Group Level Validation of Protein Intakes Estimated by 24-Hour Diet Recall and Dietary Questionnaires against 24-Hour Urinary Nitrogen in the European Prospective Investigation into Cancer and Nutrition (EPIC) Calibration Study¹

A new generation of validation and calibration designs using several reference dietary and/or biological measurements has been proposed to address the problem of measurement errors in dietary assessment methods used to investigate the relationship between dietary exposure and different diseases (1, 2, 3). Biological markers of absolute quantitative dietary intakes such as urinary N³ (4) or doubly labeled water for estimating total energy expenditure in free living people (5) are suggested as the two most accurate independent reference measurements for use in validation (or calibration) studies. In the absence of changes in N or energy balance, these markers provide a reliable estimate of individual or population N (protein) intakes and total energy expenditure (energy intakes). Of particular importance is the fact that they fulfil the fundamental statistical requirements of a reference method in validation or calibration studies, i.e., their measurement errors are independent of both the true unknown actual dietary intakes and of errors in the dietary measurements to be validated (6). Single measures have been used to verify estimated population intakes of N by various dietary methods (7, 8, 9, 10, 11), whereas repeat measures are needed to verify individual estimates (12, 13, 14).

Although studies using biological markers as a reference to validate individual measurements on restricted numbers of subjects are increasingly reported in the literature, nothing has been published on the validity of dietary measurements used as reference calibration in large nutritional surveys. In such studies, characterized by study populations from different geographical areas or ethnic groups, the calibration of dietary measurements used to assess diet in epidemiological studies has been proposed for two main purposes: (a) to correct for systematic over- or underestimation of the true intakes attributable to the different DQs used at the population level; and (b) to possibly correct for attenuation bias in the relative risk because of random errors in dietary measurements at the individual level (15, 16, 17). In practice, calibration involves applying, in addition to the DQs used to estimate individual intakes, a second highly standardized dietary method in a representative subsample as a common reference measurement across study populations. For calibration at the population level (between-cohort calibration), the selected reference dietary method (e.g., 24-HDRs, dietary records, or biological markers) must be unbiased, i.e., provide a correct estimate of the true mean of group populations. However, in the absence of any dietary method totally free of errors, the calibration of DQs will still improve the comparability of measurements across study populations if the systematic errors of the reference calibration measurements are relatively modest and constant across study populations, and the method accurately characterizes average consumption in all of the centers.

This paper presents the results of a study designed to validate the mean of single 24-HDR measurements, used as reference method to calibrate population mean estimates, against mean of single 24-h urinary N in EPIC. EPIC is a network of prospective cohort studies with approximately half a million subjects recruited from 23 European centers in 10 Western European countries (France, Italy, Spain, United Kingdom, Germany, the Netherlands, Greece, Sweden, Denmark, and Norway; Ref. 18). Diet has been measured for all of the participants using validated DQs (14, 19, 20, 21), and mean N (and total energy) intakes obtained from a subsample were also compared both to the mean 24-h dietary recall and urinary measurements.

The EPIC cohort includes 519,978 subjects from 10 countries who completed baseline dietary and other lifestyle questionnaires; ∼37,000 of them also underwent a single 24-HDR interview that was used as the reference calibration method (18). A convenient subsample from the calibration in 9 EPIC centers (or country): Paris (France); Varese, Turin, Florence, and Ragusa (Italy); Greece where the sample was recruited from different geographical regions; Cambridge and Oxford (United Kingdom); and Heidelberg and part of the participants in Postdam (Germany) were recontacted between 1995 and 1999 and asked to collect a 24-h urine specimen. Those who agreed to participate signed a consent form. In Bilthoven (the Netherlands) a strict random sample from the calibration sample was drawn from the subjects approached during 1997. In Potsdam (Germany) and Naples (Italy) the subjects involved in this study were all from the EPIC cohorts but not from the calibration sample. In Potsdam (Germany) most of the participants came from a previous validation study that was conducted among cohort members where repeated 24-HDRs and 24-h urines were collected using the same standard procedure as in the calibration study (22). In Naples, the subjects were selected from the whole cohort, and asked to collect a 24-h urine sample and a 24-HDR interview took place when they brought the urine collection back to the center. Thus, a total of 12 EPIC centers were finally involved in the statistical analysis. Initially the sample included 1386 volunteers, both men and women, but a number of them were excluded for reasons that will be explained later. Ultimately, the data from 1103 subjects were used for the statistical analysis.

