Accurate assessment of dietary intakes and dietary changes plays a central role in nutritional studies. Each tool used to evaluate dietary intakes has some strengths and limitations. Also, all standard dietary assessment methods are subjected to bias such as underreporting . In our study, we developed a 91-items interviewer-administered FFQ that was sufficiently accurate to measure intakes of nutrients in the habitual diet of subjects from the Québec City metropolitan area and changes in nutrient intakes following a 12 week intervention promoting the Mediterranean food pattern.
In the validation study, coefficients of correlation between values derived from the FFQ and those obtained by the 3-day food record ranged from 0.30 to 0.60 for macronutrients and from 0.19 to 0.56 for micronutrients at week 0. It has been previously reported that correlation coefficients for validation studies ranged from 0.4 to 0.7, similar to our results after energy adjustment [6, 15]. Also, our interviewer-administered FFQ did not significantly overestimate energy intake compared to a 3-day food record. The fact that the interviewer used food models to facilitate the estimation of portion size can contribute to explain this findings. It has been shown that FFQ can both under- and overestimate intakes of specific nutrients. In fact, many validation studies have reported that FFQ, as compared to food-record or 24-hour recall overestimate nutrients intakes as well as energy intake [16–20]. In contrast, other studies have reported that FFQ did not systematically overestimate energy and nutrients intakes [14, 21–23].
Despite the fact that we obtained similar values for energy intake with both dietary methods, we can not exclude the possibility that both tools are subjected to underreporting and therefore underestimate usual dietary intakes. It has been previously suggested that subjects may tend to underreport actual food intake by as much as 20% when completing a weighted dietary record . It has been argued that subjects who complete 3-day food record may change their nutritional food habits in order to simplify the recording of food intakes or to impress the dietitian. Also, errors in 3-day food records can be attributable to interpretation of the dietitian encoding the records. In our study, the same dietitian verified all the food records to make sure that dietary data were coded similarly for all participants. In the present study, 38% of subjects included in the validation study at week 0 had a ratio between energy intake to estimate BMR below 1.35. Considering that they had to be weight stable to be included in the study it is likely that these women were underestimating their habitual diet. Black et al concluded in a review that underreporting was observed in a great majority of nutritional surveys independently of the method used . Earlier studies conducted in lean women demonstrated that underreporting was mainly explained by undereating  or underreporting snack foods  whereas in obese subjects underreporting could be explained by an underestimation in recording portion size and to social desirability. In addition, underreporting occurs more often among foods considered 'bad' or 'unhealthy' . In our validation study, there were no significant differences between BMI of women who were considered as underreporters and women who did not underreport (not shown).
In a nutritional intervention, interpretation of the study outcomes with regard to dietary changes will depend not only of the validity and the reproducibility of the method used but also of the sensibility of the method to detect dietary changes in response to the intervention. In our nutritional intervention study, conducted in a sample of healthy women, both diet assessment methods detected similar dietary changes over the duration of the intervention. These findings suggested that our FFQ is sensitive to dietary changes in response to our intervention and could be used to assess dietary changes during a nutritional intervention. Our results are in agreement with study that showed that in response to a nutritional intervention a FFQ measured similar dietary changes as compared to 24-hour recalls  or 4-day food records .
The major differences between the two methods in our study were noted for total lipids and MUFA intakes. Our FFQ was designed to assess precisely lipid intake and many questions were asked about types of fat used to spread or to cook. The more important differences between FFQ and 3-day food record for MUFA and lipids could therefore be explained by the fact that it was difficult for participants to report precisely their lipid consumption when completing the FFQ. It has also been reported in obese men that underreporting of food record is usually specific to lipid intake  and it is thus possible that some women did not record all fats or foods high in fat consumed when completing their 3-day food record. Therefore, it is difficult to determine whether our FFQ tended to overestimate lipid or whether the 3-day food record tended to underestimate it. Also, dietary changes for these nutrients were in the same magnitude in response to our intervention with both methods. On the other hand, Mediterranean diet is usually considered high in MUFA. In North America, MUFA are mostly provided by partially hydrogenated vegetable oils and animals products . In that context, MUFA to SFA ratio could be considered as a better indicator of a Mediterranean diet. In our study, we noted that this ratio was not different between the two methods at baseline and changes observed in response to the nutritional intervention did not differ significantly (not shown).
The agreement in quartile classification was acceptable for selected nutrients with a mean of 35.1% of subjects who were in exact agreement and 5.1% who were misclassified in extreme quartiles. This finding is similar to previous observations [16, 19, 32, 33]. In many studies, classification in the same segment of the distribution using two different methods is found in 30% to 40% of subjects [16, 32, 34].
When analyses were performed at week 6 and 12 after the beginning of the nutritional intervention, coefficients of correlation were slightly higher than at week 0. We suggest that this finding be partly explained by the intervention effect. As previously reported  subjects could be influenced by a learning effect. In fact, subjects could be influenced by the first FFQ experience and be more adequately prepared for the second FFQ. The nutritional intervention may have also influenced the manner in which subjects were completing their 3-day food records during the study.
In a nutritional intervention, it is important to use a reproducible method to insure that dietary changes observed are due to the intervention effects and not to the instrument error. Our study suggests that the FFQ presents a good degree of reproducibility. In fact, in reproducibility studies the coefficients of correlation generally ranged from 0.5 to 0.7 . In our study, coefficients of correlation ranged, after energy adjustment, from 0.62 for vitamin C to 0.83 for protein intakes. These values are similar to correlations reported by others [14, 18, 19, 21, 22, 33, 35–37].
In our reproducibility study, lower mean energy intake and nutrient intakes were found at the second administration of the FFQ as compared to the first FFQ (difference of approximately 10%). However, relatively high and uniform correlation coefficients for values derived from the two FFQs were observed. Riley et al  also reported with an administered FFQ that energy intake was 10% lower at the second FFQ administration and this reduction was uniform for all nutrients studied. In our study, intakes of most nutrients were systematically higher when measured with the first FFQ compared to the second one, except for alcohol consumption, which remained the same. Seasonal variation can not explain this difference because both FFQs were administered during the same season. The fact that subjects estimated a lower frequency of intake during the second administration of the FFQ may be explained by their earlier experience in completing the FFQ. Better general knowledge of dietary intakes could lead to a readjustment in estimation of intakes after the first administration of the FFQ and therefore changes in estimated energy intake.