FFQ are widely used to assess dietary intake(Reference Thompson, Moler and Freedman1, Reference Cade, Thompson and Burley2). They are relatively low cost, are easy to administer and can assess dietary intake over a long period of time(Reference Nelson and Bingham3). In Australia, there are two FFQ that are frequently used to assess dietary intake: the Commonwealth Scientific and Industrial Research Organisation (CSIRO) FFQ and the Cancer Council of Victoria (CCV) FFQ(Reference Baghurst and Record4, Reference Ireland, Jolley and Giles5). The characteristics of each instrument are different. The CCV FFQ is a seventy-four-item self-administered FFQ, which is optically scannable to provide an analysis of dietary intake. It is quick and easy to use and has been validated in relation to 7 d weighed food records in premenopausal women. However, it is thought to underestimate carbohydrate intake; it has a truncated upper range of frequency categories and does not include some common foods(Reference Xinying, Noakes and Keogh6). The CSIRO FFQ contains 200 items, has serving sizes that can be altered by the respondent and can be modified, i.e. extra questions can be added according to the particular focus of the study. However, it is a booklet that takes 45 min to complete and has to be manually coded before the manual data entry, increasing both the respondent burden and the resources required to analyse the FFQ. The CSIRO FFQ had been validated by ourselves and others(Reference Rohan, Record and Cook7–Reference Ambrosini, Mackerras and de Klerk10).
The objectives of the present study were to examine the utility of the shorter FFQ compared with the longer FFQ and to establish the agreement between the CSIRO FFQ and the CCV FFQ, as there are advantages of reduced respondent burden and fewer professional resources of using a machine-readable FFQ.
Methods
Subjects
The subjects in the present study were men (n 159) who had volunteered to be screened for a dietary intervention study to establish the bioavailability and bioefficacy in humans of selenium supplied in high-selenium wheat. The inclusion criteria were healthy men aged 40–70 years, not supplementing with selenium or other vitamins and antioxidants. Exclusion criteria were sensitivity to study foods, i.e. gluten/wheat intolerance, or unable to comprehend or comply with the study protocol. The protocol was approved by the CSIRO Human Nutrition Human Research Ethics Committee and all subjects gave written informed consent.
FFQ
The CSIRO FFQ is a quantified FFQ containing a list of 200 items of food and drink. As a guide to serving size estimates, each food has a designated ‘standard’ serving or portion size described in household measures and/or by weight and volume alongside. This could be changed where necessary by the respondent. Respondents were asked to indicate how often they ate each of these foods using an open-ended alpha-numeric scale (N for never; R for rarely; 1, 2, 3, etc. per D, W and M for once, twice, three times, etc. per day, week and month, respectively). There was a series of qualitative questions related to food preparation and cooking techniques, the use of salt or sugar in beverages, types of milk or breads and types of fats used as spreads or in cooking. The answers to those questions were used where necessary to modify the nutrient profile of the foods in the main frequency list. Information on other commonly eaten foods, not in the main list, was also sought and included in the analysis. The responses were analysed for usual daily nutrient intake and grams consumed at CSIRO (Adelaide, South Australia) using a modification of the FREQUAN dietary analysis programme(Reference Baghurst and Record4). Nutrient composition was derived from the four sources: the Australian nutrient database (NUTTAB)(Reference Lewis, Milligan and Hunt11), the British Food Composition Tables(Reference Paul and Southgate12), the US Department of Agriculture (USDA) food tables(13) and the manufacturers’ data. The range of nutrients that can be obtained from the CSIRO FFQ is greater than the CCV FFQ and can include, for example, selenium, vitamin B12, pantothenic acid, biotin, vitamin B6 and fatty acids. The CCV FFQ had been validated in a number of studies(Reference Rohan, Record and Cook7–Reference Ambrosini, Mackerras and de Klerk10).
The CCV FFQ is a seventy-four-item semi-quantitative machine-scannable FFQ that has been described in detail elsewhere(Reference Hodge, Patterson and Brown14). In short, the first page of the four-page FFQ includes questions on how many pieces of fresh fruit and how many types of vegetables are consumed daily. Other questions are about the amount and type of milk and bread eaten, type of spread used, amount of sugar consumed daily, weekly egg intake and type of cheese eaten. The second page of the FFQ consists of four sets of photos depicting serving sizes; pages three and four of the CCV FFQ are a list of seventy-four items with frequency options ranging from never to three or more times per day, and more detailed information about alcohol.