Information on usual individual dietary intakes was assessed at baseline for each subject entering the EPIC cohorts using exhaustive quantitative DQs (France, the Netherlands, Germany, Greece, and Italy, except Naples) or a semiquantitative food frequency questionnaire (United Kingdom and Naples) developed and validated in each participating country (14, 20). These questionnaires contain 158–266 food and recipe items, and the portion sizes were estimated either by photographs or standard units, except in the United Kingdom and Naples where standard portions were used.

In addition, a single 24-HDR interview was used as the EPIC reference calibration measurement and collected from a subsample of each cohort (23). In contrast to the baseline DQs, the 24-HDR interviews were highly standardized across countries, using a computerized program (EPIC-SOFT) with the same structure and interview procedure across countries. At the ecological level, the aim of the calibration was to estimate mean population intakes, and only one 24-HDR measurement was collected from a large stratified representative sample of each of the EPIC cohorts (5–12%). Information on all of the foods and beverages consumed during the previous day was collected, entered, and coded automatically according to common rules. Food portion sizes were estimated using a common picture book containing sets of photographs of 140 foods and recipes (24), and other available methods such as standard units and household measurements. Trained dieticians conducted all of the interviews face-to-face. More details on the concept of standardization and the structure of EPIC-SOFT are described in detail elsewhere (25, 26).

In the absence of a standardized European nutrient database (27, 28), the average N and energy intakes were calculated using country-specific food composition tables for both dietary methods. A review on the comparability of food composition tables available in countries participating in EPIC (29) suggested that both protein and energy values reported in the food composition tables used in this analysis are reasonably comparable for between-country comparisons.

Before the urine collection, the subjects were instructed on how to collect a complete 24-h urine sample after a standardized protocol (13). Each subject was provided with two 2-liter bottles for urine collection, each containing 5 g of boric acid as a preservative. To evaluate the completeness of the 24-h urine collections, the subjects were given three 80-mg tablets of PABA to take at mealtimes during the course of the urine collection (4). PABA is absorbed by simple diffusion throughout the small intestines, and its metabolites are actively secreted by the renal tubule. Therefore, it is completely and quantitatively excreted in urine over the 24-h period, and the completeness of collection is estimated by measuring the PABA metabolite recovery in urine. After collection, the 24-h urine was stored at −20°C at the local center, then samples were packed in dry ice and shipped within 24 h to a central laboratory in Cambridge where all of the chemical analyses were performed according to standard procedures (13). In urine, total N was determined by the Kjeldahl technique (Tecator 1002; Perstorp Analytical, Bristol, Avon, United Kingdom), and PABA was measured by colorimetry (30). Twenty-four-h urine collections containing between 85% and 110% of the PABA marker were considered complete. Urine specimens containing <70% PABA recovery (when <3 tablets may have been taken) and >110% (when subjects may have taken drugs that interfere with the colorimetric analysis) were excluded from the analysis. Additional urine samples (70–84% PABA recovery) were used after correction for urinary electrolytes up to the expected values at 93% of PABA recovery (31). To allow for extrarenal losses in estimating 24-h N excretion, each individual 24-h urine N was divided by 0.81 (4).

