Development of a 20-item food frequency questionnaire to assess a ‘prudent’ dietary pattern among young women in Southampton
To develop a short food frequency questionnaire (FFQ) that can be used among young women in Southampton to assess compliance with a prudent dietary pattern characterized by high consumption of wholemeal bread, fruit and vegetables, and low consumption of sugar, white bread, and red and processed meat.
Diet was assessed using a 100-item interviewer-administered FFQ in 6129 non-pregnant women aged 20–34 years. In total, 94 of these women were re-interviewed 2 years later using the same FFQ. Subsequently, diet was assessed in 378 women attending SureStart Children's Centres in the Nutrition and Well-being Study (NWS) using a 20-item FFQ. The 20 foods included were those that characterized the prudent dietary pattern.
The 20-item prudent diet score was highly correlated with the full 100-item score (r=0.94) in the Southampton Women's Survey (SWS). Both scores were correlated with red blood cell folate (r=0.28 for the 100-item score and r=0.25 for the 20-item score). Among the women re-interviewed after 2 years, the change in prudent diet score was correlated with change in red cell folate for both the 20-item (rS=0.31) and 100-item scores (rS=0.32). In the NWS a strong association between the 20-item prudent diet score and educational attainment (r=0.41) was observed, similar to that seen in the SWS (r=0.47).
The prudent diet pattern describes a robust axis of variation in diet. A 20-item FFQ based on the foods that characterize the prudent diet pattern has clear advantages in terms of time and resources, and is a helpful tool to characterize the diets of young women in Southampton.
Food frequency questionnaires (FFQs) are a popular tool in nutritional epidemiology and enable estimates of habitual diet in a large population. The aim of FFQs is usually to gain information regarding the whole diet, so that intakes of a broad range of nutrients can be estimated. However, in some situations there is interest in only one aspect of diet, such as vitamin D and calcium (Severo et al., 2009) or fat (Rohrmann and Klein, 2003) intake, and thus the use of a short FFQ, which determines consumption of only those foods most relevant to the nutrients under consideration, can be as efficient (Byers et al., 1985) but require fewer resources.
Dietary patterns analysis is becoming more popular in nutritional epidemiology. Kant reviews various methods of dietary patterns analysis and describes how patterns defined using cluster analysis and principal component analysis (PCA), or the related technique of factor analysis, have been shown to be predictive of mortality, morbidity and disease-related biomarkers (Kant, 2004). A common pattern described among populations from several countries and with varying demographics is a prudent/healthful dietary pattern, characterized by high intakes of fruit, vegetables, whole grain, fish and poultry (Kant, 2004). This pattern has been found to be positively associated with micronutrient intake (Kant, 2004) and negatively associated with saturated fat intake (Crozier et al., 2006). As a prudent dietary pattern has been observed earlier to be the most important dietary pattern among young women in Southampton (Robinson et al., 2004; Crozier et al., 2006, 2008), the question arises whether a short FFQ that only includes these characteristic foods could be used to ascertain this same information. If dietary pattern scores using a reduced number of items are highly correlated with those using all items, a short FFQ could be a useful tool to provide information regarding variation in the prudent dietary pattern.
Two earlier studies have analysed the use of reduced-item scores as derived by factor analysis. Schulze et al. (2003) used the six most influential variables from a factor analysis of 49 variables, and created a ‘simplified score’ that summed the standardized frequencies of consumption, without using the coefficients from the factor analysis; this simplified reduced-item score had a correlation of >0.95 with the original factor. Mishra et al. (2006) used the same method on binary food intake data in an analysis of changes in adult dietary patterns in a large birth cohort study in the United Kingdom. From a factor analysis of 126 variables they used between 19 and 32 variables to create simplified reduced-item scores for patterns in both men and women. However, they did not report the association between the reduced-item scores and the full factor scores.