The food composition data used to calculate nutrients are from NUTTAB95(Reference Lewis, Milligan and Hunt11), supplemented by other data where necessary(Reference Paul and Southgate12, 15, Reference Foster-Powell, Holt and Brand-Miller16, 17). The output includes water, kilojoules, fat (total), protein, carbohydrate (total), sugar, starch and dextrins, dietary fibre, cholesterol, sodium, potassium, calcium, phosphorus, magnesium, iron, zinc, retinol equivalent, retinol, thiamin, β-carotene equivalents, riboflavin, niacin, niacin equivalents, vitamin C, alcohol, SFA (total), MUFA (total) and PUFA (total). Following the present study, the CCV extended the calculations to include individual fatty acids, carotenoids, glycaemic index and glycaemic load. The FFQ and participant barcodes were obtained from the CCV, and the completed FFQ were returned to the CCV for analysis using the software based on the NUTTAB95 food composition database.
The recall period for both FFQ was usual intake defined as during the past year.
Administration of the FFQ
Both the CCV FFQ and the CSIRO FFQ were completed on one occasion in the CSIRO research unit when the volunteers attended for the screening visit for the intervention trial. They were given clear instructions to recall their usual dietary intake over the past year. The questionnaires were not completed in a specified order.
Statistical analysis
Statistical analysis was performed using the Statistical Package for the Social Sciences statistical software package version 16·0 for Windows™ (2008, SPSS Inc., Chicago, IL, USA). In comparison with earlier studies, Pearson correlation coefficients and paired t tests were used to compare the questionnaires; however, the limited usefulness of these methods is acknowledged. Agreement between both questionnaires was assessed according to the Bland–Altman plots and limits of agreement (LOA)(Reference Bland and Altman18) and ordinary least products regression to test for the presence of fixed and proportional bias(Reference Ludbrook19). Proportional bias exists when there is a relationship between the extent of agreement and the level of intake. Anti-logging the Bland–Altman mean agreement and LOA resulted in values being reported as CCV FFQ intake as a percentage of CSIRO FFQ intake. Exact agreement between both questionnaires was indicated by 100 % mean agreement. The criteria for agreement were defined as maximum LOA of 50–200 %, and there was no significant proportional bias. The agreement between the relative rankings of the questionnaires was assessed by classifying subjects into quintiles and cross-tabulating. Weighted κ statistics were also calculated for the strength of agreement. Data are presented as means and sem unless otherwise stated. All intake data were log transformed before the analysis and statistical significance was determined by P < 0·05.
Results
One hundred and fifty-nine men aged 55 (sem 0·5) years, with a BMI of 27·4 (sem 0·3) kg/m2 completed both FFQ. As this was a comparison between two similar methods, it was considered valid to include all responses in the analysis regardless of energy intake.
Pearson’s correlation coefficients were highly statistically significant (all P < 0·001) and for the majority of nutrients ranged from r = 0·512 to r = 0·883 except for retinol for which r = 0·297. Agreement between both questionnaires, according to the Bland and Altman LOA(Reference Bland and Altman18) and ordinary least products regression(Reference Ludbrook19) are shown in Table 1.
Agreement between the CSIRO FFQ and the CCV FFQ according to Bland–Altman limits of agreement(Reference Bland and Altman18) (LOA) and ordinary least products regression(Reference Ludbrook19)
*Pearson correlation coefficient; P < 0·001 for all nutrients.
†Expressed as (mean (CSIRO FFQ − CCV FFQ)) as a percentage of CSIRO FFQ.
‡95 % LOA = mean difference ± 2 (sd differences).
§Slope from linear regression of the mean of the methods against the difference between methods.
||Significant fixed bias if 95 % CI of the intercept from least products regression of CSIRO FFQ on CCV FFQ contains 0.
¶Significant proportional bias if 95 % CI of the slope from least products regression of CSIRO FFQ on CCV FFQ contains 1.
There is significant evidence of a fixed bias for protein, carbohydrate, fibre and calcium with 3–16 % lower values for the CCV FFQ, with the exception of protein for which the CCV FFQ averaged 1 % higher. Protein and calcium also exhibited significant proportional bias. At low calcium intakes, the CCV FFQ reported higher intakes than the CSIRO FFQ and the opposite was true at high calcium intakes. The opposite was true for protein, i.e. at low protein intakes the CCV FFQ reported lower intakes than the CSIRO FFQ and at high protein intakes the CSIRO FFQ reported lower intakes than the CCV FFQ.
The cumulative percentage agreement between the quintiles of nutrient intakes derived from both questionnaires is shown in Table 2. When the data were analysed according to quintiles, approximately 80 % of subjects were either in exactly the same quintile or within one quintile for most nutrients, agreement was 69–74 % for six of the nutrients and 1–2 % were grossly misclassified by three or four quintiles.