To estimate the impact on the mean of extreme reported energy intakes that were physiologically implausible, high under- or over-reporters were defined according to Goldberg et al. (32). Different exclusion cutoff points were used for 24-HDRs (one single observation per individual) and for the DQ (long-term usual diet) to take into account the different day-to-day variations in dietary intakes and physical activity assumed for these two dietary methods, which are on different time scales (32).

We first compared the mean N (and energy) intakes estimated from the DQs, 24-HDRs, and urine. To approximate normality of the distributions, the individual N (and energy) measurements obtained from the DQs (DQ and 24-HDRs) and the urine were log-transformed, and the geometric means and their confidence intervals were calculated. The statistical analysis was stratified by gender and center to take into account the different study designs and populations involved, and any center-related residual confounding. Although the total number of subjects was low in Ragusa and Naples, it was decided not to merge these two Southern centers, because different baseline (DQ) dietary measurements were used. To compare urinary N with the N (and energy) estimates from the two dietary methods, independently of each other, noncalibrated DQ means were used in this analysis. The degrees of under- or over-reporting of N in dietary measurements were expressed as the ratio of the mean of 24-HDRs or DQ intakes divided by mean urinary N. For total energy, this ratio was computed as the mean total energy intakes estimated by each dietary method divided by the mean of individually calculated BMRs multiplied by a constant physical activity level of 1.55, used as a conservative lower value for sedentary populations (33).

A pairwise t test was used to determine statistically the differences between, respectively, the mean N DQs and 24-HDRs, and the reference mean urinary N estimates. The test was performed on the means of log-transformed individual values, and Ps < 0.05 were considered statistically significant.

We then performed an ecological analysis between mean center N and energy intakes estimated from DQs and 24-HDRs, and urinary N. Because one of the aims of the EPIC between-cohort calibration is to correct for center differences in systematic dietary measurement errors, our objective was to estimate the validity of the two dietary methods at the ecological level, taking urinary N as the reference. For this purpose, Pearson correlation coefficients were calculated, and results presented before and after exclusion of 1 or 2 centers with outlier values identified by comparing ratios of center mean dietary (24-HDRs or DQs) and urinary N reported in Table 3, and confirmed by visual inspection of the figures. Center means of log-transformed values were considered the units of observation for statistical analyses. Ps < 0.05 were considered statistically significant. Statistical analyses were consistently weighted by the center sample size to take into account heterogeneity in the number of subjects participating in the validation study per center. Results are presented in Figs. 1 and 2 for men and women together and in Fig. 3, a and b stratified by genders. In addition, we performed linear regression analysis to study the relationship between the center means for urine N and dietary N (energy intakes), when dietary measurements (24-HDRs or DQs) were used as the dependent variable. A test was performed to indicate whether the slope of the regression line was significantly different from 1. All of the analyses were performed using the SAS statistical package, version 8 (SAS Institute Inc., Cary, NC).

Between 80 and 96% of the urine samples in Paris, Italy (except Naples), the Netherlands, Oxford, and Potsdam, and slightly less in Heidelberg (78%), Cambridge (72%), and Naples (67%) were considered complete with >85% recovery of PABA (Table 1). In Greece, only ∼20% of the urine specimens were considered complete. After adjustment of N values for subjects with incomplete urine samples but with at least 70–84% PABA recovery, all of the centers had a PABA recovery ranging between 82 and 98% except Greece (49%).

Forty-five subjects without one of the two dietary measurements (i.e., 24-HDR or DQ) were excluded from the statistical analyses, as were individuals with total energy:BMR ratio from the DQ in the top and bottom 1% of the cohort distribution by center and gender, a routine exclusion made on the EPIC baseline questionnaire data. However, such exclusions were not performed on 24-HDRs. Furthermore, to compare only subjects with a balanced protein-energy metabolism, the subjects on a diet during the 24-HDR interview (n = 52) were also excluded from the statistical analysis. Overall, a total of 283 subjects (20.4%) were excluded, and a final sample of 1103 was used for the statistical analysis (Table 1).