It is important to understand the strengths and limitations of reduced-item scores in capturing dietary patterns if they are to be used to develop short FFQs. In addition, no earlier work has compared a reduced-item score with biomarkers, or determined whether reduced-item scores can capture change in diet over time. We have used data from a large sample of non-pregnant women to analyse these issues; a subset of the women had repeat data 2 years later. We have also developed a short FFQ based on the discriminating foods to assess dietary patterns in another study of women in the same city.
Materials and methods
The Southampton Women's Survey
The Southampton Women's Survey (SWS) has assessed the diet, body composition, physical activity and social circumstances of a large group of non-pregnant women aged 20–34 years living in the city of Southampton, UK. Full details of the study have been published earlier (Inskip et al., 2006). Women were recruited between April 1998 and December 2002 through general practices across the city. In total, 12 583 women agreed to take part in the survey, 75% of all women contacted. Trained research nurses obtained dietary information using a 100-item validated FFQ. Prompt cards were used to ensure standardized responses to the FFQ (Robinson et al., 1996). Data are presented here from the first 6129 women who were recruited from practices in the western half of the city between April 1998 and June 2000. A subgroup of 94 SWS women who had not become pregnant were followed up 2 years after their initial interview for a second assessment of diet, lifestyle and body composition (between November 2000 and February 2001) (Borland et al., 2008). Women were asked to give a blood sample during the second half of their menstrual cycle and again if they were part of the repeat study. Immediately after collection the samples had a full blood count performed and red cell folate was assessed by microparticle enzyme immunoassay using an Abbott IMX machine (Abbott, Columbus, OH, USA) (Steijns et al., 1996). Coefficients of variation of the assays were <10%. Measurements are available for 3983 (65%) of the initial sample and 51 (54%) of the repeat sample. The SWS was approved by the Southampton and South West Hampshire Local Research Ethics Committee and written consent was obtained from participants.
Nutrition and Well-being Study
The Nutrition and Well-being Study (NWS) has assessed diet and a range of psychological measures among a group of 378 women attending SureStart Children's Centres in 19 locations around the city of Southampton. SureStart Centres provide universal and social care services for families with children under the age of five. They offer a range of services such as parenting support, employment advice, women's groups, cooking and food preparation classes. Women in these centres were interviewed between June 2007 and November 2007 (Barker et al., 2009). The NWS was approved by the University of Southampton School of Medicine Research Ethics Committee and written consent was obtained from participants.
Principal component analysis of dietary data
Food frequency responses for the SWS initial and repeat data, and in the NWS, were standardized using the mean and s.d. of the relevant food item in the SWS initial data. PCA is a statistical technique that produces new variables that are uncorrelated linear combinations of the dietary variables that maximize the explained variance (Joliffe and Morgan, 1992). PCA was performed on the reported weekly frequencies of consumption of 100 foods in the SWS. Individual dietary pattern scores were calculated by multiplying the coefficients for the 100 foods by each individual's standardized reported frequencies of consumption, to provide a score for every participant. On the repeat data ‘applied’ scores (Northstone and Emmett, 2008) were generated using the coefficients from the initial data; both the initial and repeat scores were standardized using the mean and s.d. of the initial scores. The reduced-item scores were calculated by multiplying the coefficients for the reduced number of foods by each individual's standardized reported frequencies of consumption. Both initial and repeat reduced-item scores were standardized using the mean and s.d. of the initial reduced-item score.
The red blood cell folate measures were transformed to normality using a logarithm transformation. T-tests were used to compare two groups of normally distributed variables and Mann–Whitney rank-sum tests to compare two groups of non-normally distributed variables. Pearson's correlation coefficients were used to measure association between two normally distributed continuous variables and Spearman's correlation coefficient in which at least one variable was not normally distributed. Normally distributed variables were age, educational attainment, prudent diet scores and log red blood cell folate levels; non-normally distributed variables were number of children in the household, change in prudent diet scores, change in red blood cell folate levels and frequency of food intakes. Agreement was assessed using Bland–Altman limits of agreement (Bland and Altman, 1986). Statistical analysis was performed using Stata 10.0 (StataCorp, College Station, TX, USA) (StataCorp, 2007).
