The aim of this study was to evaluate the performance of a modified FFQ used in the NESCAV study, against several nutritional biomarkers. Overall, our FFQ performed well in assessing intakes of fruits and vegetables and several micro-nutrients as correlations were within the range noted by others investigators [2, 25, 28–35].
Although, biomarkers provide an objective measure of intake of which the errors are largely independent of the errors associated with FFQ [5, 11], they have several drawbacks. Indeed, while FFQ measures intake, biomarkers measures circulating concentrations that are influenced not only by dietary intake but also by a number of physiological and environmental factors . The effects of genetic, lifestyle, physiologic and others dietary factors may also bear on the relationship between the amount ingested and the biochemical measurement [11, 37]. Although, some of these factors were taken into account by statistical adjustment or restriction, the weak correlations observed between biomarkers and the FFQ-estimated values would be related to their effect . We thus consider the observed correlations to be the lower limit of the ability of the FFQ to measure considered nutrients.
Plasma β-carotene is very sensitive to dietary intake as it is not closely regulated by a homeostatic mechanism . Similar to other studies [31–33], we found correlations around 0.20.
Concerning estimates of fruits and vegetables, although others authors reported higher correlations [40–43], significant positive correlations were found between plasma β-carotene concentration and estimated intakes of vegetables and fruits (r = 0.17 in men and r = 0.1 in women), vegetables alone (r = 0.18 in men and r = 0.11 in women) and fruits alone (r = 0.11 in men and r = 0.1 in women).
For Vitamin B9 intake, we used both serum folate which indicates recent dietary folate intake, and erythrocytes folate, which is an indicator of long-term status . Of the nutrient we considered, folates displayed the highest correlation and kappa coefficients. In men, we obtained adjusted-correlations of 0.25 and 0.29 for energy-adjusted erythrocytes and serum folates respectively. In women, these correlations were equal to 0.25 and 0.32 respectively. These correlations were of the same magnitude to those obtained in previous studies  and even higher than others [29, 30].
For vitamin B12, the observed correlation, around 0.10 suggests a responsiveness of this biomarker to dietary intake. This result was in the range of other studies excluding supplement users .
A single measurement of plasma α-tocopherol, adjusted for blood lipids, appears to be able to represent long-term vitamin E intake to a modest degree [45–47]. However, in several studies, even if total vitamin E intake was positively associated with plasma concentrations of α-tocopherol, this was primarily due to vitamin E supplements since no[38, 48, 49] association were observed among persons who do not use vitamin E-containing supplements. As most of these studies, our correlations coefficients decreased significantly when adjusted for the use of vitamin E supplement (from 0.13 to 0.09 in men; from 0.11 to 0.05 in women). It is possible that there is a poor relationship between intake and serum levels at the range of intakes from diet alone and a good relationship at the range of intakes achieved through dietary supplements use.
Because plasma vitamin D concentrations are influenced by both diet and sunlight exposure , validation of vitamin D intake was undertaken in the late winter when sun synthesis of vitamin D would not be a confounder . In late winter, skin production of vitamin D is nearly null and previous serum vitamin D stores had been depleted or nearly depleted. The observed energy-adjusted correlations (0.28 in men and 0.54 in women), between intake and biomarker suggest a responsiveness of plasma concentrations to dietary intake. But these associations were no longer significant when adjusted for use of vitamin D supplements. Since others [2, 51] reported significant correlations, even after excluding supplement users, we think that our FFQ is not performing well in ranking individuals according to vitamin D intake.
Concerning correlations between urinary NA level and sodium intake, we found correlations of 0.10 in women and equal to 0.07 in men. This is not surprising since there are no questions on sodium intake or use of table salt in our FFQ. Similar validation studies assessing the association between FFQ-derived estimates of NA and urinary NA levels are rare. In a Brazilian study among hypertensive subjects , the FFQ revealed no significant correlations with 24 h urinary Na while another one  found only weak correlation (r = 0.29). An accurate assessment of Na intake implies identification of the sources of Na in the standard diet, yet this is different from the situation for other nutrients which are supplied largely by intrinsic nutrients in specific foods. Na constitutes a part of almost all fresh foods, and it is a major component of industrialized canned and pre-prepared foods. Moreover, table salt and that added when preparing foods are also an important source of Na in the individual diet. Therefore, in order to estimate Na consumption, it is necessary to consider all different sources of dietary Na. It is widely recognized that the observed weak relation between dietary and urinary sodium is attributed to the poor assessment of salt intake by dietary assessment methods, the lack of inclusion of foods prepared with salt in food-composition tables, and the high within-person variability of urinary sodium [54, 55]. It is therefore no surprise that the association between FFQ estimate of sodium intake and excretion is so weak.
In our study, no correlation was found between dietary iron intake and the three selected biomarkers: iron, serum ferritin and transferrin. Some investigators have found correlations between dietary iron intake or iron-rich foods and serum ferritin levels [56, 57]. Reliable surrogate biomarkers of the total quantity of dietary iron are unavailable because of the wide variation in bioavailability of dietary iron (for instance heme vs non-heme iron), inter-individual variation in biological availability of dietary iron, interactions between dietary iron and absorption enhancers and inhibitors, variations in physiological (menstruation, childbirth) or unphysiological (blood donation) iron losses and uncertain food composition data. Therefore, the lack of correlation for iron and its biomarkers is probably because of the insensitivity of plasma concentrations to intake of iron.
Concerning urine iodine:creatinine ratio, correlations were higher in women (r > 0.20) than in men (r around 0.10). Although, one study reported a correlation coefficient of 0.66 , our results are similar to those previously published (r = 0.16 (34), r = 0.24(35)).
The main strength of this study is that we compared FFQ-estimated intakes with objective biochemical measurements. Although these analyses give the lower limit of the FFQ validity, they allow to avoiding correlated errors between FFQ and others self-reported measures. In addition, the size of the sample allows investigation of the validity across important subject characteristics such as gender, age, smoking status, BMI, diploma and supplements’ use. No notable differences were observed between the different characteristics except between men and women, smokers and no smokers for vitamin E and between supplements’ users for vitamin B9 and D. Intakes of fruits and vegetables were better measured in men than in women but this did not reflect a clear gender difference in the quality of response as the correlation for micro-nutrient tend to be higher in women. Thirdly, as the recruitment of participants took place throughout the year, the four seasons were well represented and therefore the intakes of all foods were covered. Finally, the socio-demographic characteristics of participants in the validation study were very similar to those of the overall study participants (data not shown). Therefore, we can assume that the rest of the NESCAV sample performs in the same way.
The limitation of this study was that we did not use recovery biomarkers. Validation study should ideally be carried out using recovery biomarkers such as doubly labeled water, markers of potassium and nitrogen in 24 h urine collections to validate total energy intake, potassium and protein intake respectively. Recovery biomarkers are considered the gold standard but the availability and expense of those biomarkers made their use not possible for the validation of this questionnaire.
In summary, results of the biomarker-analyses, in combination with our previous findings from comparison between the FFQ and DR allow to evaluate the overall validity of our FFQ. In the previous study comparing estimates of several macro- and micro-nutrients computed from the FFQ and DR, the relationships between the two measurement tools were satisfactory . However, Results for protein, cholesterol, starch, vitamins A, E and B12 ought to be interpreted with caution. In the present validation study, our FFQ performed well in ranking most of micro-nutrient, particularly fruits and vegetables intakes. Worth noted, despite the absence of agreement for vitamin B12 in the comparison with DR, significant correlations were observed in biomarkers study. However, sodium and vitamin D and E will have to be use carefully since no significant associations were observed after adjustment.