Table 2 summarizes some general characteristics of the study populations, and the time interval between dietary and urinary measurements. This study population involved middle-aged people (approximately 53–62 years old), except men and women in the Netherlands, women in Heidelberg, and men in Ragusa who were younger. Overall, there were more women (∼59%) than men, reflecting the different gender distributions in EPIC (∼70% women). The range in average BMI was narrow across countries in men (24.8–26.8 kg/m²) but broader in women (23.2–27.5 kg/m²). BMI in both men (29.0 kg/m²) and women (30.0 kg/m²) was much greater in participants in Greece.

The primary purpose of our analysis was to validate the mean 24-HDR versus 24-h urinary measurements. Therefore, the urine samples were collected at the same time as the 24-HDR interview or within a maximum of 6 days afterward. The time interval between the two DQ collections, and, hence, between the DQ and urine measurements, is longer and varies across countries. In Bilthoven and the German centers, the 24-HDRs were collected from a few days to <2 months after the baseline DQ. In the other centers, 24-HDRs and urine samples were obtained 11–21 months later and >2–5 years after in Turin (men), France, and Naples.

Table 3 reports the geometric means and the ratio between mean N (and total energy) estimated from either the 24-HDR or the DQs and 24-h urine stratified by center and gender. The reported ratios, indicating the degree of systematic under- or over-reporting of dietary N and total energy intakes compared with the reference measurements are presented with and without exclusion of extreme under- or over-reporters according to Goldberg et al. (32). The results presented after exclusion were obtained using the Goldberg cutoff based on either 24-HDR or DQ-specific values (not both).

The ratio of mean N 24-HDR:N urinary ranged from 0.69 (Greece) to 0.99 (Ragusa) for men and from 0.54 (Greece) to 0.92 (Paris) for women. The pairwise t test consistently showed a statistically significant difference between log-transformed means in all of the centers except in men in Italy (except Turin) and women in Ragusa, with average under-reporting of 13% in men and 19% in women when Greece was omitted. The exclusion according to the 24-HDR cutoff of Goldberg et al. (32) did not change the significance of the results except for men in the Netherlands and Heidelberg where the difference between means is no longer statistically different from zero after the exclusion of 14 and 18% of subjects, respectively. For DQs, the ratio of mean N DQ:N urinary ranged from 0.75 (Heidelberg) to 0.90 (Oxford) for men and 0.76 (Heidelberg) to ≥1.04 (Oxford, Naples, France, Florence, and Cambridge) for women. The difference between means is not statistically significant for men in Florence and Ragusa, and for women in Oxford and all of the Italian centers. The exclusion according to Goldberg et al. (32) cut points (percentage of exclusion indicated in brackets) changes the significance of the results in men in Turin (23%) and Cambridge (47%), and in women in France (27%), Heidelberg (47%), Potsdam (32%), and Greece (32%).

The ratio of energy intake:BMR calculated from 24-HDRs ranges from 0.86 to 0.96, except in Varese (1.07) where it is higher but not statistically significant, and lower in Greece (0.70). In women, the ratios are from 0.74 (Oxford) to 0.99 (Turin), and 0.60 in Greece. The mean difference is not statistically significant in the Italian centers (except Varese and Florence in women) and Oxford (men), and after exclusion of men in Bilthoven (14%) and Heidelberg (18%), and women in France (13%).

In the DQ, the ratio of energy intake:BMR ranged from 0.78 (Cambridge) to 1.09 (Ragusa) for men and 0.83 (Bilthoven and Greece) to 1.11 (Naples) for women. Pairwise t test comparison showed no statistically significant difference between means in either gender in Italy. After Goldberg exclusions, the differences in means were no longer statistically significant for women in France (27% exclusion), Cambridge (26%), Greece and Potsdam (32%), and Heidelberg (47%).

