Analysing dietary data to capture how individuals typically consume foods is dependent on the coding variables used. Individual foods consumed simultaneously, like coffee with milk, are given codes to identify these combinations. Our literature review revealed a lack of discussion about using combination codes in analysis. The present study identified foods consumed at mealtimes and by race when combination codes were or were not utilized.

Duplicate analysis methods were performed on separate data sets. The original data set consisted of all foods reported; each food was coded as if it was consumed individually. The revised data set was derived from the original data set by first isolating coded foods consumed as individual items from those foods consumed simultaneously and assigning a code to designate a combination. Foods assigned a combination code, like pancakes with syrup, were aggregated and associated with a food group, defined by the major food component (i.e. pancakes), and then appended to the isolated coded foods.

Healthy Aging in Neighborhoods of Diversity across the Life Span study.

African-American and White adults with two dietary recalls (n 2177).

Differences existed in lists of foods most frequently consumed by mealtime and race when comparing results based on original and revised data sets. African Americans reported consumption of sausage/luncheon meat and poultry, while ready-to-eat cereals and cakes/doughnuts/pastries were reported by Whites on recalls.

Use of combination codes provided more accurate representation of how foods were consumed by populations. This information is beneficial when creating interventions and exploring diet–health relationships.

To validate a diet history questionnaire (DHQ) using a weighed food record (WFR) as the standard method in the estimation of food consumption and nutrient intake in a group of adults.

WFR: all foods consumed by subjects during 7 consecutive days were weighed and recorded by nutrition students. Two DHQ interviews were carried out on days 1 (first diet history questionnaire, DHQ1) and 28 (second diet history questionnaire, DHQ2).

Costa Rica.

Sixty adults: 30 men and 30 women; 30 living in urban and 30 in rural areas.

In comparison to the WFR, the DHQ1 gave statistically significant higher estimates of the mean intake of 19 nutrients for men and of three nutrients for women. The uncorrected correlation coefficients for nutrient intake according to both methods ranged from 0.40 to 0.83 for males and from 0.22 to 0.62 for females. Percentage of subjects classified in the same quartiles of nutrient intake according to each method ranged from 33.3% to 63.3% for males and from 23.3% to 53.3% for females. Misclassification in extreme quartiles ranged from 0% to 13.3% for both sexes. The mean food group consumption, according to the DHQ1, when compared with the WFR, gave statistically significant differences for three of the 18 food groups for men and for two groups in the case of women. The two applications of the DHQ gave similar results.

Validation of a DHQ using a WFR as the standard method gave results that compare favourably with those reported by other authors. This study found important differences in the response of men and women to the DHQ: among men, the estimates of mean nutrient intake from DHQ1 were significantly greater than those of the WFR, while in the case of women, the mean nutrient intake estimates from both methods were not significantly different. There was a higher degree of correlation between the DHQ1 and the WFR mean nutrient intakes among men than among women. The DHQ showed good reproducibility.

To describe the main finding produced so far by the EPIC project, which aims to improve the scientific knowledge of nutritional factors involved in cancer.

Prospective cohort study in 23 centres from 10 European countries. Dietary assessment method varied by countries: in some cases a diet history by personal interview was carried out, while in most countries a self-administered semi-quantitative food-frequency questionnaire was used; in one case the questionnaire was combined with a food record. Information on a wide range of lifestyle factors and anthropometric measurements were also collected for most participants.

The cohort includes 521 468 men and women mostly aged between 39 and 69 years, whereas blood samples are available for 387 889 subjects.

To date 24 185 cancer cases have been identified in the follow-up of the cohort. The publications produced by each centre can be consulted on the EPIC website (http://www.iarc.fr/epic/). Among initial findings concerning the associations between cancer and dietary factors, one of the most important results is a protective effect of high fibre intake and fish consumption against colorectal cancer, while high red and processed meat intake increase the risk. Regarding lung cancer the first analyses found a protective effect of fruit intake but no association with vegetable consumption. No association was observed between vegetables and fruit intake and the risk of prostate cancer or breast cancer. Finally, data from Cambridge point to an interesting result regarding breast cancer: no association was observed with saturated fat intake measured by food-frequency questionnaire, but according to the food diary a daily intake of 35 g doubles the risk of breast cancer compared to women with daily intake of 10 g or less.

