There is ample evidence regarding the role of dietary factors on the development of different chronic diseases. Nevertheless, assessing dietary patterns both at the individual and population levels is a difficult task due to the extensive variability of intake.
Multiple methods have been described to ascertain nutrient intake, with the Food Frequency Questionnaire (FFQ) being the instrument of choice in large nutritional epidemiological studies. Two other dietary instruments that are commonly used are the 24-h dietary recall and the dietary record (DR). It is generally accepted that all these methods have advantages and limitations and none of them is entirely satisfactory.
A universal epidemiological method to ascertain individual dietary intake does not exist. All methods have some kind of errors, although these are not always of a similar magnitude. Thus, it is a difficult task to know what the best method consists of if all of them are imperfect. The objective of a validation method is to compare a nutritional assessment method with another considered as superior, but never as the absolute truth(Reference Willett, Lenart and Willett1). One of the most important issues for avoiding false interpretations is the independence of errors among the different methods evaluated.
Numerous validation studies designed to compare the available methods to assess dietary intake are cited in the scientific literature. The most frequently applied method to ascertain dietary intake is the FFQ, whereas DR and recalls are most utilised as reference methods. The diversity of the results obtained does not allow us to draw conclusions about the selection of an ideal dietary assessment method.
Thus, the aim of this analysis was to identify the most accurate method to assess vitamin intake in the adult population through an extensive review of the literature.
Material and methods
A MEDLINE and EMBASE literature search was carried out between July 2007 and March 2008. The procedure for the identification and selection of articles was performed in three steps:
At stage 1, a search strategy was established to identify the most relevant studies in the electronic databases.
The search terms used in the electronic databases were divided into two categorical strategies:
(1) General. The MeSH terms applied in the general search were: nutrient terms (‘nutritional assessment’ OR ‘diet’ OR ‘nutritional status’ OR ‘dietary intake’ OR ‘food intake’); validity terms (‘validity’ OR ‘validation study’ OR ‘reproducibility’ OR ‘replication study’ OR ‘correlation coefficient’ OR ‘correlation study’) and human studies.
(2) Vitamin specific. The MeSH terms applied were: nutrient terms (vitamin names and synonyms) and intake terms: (intake* OR diet).
At stage 2 of the review, titles and abstracts of the selected articles were read by two independent reviewers. Only when they both determined that titles/abstracts met the exclusion criteria were the articles excluded. When a title/abstract could not be rejected with certainty, the full text of the article was obtained and further evaluated. We applied the following exclusion criteria:
(1) Articles exclusively assessing macronutrients and/or energy.
(2) Studies describing the content of foods in nutrients, additives or contaminants.
(3) Studies in diseased or institutionalised persons exclusively.
(4) Articles presenting reference values for food consumption, nutrient intake, biochemical markers and anthropometric measurements.
(5) Articles establishing associations between food consumption, nutrient intake, biological variables, biochemical markers and anthropometric measurements.
(6) Studies relating diseases to food consumption or nutrient intake.
(7) Intervention studies and other therapeutic studies with nutrients or drugs related to the metabolism of these nutrients.
(8) Calibration studies and those discussing statistical methods.
(9) Studies evaluating the physiological effects of foods, nutrients and in relation to their genetic determinants.
(10) Studies in animals and those without abstracts in PubMed.
At stage 3, the following criteria were considered to select the articles for inclusion:
(1) Studies regarding validation results for vitamin intake (those articles analysing only reproducibility or supplement use were excluded from the present analysis).
(2) Studies based on the adult population (those articles based on children, adolescents, pregnant women or the elderly were excluded from the present analysis).
The full text of all articles collected was screened for definitive exclusion or data extraction by a different reviewer from those involved in the acceptance process, with independent duplicate assessment of a random sample of 25 % by a second reviewer. Where the two reviewers disagreed the study was discussed and a consensus decision reached where possible. If this was not possible, then a third reviewer was asked to arbitrate.
The articles included in the present study were categorised according to the total number of days over which the reference methods were applied:
(1) Long-term intake. If the reference method was a dietary assessment method (including 24-h recall and estimated and weighed DR) applied 7 or more days.