Of the first 6129 women recruited to the SWS, 6125 provided complete dietary information. These 6125 women form the SWS analysis sample in this paper. Similarly, of the 378 NWS women, 377 provided complete dietary information and form the NWS analysis sample in this paper. Study characteristics of the 6125 SWS participants and 377 NWS participants are given in Table 1. The participants in the two studies were of a similar age with a mean around 28 years. The SWS women had somewhat higher educational attainment and tended to live with fewer children (both P<0.001). In all, 33% of the SWS women smoked at the time of interview and they had a median body mass index of 23.9; this information was not collected in the NWS.
Table 1: Descriptive statistics for 6125 women in the SWS cohort and the 377 women in the NWS cohort
Full size table
Principal component analysis in the Southampton Women's Survey
Principal component analysis was performed in the SWS using the 100 foods in the FFQ. This revealed a first principal component that was characterized by high intakes of vegetables (peppers, tomatoes, vegetable dishes, courgettes, green salad, onions and spinach), wholemeal bread, vegetarian food and pasta, and low intakes of full-fat milk, beef, crisps and savoury snacks, Yorkshire pudding and savoury pancakes, white bread, sugar, gravy, sausages, meat pies and roast potatoes. The coefficients of the 20 most influential foods (those with coefficients of highest magnitude) are given in Table 2, alongside the means and s.d. used to standardize the foods in the analysis.
Table 2: Mean (s.d.) frequency per week and PCA coefficients for the 20 most influential foods and food groups in the 100-item SWS prudent diet score (n=6125)
Full size table
The first component was termed the ‘100-item prudent diet score’, in line with published data (Slattery et al., 1998; Hu et al., 1999; Osler et al., 2001); women with high scores had diets in accordance with recommendations from the Department of Health (Department of Health, 1994, 1998) and other agencies. The 100-item prudent diet score had a strong and graded association with educational qualifications (r=0.47) (Figure 1a), as seen earlier (Robinson et al., 2004). In all, 56% of the women who left school with no qualifications were in the lowest quarter of the prudent diet score, compared with only 3% of women with university degrees. The correlation between the 100-item prudent diet score and women's red blood cell folate measurements was 0.28 (P<0.0001).
(a) Mean (95% confidence interval (CI)) 100-item prudent diet score by educational qualifications in the Southampton Women's Survey (SWS). (b) Mean (95% CI) 20-item prudent diet score by educational qualifications in the Nutrition and Well-being Study (NWS).
Full size image
Principal component analyses in the Southampton Women's Survey repeat study
For women in the SWS repeat study, it was possible to calculate change in 100-item prudent diet score and change in red blood cell folate measures. Among these 51 women, Spearman's correlation between change in 100-item prudent diet score and change in red blood cell folate measurements was 0.32 (P=0.02), indicating that changes in prudent diet score over time were associated with changes in measurements of this biomarker.
Development of the 20-item food frequency questionnaire
To analyse how many foods are optimal to include in a short FFQ, principal component scores based on various reduced numbers of foods were derived. Scores were derived for between 2 and 30 items and correlations calculated between these scores and the SWS 100-item prudent diet score. Figure 2 shows how the correlation increases as the number of items increases, as would be expected. However, the aim of a short FFQ is to have a substantially reduced number of foods. The 20 most influential foods (Table 2) have a correlation with the 100-item score of 0.94 and seem to be a pragmatic choice when comparing number of items with association (Figure 2). As it is possible to have a high correlation without high agreement, Bland and Altman limits of agreement were calculated; there was no average difference in the scores, with 95% limits of agreement (from −0.7 to 0.7 s.d.).
The correlation between Southampton Women's Survey (SWS) 100-item prudent diet score and score with reduced number of food variables.