To develop and test a quantitative, interviewer-administered food frequency questionnaire (FFQ) to ascertain nutrient intakes of individuals in northern India.

A 92-item FFQ was developed based on food use and market surveys of the study area. A validation study was conducted consisting of 24-h diet recalls (24HR) administered on 6 randomly selected days over 1 year. Two FFQs were administered, one each at the beginning and end of the 1-year period. FFQ and 24HR-derived nutrient scores were compared using correlation and regression analyses and by computing differences between nutrient intakes estimated by the two methods.

Rural villages in Bhavnagar District, Gujarat, North India.

60 individuals who agreed to provide all necessary data.

Pearson (parametric) correlation coefficients averaged 0.69 in comparing nutrient scores derived from the 24HR with those from the first FFQ and 0.72 in comparing the second FFQ (P < 0.0001). Spearman correlation coefficients were virtually identical to the Pearson correlations, averaging 0.68 and 0.72, respectively. In regression analyses, most coefficients were close to 1.0 (perfect linear association). Nutrient scores were significantly and consistently higher on both FFQs relative to the 24HR.

This FFQ produces results broadly comparable, and superior in some respects, to those commonly used in the West. Higher than average measures of association indicate its suitability for comparing exposures within this study population in reference to health-related endpoints.

To validate an estimated food record (EFR), using a weighed food record (WFR) as the reference method, for the determination of food consumption and nutrient intakes in a group of Costa Rican adults.

For the WFR, all foods consumed by subjects during seven consecutive days were weighed and recorded by nutrition students. Two EFRs (a 4-day and a 3-day record) were recorded by subjects with the use of home measures and photographs to estimate amounts.

Costa Rica.

Sixty adults: 30 men and 30 women; 30 living in urban and 30 in rural areas.

The EFR gave statistically significant lower average intake estimates for energy and 12 of the 22 nutrients examined. The correlation coefficients ranged from 0.68 (polyunsaturated fats) to 0.87 (calcium). The percentage of subjects classified into the same quartile ranged from 45.0% (polyunsaturated fats) to 68.3% (vitamin B12). For all nutrients except vitamin C, 0 or 1.7% were misclassified into extreme quartiles. For food group consumption, the EFR gave statistically significant lower estimates for six of the 17 groups and correlation coefficients ranged from 0.22 (fish) to 0.93 (beverages). Greater differences in estimates of mean energy and nutrient intakes were detected among subjects from rural areas, caused in part by a tendency to underestimate the amounts of rice and beans consumed.

Validation of the EFR using a WFR as the reference method gave results that compare favourably with those reported by other authors, and support the use of the EFR for dietary surveys among Costa Rican adults.

The purpose of this study was to validate a quantitative food-frequency questionnaire (QFFQ) created for assessing the usual intake of foods and nutrients in the prevailing season in Western Mali.

Intake of foods and nutrients over the week preceding the interview was measured with a 69-item QFFQ. Intakes were compared with intakes as measured with 2-day combined weighed and recalled diet records.

A rural village in Western Mali, West Africa.

Twenty-seven men and 48 women (15–59 years of age) representing 18 households.

Spearman rank correlations between intake of food groups from the QFFQ and the diet record ranged from 0.09 (meat/fish) to 0.58 (tea/coffee). Median coefficient was 0.37. Median Spearman correlation coefficient for nutrient intake was 0.40. Men had higher median correlation coefficients than did women. The proportion of subjects being classified into the same quartile of food intake was on median 33%, while a median of 7% was misclassified into extreme quartiles. Correct classification into the same quartile for intake of nutrients was on median 34% while a median of 4% was grossly misclassified. Intakes of most food groups and nutrients as measured by the QFFQ were higher than those measured by the diet records. However, while men had higher estimated intakes for foods eaten in-between meals, women in general had higher intake of foods eaten in the main meals.