(2) Short-term intake. If the reference method was a dietary assessment method (including 24-h recall and estimated and weighed DR) applied less than 7 d.
(3) Biomarker. If the reference method was a biomarker.
To assess the quality of the different validation studies, a quality score system was developed(Reference Serra-Majem, Frost Andersen and Henriquez- Sánchez2). The studies were scored considering its sample size, the statistics used to validate the method, the procedure of data collection, and the inclusion or not of seasonality and vitamin supplements use, according to the values described in Table 1. The CC obtained for each study were weighed in proportion to the article quality index.
SES, Socioeconomic status.
Results
Five thousand four-hundred and seventy-six articles were identified in the initial search strategy. After applying the exclusion criteria, 392 articles from the general search remained in the review. We applied the inclusion criteria in stage 3 obtaining the following articles for each vitamin: 76 articles for vitamin A; 108 for vitamin C; 21 for vitamin D; 75 for vitamin E; 47 for folic acid; 19 for vitamin B12; 21 for vitamin B6; 49 for thiamine; 49 for riboflavin: 30 for niacin, extracted from a total of 124 studies.
In the present review, most of the validation studies used a FFQ as the intake assessment method. For each vitamin, the methods used as gold standards to validate the FFQ are described in Table 2.
To measure the validity of the different FFQ according to the type of reference method utilised, weighted means of the CC were calculated using the quality index value of each study as the weight (Table 3). Weighted CC for the FFQ ranged from 0·41 to 0·53 when the reference method was the DR, and from 0·43 to 0·67 when the recalls were utilised as the gold standard. Only a few studies (n 33) used biomarkers as a validation method, which yielded lower correlation values (coefficients between 0·26 and 0·38).
Tables 4–11 describe the purpose and scope of the literature examined in the present review: author and year of publication, population size (ranged from 20 to 860 persons) and sex (majority were females), characteristics of the FFQ (mode of administration, number of food items included in the questionnaire and reference period), characteristics of the reference method (collection of information and total number of days over which the reference method was applied), the use or not of supplements, the correlation coefficient and the quality index.
No., number; Suppl., supplements; W, women; M, men; DR, dietary record; 24-h R, 24-h recall.
* Age, sex and energy-adjusted, Pearson correlation coefficient.
† Crude, Spearman correlation coefficient.
‡ Crude, Pearson correlation coefficient.
§ Energy-adjusted, deattenuated correlation coefficient.
∥ Energy-adjusted, Spearman correlation coefficient.
¶ Intra-class correlation coefficient.
** Energy-adjusted, Pearson correlation coefficient.
†† Low-food diversity.
‡‡ High-food diversity.
No., number; Suppl., supplements; W, women; M, men.
* Energy-adjusted, deattenuated correlation coefficient.
† Energy-adjusted, Pearson correlation coefficient.
‡ Crude, Pearson correlation coefficient.
§ Crude, Spearman correlation coefficient.
No, number; Suppl., supplements; W, women; M, men, Bl, black; Wh, white.
* Crude, Spearman correlation coefficient.
† Multivariable-adjusted correlation coefficient.
‡ Energy-adjusted, deattenuated correlation coefficient.
§ Energy-adjusted, Spearman correlation coefficient.
∥ Crude, Pearson correlation coefficient.
No., number; Suppl., supplements; W, women; M, men; DR, dietary record; Bl, black; Wh, white*; (24-, 48-, 72-) h R, (24-, 48-, 72-) h recall.
* Multivariable-adjusted correlation coefficient.
† Crude, Spearman correlation coefficient.
‡ Energy-adjusted, Pearson correlation coefficient.
§ Intra-class correlation coefficient.
∥ Energy-adjusted, deattenuated correlation coefficient.
¶ Energy-adjusted, Spearman correlation coefficient.
** Crude, Pearson correlation coefficient.
No., number; Suppl., supplements; M, men; DR, dietary record; W, women; 24-h R, 24-h recall.
* Crude, Spearman correlation coefficient.
† Crude, Pearson correlation coefficient.