Full size image
Like the 100-item prudent diet score, the 20-item prudent diet score was correlated with red blood cell folate levels at the initial time point (r=0.25, P<0.0001). Similarly, the change in 20-item prudent diet score was associated with change in red blood cell folate levels (rS=0.31, P=0.03).
Calculating the 20-item prudent diet score in the nutrition and well-being study
The 20-item FFQ was used in the NWS. The mean (s.d.) prudent diet score based on the 20 foods was 0.00 (0.99) in the SWS and −0.01 (1.14) in the NWS. Thus, the variability in scores in the NWS was as great as that within the SWS and there was no difference in mean scores between the two studies (P=0.85). The association with educational qualifications in the NWS was very similar to that in the SWS (r=0.41) (Figure 1b).
Using PCA, we have found that prudent diet scores calculated using the 100-item SWS FFQ are correlated with red blood cell folate status. A 20-item prudent diet score, based on the 20 most influential foods for this pattern, was highly correlated with the full 100-item score and was similarly correlated with red blood cell folate. Furthermore, among a subset of women (Borland et al., 2008), the change in prudent diet score over 2 years was correlated with change in red cell folate for both the 20- and 100-item scores. A 20-item FFQ can capture very similar information regarding the prudent diet score, but has clear advantages of being simpler to administer and requiring fewer resources.
The SWS had a good response rate (75%) and participants have been shown to be broadly representative of young women in England and Wales, although those from ethnic minorities are somewhat under-represented (Inskip et al., 2006). The FFQ was interviewer-administered meaning that there was little missing information, a particular advantage for principal component analyses in which complete dietary data are required. There is concern that FFQs may be subject to bias (Byers, 2004). However, in the context of dietary patterns analysis, Hu et al. (1999) showed that an FFQ revealed similar patterns of diet as weighed diet records and that individuals' scores on both were strongly positively correlated. In a Southampton cohort, we have also shown that prudent diet scores calculated from FFQ and diary data are highly correlated (Crozier et al., 2008).
A reduced-item score is able to account for a large proportion of the variability in diet because the food consumption frequencies are correlated. In the SWS, for example, mushroom consumption, which is not included in the 20-item score, is correlated with pepper consumption (rS=0.31), which is included. Byers et al. (1985) suggest that between 15 and 20 items may be all that is required for assessment of a single nutrient for epidemiological purposes.
Principal component analysis has earlier been performed on 49 foods and food groups formed by grouping the 100 foods in the SWS data (Robinson et al., 2004; Crozier et al., 2006). The 100-item and 49-item prudent diet scores are strongly correlated (r=0.97) (Crozier et al., 2006). As we aim at producing a simple short FFQ, we based our investigations on the 100-item prudent diet score.
We have shown how using the most influential foods from a 100-item prudent diet score in the SWS we were able to generate a 20-item score with necessarily reduced information, but able to capture a large amount of the variability in the 100-item prudent diet score, with which it was highly correlated (r=0.94) and had strong agreement (Bland–Altman 95% limits of agreement=0.0 (−0.7, 0.7) s.d.). The correlation of our 20-item score with the full score compares well with that found by Schulze et al. (2003) (r>0.95), although these results are not directly comparable because different numbers of foods were chosen for the reduced-item score. In contrast to Mishra et al. (2006), we used the standardized food frequencies, rather than binary intake data. When using standardized food frequencies it is important to standardize the foods at a second time point or in an alternative study using the mean and s.d. from the initial foods (as shown in Table 2), because otherwise it is not possible to assess change or differences (if foods are standardized to an internal mean of zero there is no change or difference by definition).