Figs. 1 and 2 show the Pearson’s correlation at the group mean level, between urinary N and N, or total energy intake, from 24-HDRs and DQs for both sexes (n = 22). The Pearson’s correlation was weighted according to the number of subjects per center. The correlation between mean center urinary and dietary N was higher with 24-HDRs (0.86) than DQs (0.53). When values for women in Varese and Greece in both genders, which deviated from the others, were excluded, the correlation between urinary N and 24-HDR means increased to 0.94. The correlations between total energy intakes from the 24-HDRs and urinary N were 0.91 (0.95 after exclusion of Greece), and (0.86) for the DQs. When genders were considered separately, the ecological correlation was higher in men than in women with both dietary methods (Fig. 3,a). In men a relatively high correlation was observed for both 24-HDRs (0.87 and 0.92 after exclusion of Greece) and DQ (0.77). In women the correlation was significantly better with the 24-HDRs (0.72) than with the DQs (0.41; P > 0.05), particularly when Varese and Greece are excluded (0.89). The correlation between urinary N was also moderately higher with energy intake calculated from the 24-HDRs (0.71 and 0.76 after exclusion of Greece and Varese) than with energy from DQs (0.55; P > 0.05) in women, but higher for men with DQs (0.89) than with 24-HDRs (0.70; Fig. 3 b). The latter correlation became similar between the two dietary methods after exclusion of Greece in the 24-HDRs (0.88).

In linear regression analyses the β coefficients were not statistically significantly different from 1, the ideal value of proportional agreement, when mean dietary N from the DQs was regressed on urinary N in men (β = 0.92 ± 0.24) and women (β = 0.85 ± 0.55). When N was calculated from the 24-HDRs, β coefficients were also not significantly different from 1 in women (β = 1.73 ± 0.47) and in both genders together (β = 1.29 ± 0.17), but this was not the case in men (1.67 ± 0.29). Additional adjustments for age and BMI considered together or separately in the model did not change the significance of the test in the analysis stratified by gender. However, after adjustment for BMI in women, the β coefficient value obtained by regressing dietary N on urinary N becomes closer to 1 for 24-HDRs (from 1.73 ± 0.47 to 0.99 ± 0.36), whereas moving far from 1 for DQs (from 0.85 ± 0.55 to 0.43 ± 0.59). For the regression of total energy intakes on urinary N, the β coefficients were never statistically significantly different from 1, whatever the gender, the dietary method, and the type of adjustment considered.

When genders were considered together, nonstatistical differences of the β coefficients from 1 were shown consistently, for both dietary methods, and N and total energy intakes, when age, BMI, and sex were added to the model at the same time. Different results were obtained when the adjustment variables were considered separately in the model.

The principal aim of this analysis was to evaluate the validity of mean N intake and indirectly total energy intake estimated from 24-HDRs, and to determine whether 24-HDRs can be used as reference measurements for between-cohort calibration. The same dietary variables, obtained from the nonstandardized dietary methods (DQs) applied on all of the EPIC individuals at baseline, were also compared with the reference urinary measurements to estimate the additional benefit of the calibration approach in EPIC to correct systematic mean population bias.

To our knowledge, this is the first time that dietary measurements used for between-cohort calibration have been validated using an independent biological reference measurement such as urinary N. However, the convenience sample used in the analysis is not strictly representative of the EPIC cohorts, neither does it include all of the EPIC centers. Any extrapolation to the entire EPIC cohort or to the general population needs to be made with caution. Furthermore, in most centers it was not possible to estimate the response rate and the reasons for nonparticipation, which may have biased the results.

Although the time interval between the collection of urine specimens and dietary data, particularly DQs, varied considerably across centers, no statistically significant correlation was observed between the center mean time interval, and the difference in mean urinary and 24-HDR N measurements. This suggests the effect of time interval between measurements on correlations at population levels in this study were, at the most, very modest and undetectable.