‡ Energy-adjusted, deattenuated correlation coefficient.
§ Energy-adjusted, Spearman correlation coefficient.
∥ Energy-adjusted, Pearson correlation coefficient.
¶ Intra-class correlation coefficient.
‡‡ Age, sex and energy-adjusted, Pearson correlation coefficient.
** Low-food diversity.
†† High-food diversity.
No., number; Suppl., supplements; W, women; M, men.
* Energy-adjusted, deattenuated correlation coefficient.
† Crude, Pearson correlation coefficient.
‡ Crude, Spearman correlation coefficient.
No., number; Suppl., supplements; W, women; M, men; RBC, red blood cells.
* Crude, Pearson correlation coefficient.
† Multivariable-adjusted correlation coefficient.
‡ Energy-adjusted, deattenuated correlation coefficient.
§ Crude, Spearman correlation coefficient.
No., number; Suppl., supplements; W, women; M, men; DR, dietary record (no. of records/no. of days per record); (24-, 48-, 72-) h R, (24-, 48-, 72-) h recall; RBC, red blood cells.
* Crude, Spearman correlation coefficient.
† Energy-adjusted, deattenuated correlation coefficient.
‡ Intra-class correlation coefficient.
§ Multivariable-adjusted correlation coefficient.
FFQ v. dietary record
The DR was used as the gold standard in most of the validation studies included in the present systematic review. In the majority of the cases, information regarding dietary intake was collected through a self-administered FFQ (82 %) to assess dietary intake in the previous 12 months (58 %). The number of food items included in the questionnaire ranged between 22 and 350. The weighted CC varied according to the number of food items ( < 100 or ≥ 100 food items) included in the FFQ (Fig. 1).
Only 31 % of the studies included the intake of vitamin supplements in their analyses. There were no large differences in CC between studies that did or did not include information on vitamin supplements (Fig. 2).
In 43·7 % of the cases, the CC were higher when the DR used as the reference method was a weighed DR compared to the use of an estimated DR (Fig. 3).
Most of the DR used as reference methods collected dietary intake for 7 d or more (long-term intake; 74 % of the studies). Fig. 4 shows the difference in the weighted CC according to the number of days included in the DR (long- v. short-term intake).
FFQ v. recall
When the reference method was the recall, information was more frequently collected through an interviewer-administered FFQ (50 %) instead of being self-administered by the study participants. The number of food items included in the FFQ ranged between 47 and 222. Fig. 5 shows the weighted CC according to the number of food items included in the FFQ.
The proportion of studies in this group including information about vitamin supplement intake was lower than that of those using DR as the gold standard (23 %). Indeed, there were no data on supplement intake for the B-complex vitamins with the exception of one study (Fig. 6).
More than half of the recalls (55·6%) used to validate the FFQ collected dietary information during 7 days or more. When long-term intake was evaluated, the CC for the B-complex vitamins decreased (Fig. 7).
FFQ v. biomarkers
The use of biomarkers as the reference method was less frequent, resulting in CC often much lower than 0·40. Folate intake collected through a FFQ was validated using serum folate in three studies. One study used erythrocyte folate as the reference method, and another three studies used both serum and erythrocyte folate as biomarkers. In fifteen studies, vitamin C intake was compared with its blood levels. Plasma concentration of vitamin E was used as the gold standard in seventeen studies validating vitamin E intake. Moreover, in four other studies, adipose tissue concentration was also used as biomarker.
Discussion
Our aim was to determine the comparative efficacy of available methods to validate dietary intake. The present review shows that FFQ are the most commonly used method for assessing diet in epidemiological studies. The main advantages are their low cost and their capability to characterise the usual diet in the past, as well as to minimise the risk of serious interviewer(Reference Willett, Lenart and Willett1) and measurement bias, given that they can be self-administered.
The drawbacks of the FFQ include the use of fixed lists of foods, the effect of memory, the difficulties in portion size estimation and the interpretation of questionnaires(Reference Willett, Lenart and Willett1, Reference Bingham, Nelson, Margetts and Nelson3). Among the available reference methods for the validation of FFQ, dietary records (DR) are likely to have the least correlated errors, whereas dietary histories as gold standards are considered the least appropriate(Reference Willett, Lenart and Willett1).