The SWS 20-item prudent diet score had a correlation with red cell folate measurement very similar to that for the 100-item prudent diet score (r=0.25 compared with r=0.28), indicating that the reduced-item prudent diet score has moderate associations with a biomarker that reflects variation in intake of fruit and vegetables, wholemeal bread and breakfast cereals. Earlier analyses (Borland et al., 2008) have shown that the initial and repeat prudent diet scores are strongly associated (rS=0.81), and showed a slight increase in score over a 2-year period (mean increase=0.13 s.d.). Change in red cell folate intake between these two time points is reflected by change in the 20-item prudent diet score (rS=0.31), a very similar association as when using the 100-item prudent diet score (rS=0.32).
A particular strength of this work has been the opportunity to create a short FFQ and to use this among a cohort of young women attending SureStart Children's Centres in Southampton: the NWS. As limited time was available in each NWS interview, we were keen to develop a short FFQ with which to assess diet. Women in the NWS had the same average age as those in the SWS (28 years), and both studies included women with a wide range of educational achievements. The scores in the NWS showed the same variation as those in the SWS with the same average score. Furthermore, the 20-item prudent diet score within the NWS had a similar correlation with educational qualifications as the 100-item prudent diet score in the SWS (r=0.41 compared with r=0.47), indicating a consistent association with an important predictor of dietary quality (Robinson et al., 2004).
The prudent diet pattern has been found to describe a robust axis of variation in a range of studies (Kant, 2004) and in a variety of analyses within the SWS (Crozier et al., 2006). Among young women in Southampton a 20-item prudent diet score is strongly associated with the initial 100-item prudent diet score, and with red blood cell folate measures. There is evidence from the NWS that a 20-item FFQ based on this score might be a helpful tool for use among young women in Southampton where time available for ascertainment of diet is limited. However, it is unlikely that it would be appropriate for researchers in other settings to use a 20-item FFQ based on the foods in Table 2. Differences in food consumption patterns among men, participants of different ages and ethnicity, or those from different geographical areas mean that the 20 foods in Table 2 may not correlate as well with a full prudent diet score. In order to develop a short FFQ to assess a prudent diet score, the researcher should ideally apply the techniques described in this paper to an existing data set in order to generate a population-specific short FFQ.
The 20-item FFQ we have developed has clear advantages for use in large surveys in terms of the time and resources required for completion and analysis. Our data from young women in Southampton show that this short FFQ can be used to provide useful information regarding variation in their compliance with the prudent dietary pattern.
Conflict of interest
The authors declare no conflict of interest.
Barker M, Lawrence W, Crozier S, Robinson S, Baird J, Margetts B et al., the Food Choice Group (2009). Educational attainment, perceived control and the quality of women's diets. Appetite 52, 631–636.
Bland JM, Altman DG (1986). Statistical methods for assessing agreement between two methods of clinical measurement. Lancet 1, 307–310.
Borland SE, Robinson SM, Crozier SR, Inskip HM, the Southampton Women's Survey Study Group (2008). Stability of dietary patterns in young women over a 2-year period. Eur J Clin Nutr 62, 119–126.
Byers T (2004). Food frequency dietary assessment: how bad is good enough? Am J Epidemiol 154, 1087–1088.
Byers T, Marshall J, Fiedler R, Zielenzny M, Graham S (1985). Assessing nutrient intake with an abbreviate dietary interview. Am J Epidemiol 122, 41–49.
Crozier SR, Inskip HM, Godfrey KM, Robinson SM (2008). Dietary patterns in pregnant women: a comparison of food-frequency questionnaires and 4 d prospective diaries. Br J Nutr 99, 869–875.
Crozier SR, Robinson SM, Borland SE, Inskip HM, The Southampton Women's Survey Study Group (2006). Dietary patterns in the Southampton Women's Survey. Eur J Clin Nutr 60, 1391–1399.
Department of Health (1994). Committee on Medical Aspects of Food Policy. Nutritional Aspects of Cardiovascular Disease. HMSO: London.
Department of Health (1998). Committee on Medical Aspects of Food Policy. Nutritional Aspects of the Development of Cancer. The Stationery Office: London.
