particularly time consuming. This could have been performed less frequently given the several other quality control checks. However, the INTERMAP experience in each country confirms the value of taping all recalls, and of ongoing monitoring and use of multiple checks on data quality, including use of a system as sophisticated as NDS.

To complete the process of acquiring high-quality nutrient data, up-to-date state-of-the-art computerized information—comparable across countries— was essential on nutrient composition of all reported foods, numbering in the thousands, including hundreds of ‘new’ foods not in original country dietary databases. NCC was the pivotal resource for accomplishment of the several aspects of this task—evaluation and enhancement of quality of available data on nutrient composition of foods; acquisition of missing data (including data on new foods); expansion of data to encompass additional nutrients and maximize comparability across the four countries. These multiple efforts required painstaking work for many months by NCC jointly with Country Nutritionists and INTERMAP senior investigators. Completion of these essential tasks was an undertaking requiring substantially more time and effort than originally anticipated. The four complete databases are resources available for research transcending INTERMAP.⁵¹

Finally, it is essential explicitly to underscore a key fact about all dietary/nutrient data acquired by self-report of free-living people: There is no ‘gold standard’ for evaluation of validity, that is, objective truth. The INTERMAP Protocol stipulated use of four urinary variables (urea, sodium, potassium, Na/K) from two timed 24-h urine collections/person as relatively objective measures of intake of protein, Na, K, Na/K, to evaluate quality of data on these variables accrued with four 24-h dietary recalls/person. In so doing, INTERMAP recognized that these were not ‘gold standards’ for validity, since there was no procedure to estimate with certainty accuracy—for example, completeness—of each 24-h urine collection. Further, there is the problem—also not quantifiable—of dyssynchrony between urine and dietary measurements due to possible incomplete absorption and excretion of ingested nutrients during the specific 24-h period. As shown in the Results section, correlations between urinary and dietary values for these four variables were for all 4680 participants as high as or higher than those reported from previous population studies^38,52–55—0.51 for protein, 0.42 for sodium, 0.55 for potassium, and 0.42 for Na/K, with coefficients similar for men and women. While these r values can be assessed as indicative of relative validity of the dietary data on these variables, there are reasons for concern. Thus, the fact that the r for Na was consistently lower than the others for single nutrients (an expected finding) reflects the special difficulties in accurate assessment by dietary survey methods of individual intake of the food additive salt (NaCl), added in kitchen, at table, and in commercial food processing (source of about two-thirds or three-quarters of ingested salt).^9,56–59 For Japanese, UK, and US participants, it is reasonable to conclude that INTERMAP recalls captured this portion. In China, few commercial products were consumed by the farmers, and total salt/soy sauce intakes were estimated separately, as was done in Japan. This circumstance may have played a role in the lower Na r values for China than the other countries, but the UK r was only slightly higher than the Chinese. In general, the variation in all r values across countries can be interpreted as indicative of differing degrees of data quality (if the assumption is accepted that urine values were of similar high-level relative validity across countries).

Insofar as urine values were not absolute ‘gold standards’, correlation coefficients less than 1.00 can be attributed at least in part to this limitation; for the rest (not quantifiable), they can be attributed to possible dyssynchrony in time between ingestion and excretion (even with close time linkage of urinary and dietary data collections), and of course to limitations in validity of the dietary data, despite the multiple INTERMAP efforts to minimize them. No doubt there were such, not only for Na (as already noted), but also for protein, potassium, and other nutrients. While multiple measurements permit statistical quantitation of reliability and correction for its limitations in regression analyses of nutrient–BP relations,^2–8 there are no ways to measure and adjust for limitations in validity of the data. Insofar as these supervene, they introduce ‘noise’ into analyses of nutrient–BP relations, tending generally to result in coefficients smaller than true ones. This must be kept in mind in all work with such data.