Overall, higher Pearson correlations between center mean urinary N and N intakes were observed with 24-HDR compared with DQs, when both sexes and centers were considered together. When genders were considered separately, urinary N was well correlated with both dietary measurements in men, whereas in women the correlation was higher with 24-HDRs than with DQ, particularly when Varese and Greece with outlier values were excluded. Despite the evidence of under-reporting in mean estimates reported in Table 3, A and B, and the relatively low within- and between-subjects physiological variations associated with both N and total energy estimates reported in the literature (34, 35, 36, 37), both dietary methods showed good quantitative agreement when compared with urine measurements, particularly in men. In addition, the β coefficients estimated by regressing mean N intakes from 24-HDR and DQ on mean urinary N (total energy intakes) were not statistically different from 1, except for men for 24-HDRs. When genders were considered together, adjustment for age, BMI, and sex led to the estimation of a slope not statistically different from 1. These results suggest that, although mean center/country dietary N intakes are not correct in absolute values, the order and magnitude of the systematic bias (generally underestimation) of dietary measurements is overall comparable across centers, at least under the conditions considered in this analysis.

In the absence of a completely unbiased reference dietary method, a constant under- (or over-) estimation can still fulfil one of the basic statistical requirements of rescaling group mean values obtained with nonstandardized DQs (38). Furthermore, it suggests that possible methodological errors in N intake estimation (e.g., lack of standardized food composition tables and different baseline DQs) might be quite comparable across centers and/or are small compared with the actual differences in mean N (or total energy) intakes across centers.

Although urinary N, corrected for extrarenal losses, is a good reference measurement for estimating under-reporting of N (protein) dietary intakes and identifying gross under-reporters, it does not provide a precise estimate of the overall degree of under- or over-reporting of total energy intakes (39). Because doubly labeled water measurements were not collected in our study, urinary N was also used as an indirect and surrogate reference of total energy intakes. This assumption is based on the relatively high physiological correlation existing between the intakes of total energy and its main sources, particularly protein. We considered that some insights, even indirect, of the correlation at the center mean level between total energy intake and urinary N could provide useful information for understanding the nature and magnitude of the overall measurement errors. Indeed, a non-neutral differential bias of food and nutrient measurements has been reported in the literature when (high) under-reporters are compared with other population groups (13, 40, 41, 42, 43, 44). In most of these studies (13, 40, 41, 42, 44), under-reporting is usually associated with a relative increase in protein intakes, when expressed as a percentage of total energy, which suggests that proteins are relatively less likely to be grossly under- or overestimated than other macronutrients. This was confirmed in a small observational study where it was observed that snacks, in contrast to main meals, were not well reported, and that, although total energy was underestimated by ∼14%, protein intakes were estimated correctly (43). Studying the validity of protein intake measurements only may lead to underestimation of the overall actual bias, particularly if differential under- or over-reporting in different foods are suspected. Overall, our results show that the 24-HDRs and DQs used in EPIC discriminate centers by mean total energy intakes in close agreement with mean urinary N when all of the centers are considered together, and in men when the genders are considered separately. In women, the correlation with energy intakes was lower, particularly with DQs (0.55; P > 0.05).

In most cases, the value and the statistical significance of the β coefficients did not change significantly when center means were adjusted additionally for age and/or BMI in men, but there were changes in women, particularly with BMI. The gender differences suggested by our results may be explained by different possible factors.

First, 2 centers (Greece and Varese in Italy) were identified as outliers, which affected the Pearson correlation calculated between 24-HDR and reference measurements in women. For example, the correlation between mean urinary N and N estimated from 24-HDRs increased from 0.72 to 0.89 after exclusion of these 2 centers. In contrast to Varese, where the problem is observed only in women, Greek subjects of both sexes systematically under-report by ≥30% with the 24-HDRs but to a lesser extent with DQs. We reanalyzed the Greek 24-HDR in detail, and no systematic errors in the dietary interview procedure were found that could explain this high under-reporting. A low level of completeness of urine samples estimated with the PABA check (20%) compared with the other EPIC centers suggest that a possible effect of subject behavior cannot be excluded, as indicated by other studies on EPIC data (45).