The validity of other methods was also evaluated. However, their relatively low number and the many different reference methods that were used led us to focus on FFQ. The vast majority of FFQ were self-administered. Despite the fact that data collection was simplified, their incompleteness is a serious handicap as well as their lack of precision, given the large interpersonal variability in diet recalls(Reference Bingham, Nelson, Margetts and Nelson3). In a review of FFQ validation studies, Cade et al. (Reference Cade, Thompson and Burley4) found that CC always improved, with the exception of vitamin C, when questionnaires were administered by an interviewer compared to those that were self-administered.
Time-frame concordance between FFQ and the reference method is even more crucial. The period of time which the dietary intake is referring to depends on the objectives of the study(Reference Nelson, Margetts and Nelson5). Usually, FFQ are designated to measure diet during the preceding year, whereas the reference methods do not cover the same time period. Multiple DR or recalls must be collected during the study period, and there is also a need to take into account seasonal variability, especially if we are interested in vitamin intakes, which are highly influenced by market availability. We have found that dietary intakes correlate better when the number of days covered by the reference method increases, except for B-complex vitamins when recall methods were used.
The order in which the FFQ and the reference method are applied is also decisive, considering that the results of the first measurement can affect those collected later on. We recommend that FFQ data be collected before gold standard measurement(Reference Nelson, Margetts and Nelson5). In many studies that were reviewed, FFQ were applied twice, before and alter the reference period for the gold standard, in order to evaluate its reproducibility. However, the results depend on the vitamins considered.
The number of foods included in the FFQ has been classically considered as a key component to assure validity of dietary intakes. Using a meta-analysis, Molag et al. (Reference Molag, de Vries and Ocké6) highlighted this point as the major determinant for ranking individuals according to their intakes. In the same vein, we have also found an improvement of the correlation when the numbers of food items surpassed 100: shorter questionnaires ( < 100 food items) yielded worse CC (mean 0·47) than those including more items (0·52), the latter having less variation as well.
However, it is also clear that the administration of longer FFQ is more expensive and participation rates may decrease. The use of extensive food lists is said to give less reliable results than shorter forms(Reference Willett, Lenart and Willett1), or even less information. The foods to be included in a FFQ should be restricted to those that are the principal source of the nutrient(s) of interest, and the frequency of their consumption must also be considered.
Supplements should be present in any dietary data assessment. We observed that data from FFQ and the reference method correlate better when specific questions about supplement intake are included, provided that they are asked for with the same emphasis. We stress the need to ask for the type and dosage of supplement use.
DR were the most commonly used reference method to validate the vitamin intakes measured by FFQ. We find narrower ranges of CC when DR are used, and so they may be less variable than recalls. Correlations are probably suboptimal when methods are used which share the inherent errors associated to FFQ, such as lack of memory and estimation of portion sizes.
Biomarker characteristics are responsible for the observed low correlations. Specifically, for vitamin C, plasma levels only show recent intake. Concerning vitamin E, for which more biomarker-based validation studies have been done, correlation is even worse, given that only four studied its level in adipose tissue, which is the more appropriate marker for usual vitamin E intake.
In any case, the highest correlation coefficient observed, weighed by quality, was 0·5. In light of this, we recommend the application of correction factors to any population-based nutrition study.
Acknowledgements
The studies reported herein have been carried out within the EURRECA Network of Excellence (www.eurreca.org), financially supported by the Commission of the European Communities, specific Research, Technology and Development (RTD) Programme Quality of Life and Management of Living Resources, within the Sixth Framework Programme, contract no. 036196. This report does not necessarily reflect the Commission's views or its future policy in this area. P. H.-S., A. S.-V. and J. D.-A. were responsible for designing search strategies, retrieving references, critical reading and evaluation, and for writing the article. A. O.-A. participated in retrieving references. K. P. participated in data gathering. L. S.-M. designed the analysis and presentation of data and supervised the article editing, which has been read and approved by all the authors. The authors have no conflict of interests to report.