Hu FB, Rimm E, Smith-Warner SA, Feskanich D, Stampfer MJ, Ascherio A et al. (1999). Reproducibility and validity of dietary patterns assessed with a food-frequency questionnaire. Am J Clin Nutr 69, 243–249.
Inskip HM, Godfrey KM, Robinson SM, Law CM, Barker DJP, Cooper C, The Southampton Women's Survey Study Group (2006). Cohort profile: The Southampton Women's Survey. Int J Epidemiol 35, 42–48.
Joliffe IT, Morgan BJT (1992). Principal component analysis and exploratory factor analysis. Stat Methods Med Res 1, 69–95.
Kant AK (2004). Dietary patterns and health outcomes. J Am Diet Assoc 104, 615–635.
Mishra GD, McNaughton SA, Bramwell GD, Wadsworth MEJ (2006). Longitudinal changes in dietary patterns during adult life. Br J Nutr 96, 735–744.
Northstone K, Emmett P (2008). A comparison of methods to assess changes in dietary patterns from pregnancy to 4 years post-partum obtained using principal components analysis. Br J Nutr 99, 1099–1106.
Osler M, Heitmann BL, Gerdes LU, Jørgensen LM, Schroll M (2001). Dietary patterns and mortality in Danish men and women: a prospective observational study. Br J Nutr 85, 219–225.
Robinson S, Godfrey K, Osmond C, Cox V, Barker D (1996). Evaluation of a food frequency questionnaire used to assess nutrient intakes in pregnant women. Eur J Clin Nutr 50, 302–308.
Robinson SM, Crozier SR, Borland SE, Hammond J, Barker DJP, Inskip HM (2004). Impact of educational attainment on the quality of young women's diets. Eur J Clin Nutr 58, 1174–1180.
Rohrmann S, Klein G (2003). Development and validation of a short food list to assess the intake of total fat, saturated, mono-unsaturated, polyunsaturated fatty acids and cholesterol. Eur J Public Health 13, 262–268.
Schulze MB, Hoffman K, Kroke A, Boeing H (2003). An approach to construct simplified measures of dietary patterns from exploratory factor analysis. Br J Nutr 89, 409–418.
Severo M, Lopes C, Lucas R, Barros H (2009). Development of a tool for the assessment of calcium and vitamin D intakes in clinical settings. Osteoporosis Int 20, 231–237.
Slattery ML, Boucher KM, Caan BJ, Potter JD, Ma KN (1998). Eating patterns and risk of colon cancer. Am J Epidemiol 148, 4–15.
StataCorp (2007). Stata Statistical Software: Release 10. StataCorp: College Station, TX.
Steijns LSW, Braams-Wiatrowska JK, Luiting H-J, van der Weide J (1996). Evaluation of nonisotopic binding assays for measuring vitamin B12 and folate in serum. Clin Chim Acta 248, 135–141.
We are grateful to the women of Southampton who took part in these studies and the research nurses and other staff who collected and processed the data. The SWS was funded by the Medical Research Council, the University of Southampton and the Dunhill Medical Trust. This NWB was funded by an award from Danone Institute International.
AffiliationsMRC Epidemiology Resource Centre, University of Southampton, Southampton General Hospital, Southampton, UKS R Crozier, H M Inskip, M E Barker, W T Lawrence, C Cooper & S M Robinson
AuthorsSearch for S R Crozier in:Nature Research journals • PubMed • Google ScholarSearch for H M Inskip in:Nature Research journals • PubMed • Google ScholarSearch for M E Barker in:Nature Research journals • PubMed • Google ScholarSearch for W T Lawrence in:Nature Research journals • PubMed • Google ScholarSearch for C Cooper in:Nature Research journals • PubMed • Google ScholarSearch for S M Robinson in:Nature Research journals • PubMed • Google Scholar
Correspondence to S R Crozier.
Rights and permissions
To obtain permission to re-use content from this article visit RightsLink.
About this article
07 April 2009
07 August 2009
10 August 2009
16 September 2009