This QFFQ can be used for comparing the intake of foods and nutrients between groups within this study population. It therefore represents a useful tool in the surveillance of food intake in the population, both in identifying vulnerable groups and for tracking food intake over time. The differences between men and women in overestimating food intake need to be taken into account when using the method.

To evaluate measurement error structure in dietary assessment instruments and to investigate its implications for nutritional studies, using urinary nitrogen excretion as a reference biomarker for protein intake.

The dietary assessment methods included different food-frequency questionnaires (FFQs) and such conventional dietary-report reference instruments as a series of 24-hour recalls, 4-day weighed food records or 7-day diaries.

Six original pilot validation studies within the European Prospective Investigation of Cancer (EPIC), and two validation studies conducted by the British Medical Research Council (MRC) within the Norfolk cohort that later joined as a collaborative component cohort of EPIC.

A sample of approximately 100 to 200 women and men, aged 35–74 years, from each of eight validation studies.

In assessing protein intake, all conventional dietary-report reference methods violated the critical requirements for a valid reference instrument for evaluating, and adjusting for, dietary measurement error in an FFQ. They displayed systematic bias that depended partly on true intake and partly was person-specific, correlated with person-specific bias in the FFQ. Using the dietary-report methods as reference instruments produced substantial overestimation (up to 230%) of the FFQ correlation with true usual intake and serious underestimation (up to 240%) of the degree of attenuation of FFQ-based log relative risks.

The impact of measurement error in dietary assessment instruments on the design, analysis and interpretation of nutritional studies may be much greater than has been previously estimated, at least regarding protein intake.

In the vast majority of surveys and research in humans, dietary data are obtained from self-reports: recalls; records; or historical methods, usually food frequency questionnaires (FFQ). This study provides a rare opportunity to compare data derived from all three methods.

A crossover study of dietary fat in which data were collected using an average of 11.4 food records and 11.7 24-h diet recalls. Using simple subtraction and correlation, energy and nutrient intakes derived from the three methods were compared to each other and with those derived from a single FFQ. Analysis of variance was used to evaluate sources of variability in nutrient intakes estimated from the individual days of records and recalls.

An independent, free-standing medical research institute.

13 men who were compliant with study procedures.

FFQ-derived estimates of energy and nutrient intake were highest (e.g. 1967 kcal versus 1858 kcal and 1936 kcal for the records and recalls, respectively). Mean differences in energy and nutrient intakes and their variances were lowest and correlation coefficients highest in comparing the records and recalls (e.g. for fat the mean difference was 5.0 g, and r=0.85). Analysis of variance of individual days of record- and recall-derived datd (n=300) revealed that there was no effect due to either method (record or recall) or the sequence of administration.

Results of this study indicate that the FFQ overestimated dietary intake. Energy and nutrient results obtained from the records and recalls were interchangeable. However, based on smaller SDs around the means, it appears that the recalls may perform slightly better in estimating dietary intake in groups such as these well-educated, highly compliant men.

To develop and test a food frequency questionnaire (FFQ) for use in rural areas of Kerala, India.

Based on food use and market surveys of the study area, a quantitative 81-item interviewer-administered FFQ was developed. A validation study was conducted consisting of 24-h diet recalls (24HR) administered on 8 days randomly selected over an entire year and two administrations of the FFQ, one at the beginning of the l-year period and the other at the end. FFQ and 24HR-derived nutrient scores were compared using correlation and regression analyses and by examining differences in the nutrient scores.

Rural villages in Ernakulum district, Kerala, South India.

In each of 30 households, the male head of household and female food preparer were enrolled.

Pearson (parametric) correlation coefficients (rp) averaged about 0.50 in comparing nutrient scores derived from the 24HR with those from the first FFQ and about 0.55 in comparing the second FFQ. On average, Spearman correlation coefficients (rs) were slightly lower than the rp in comparing the scores derived from the first FFQ, but virtually identical for the second FFQ. Regression analyses indicated better agreement in the comparison of the 24HR-derived scores with the first FFQ than the second FFQ. Difference scores, however, tended to be larger in comparing the first FFQ scores with the 24HR.

This FFQ produces results broadly comparable to those used in Europe and North America, indicating its suitability for comparing exposures within a study population in reference to health-related endpoints.