Secondly, we observed that men under-report protein intake systematically with both dietary methods compared with urinary measurements, whereas women systematically under-report with 24-HDRs, but over- or under-report in DQ depending on the center. In Bilthoven, Germany, Greece, and certain Italian centers, women under-report, whereas in the 2 British centers, France, Florence, and Naples, they tend to over-report by about 7–9%, although the difference is not always statistically significant from zero.

Third, as suggested by the reference urinary measurements, the range of mean N intakes is quite narrow (∼23% in both genders) compared with other dietary components, but does not overlap between women (11.44 to 14.8 g) and men (15.1 to 19.7 g). Apart from Greece, the same figure is observed with 24-HDRs, whereas for DQs several centers have averages contained within the range of the opposite gender. However, compared with urine measurements, the range of consumption observed is greater with both dietary methods, highlighting random errors in the measurements, particularly with DQs in women.

Fourth, the absolute range of variation, particularly of total energy intake, is narrower in women, so random measurement errors are more likely to distort the correlation in women than in men. This may be because of the greater range of average work and leisure time activities in men than in women, which has been observed across EPIC centers (46). Using dietary intake and doubly labeled water as reference measurements, Black et al. (47) showed that women tend to report on average 11% less energy than men, after weight, height, and age (i.e., for gender physiological differences) are controlled for.

Fifth, these results may also suggest that women are more likely, consciously or unconsciously, to under- (or mis-) report their diet than men. This is supported by other analyses on the EPIC data (45) and some other studies (48, 49, 50). In respiratory room conditions, it has been shown that there are no gender differences in energy expenditure, after body composition and physical activity are controlled for (51).

The approach of Goldberg et al. (32), expressed as the percentage of subjects with nonplausible physiological total energy values at the individual level, was used as an indirect (but not exhaustive) empirical indicator of the overall quality of the data, and as a possible criterion for exclusion of extreme values. In our analysis, the exclusion according to the Goldberg cutoff showed that under- (or over-) estimation of total energy or N intakes of the mean groups were unlikely to be attributable exclusively to subjects reporting extreme levels of consumption. In most cases, the exclusions according to Goldberg et al. (32) did not change the statistical significance of the test difference between mean dietary N and total energy intakes, and mean reference measurements, except when a high proportion of subjects were excluded, particularly with DQs. The lower number of subjects outside plausible physiological ranges according to the Goldberg cutoff suggests that 24-HDRs are less prone to such extreme values than DQs, which, in addition, vary to a greater extent across centers. Although our results suggest a more systematic problem of measurement errors not specifically attributable, in most cases, to extreme misreporters, the lack of sensitivity of the current Goldberg cutoff points reported recently by Black et al. (52) cannot be disregarded as a possible cause.

In conclusion, the analyses comparing urinary N output with two different methods used to assess N intake showed overall better quantitative agreement when intake was obtained using a highly standardized 24-HDR than with DQs. Surprisingly, despite the lack of standardization, the DQs considered in this analysis also provide relatively good agreement of center mean total N or energy intakes in men when compared with urinary N. However, in contrast to 24-HDRs, errors in DQs are not only different in magnitude across countries but also in direction (over- and underestimation), particularly in women. This may cause problems of misclassification, when pooling different individual questionnaire data without any correction for systematic bias in the baseline measurements, which need to be investigated additionally. Future validation studies on calibration measurements should be designed to have greater statistical power and include more representative samples from all of the EPIC centers, particularly those from the Nordic countries, which joined EPIC later and used different DQs (semiquantitative food frequency questionnaires) than the more extensive quantitative DQs used in this analysis (18). Furthermore, studies using doubly labeled water measurements are needed to better evaluate the validity of total energy intakes.

We thank all of the study participants for their cooperation and all of the persons who participated in the field work studies in each EPIC center.