Quintiles cumulative percentage agreement between nutrient intakes derived from the CSIRO FFQ and the CCV FFQ
*Gross misclassification = disagreement by three or four quintiles.
Discussion
The finding that the CCV FFQ reported lower carbohydrate was in agreement with our previous findings(Reference Xinying, Noakes and Keogh6). The underestimation of carbohydrate suggests that some key foods, e.g. soft drinks, are missing from this FFQ. However, as fibre also appears to be underestimated, it appears that other carbohydrate-containing foods, e.g. wholemeal bread and fibre-containing cereals, may also be underestimated.
The differences in calcium intake between both questionnaires are more difficult to explain, especially as there is proportional as well as fixed bias such that at low calcium intakes the CCV FFQ reports higher intakes than the CSIRO FFQ, and the opposite is true at high calcium intakes.
FFQ are often used in epidemiological studies and have been shown to perform well in comparison with other dietary intake methods, e.g. a 7 d diet diary, dietary history and biochemical markers and repeated 24 h diet recalls(Reference Brunner, Stallone and Juneja20–Reference Bohlscheid-Thomas, Hoting and Boeing22). A longer version of the CSIRO FFQ was used in an Australian study that linked consumption of trans fatty acids with an increased risk of coronary artery disease(Reference Clifton, Keogh and Noakes23). The CCV FFQ has helped to establish that fruit and vegetable intakes are inversely associated with CVD mortality(Reference Harriss, English and Powles24). An FFQ is being used in the analysis of risk from the diet in relation to cancer incidence in the large European Prospective Investigation into Cancer and Nutrition (EPIC) Study, as it is considered that measurement error will be overcome by the large dietary heterogeneity(Reference Bingham, Welch and McTaggart25). This FFQ has been validated against ten 24 h recalls and found to have an estimated level of validity similar to other FFQ measurements(Reference Johansson, Hallmans and Wikman26). However, in the UK EPIC study, dietary data are collected using a 7 d diary, as correlations between dietary intake and urinary excretion of nitrogen and potassium were low(Reference Bingham, Welch and McTaggart25).
Day et al.(Reference Day, McKeown and Wong27) compared urinary nitrogen, potassium and sodium with intakes from an FFQ and a 7 d diet diary. They found that the bias within the FFQ was much greater than the 7 d diet diary and was greater than that would be deduced if validation studies were based solely on another diet methodology. We have published data showing that protein intake derived from the CSIRO FFQ was correlated significantly with urinary urea and sodium, but not with potassium(Reference Lassale, Guilbert and Keogh8).
Despite their limitations, FFQ continue to be used in large studies such as the Health Professionals’ Follow-up Study, the Nurses’ Health Study and the Physicians’ Health Study(Reference Willett, Stampfer and Chu28–Reference Hall, Chavarro and Lee30). A 124-item food frequency has been used to evaluate the relationship between red meat intake and breast cancer(Reference Linos, Willett and Cho29). A relative risk of 1·36 could be shown when women in the lowest quintile of red meat intake during adolescence were compared with women in the highest quintile. A shortened FFQ was used to examine the association between fish and n-3 fatty acid intake and colorectal cancer risk(Reference Hall, Chavarro and Lee30). In the present study, a relative risk of 0·60 was shown for the highest v. the lowest category of fish intake. Thus, despite their imprecision, FFQ are capable of demonstrating the diet–disease relationships. Both the CSIRO FFQ and the CCV FFQ have been used in a similar way to investigate the diet–disease relationships in adults and children, respectively(Reference Giles, Simpson and English31, Reference Almqvist, Garden and Xuan32).
It is recommended in validation studies that the test measure should be administered before the assessment of the reference measure(Reference Cade, Thompson and Burley2). However, in the present study, the two FFQ were being compared to one another and we believed that it was not necessary to follow this sequence.
We conclude that while both questionnaires rank the majority of subjects in either exactly the same quintile or within one quintile for all nutrients, they are not interchangeable for estimation of individual dietary intakes. However, the study was only in men and the conclusions cannot be generalised to women.
There is a continuing concern that the CCV FFQ appears to underestimate carbohydrate intake and this needs to be taken into account if this methodology is being used. The larger range of nutrients and greater flexibility of the CSIRO FFQ make it a useful instrument.
Acknowledgements
There are no conflicts of interest. The study was partly funded by the National Centre of Excellence for Functional Foods, Australia. J.B.K. designed the study, contributed to the data analysis and interpretation and wrote the manuscript; K.L. provided expert advice on statistical analysis and undertook some of the data analysis; J.S. assisted with the design of the study and contributed to the data analysis. The authors acknowledge the CSIRO Clinical Research Staff for their assistance with the present study.
