British Journal of Nutrition (2014), 112, 609–615 doi:10.

1017/S0007114514001111
q The Authors 2014. This is an Open Access article, distributed under the terms of the Creative Commons Attribution licence (http://
creativecommons.org/licenses/by/3.0/), which permits unrestricted re-use, distribution, and reproduction in any medium, provided the
original work is properly cited.

The reliability of an adolescent dietary pattern identified using

reduced-rank regression: comparison of a FFQ and 3 d food record

Geeta Appannah1,2, Gerda Karolien Pot1,3, Therese Anne O’Sullivan4,5, Wendy Hazel Oddy5,
Susan Ann Jebb1 and Gina Leslie Ambrosini1,6*
1
MRC Human Nutrition Research, Cambridge, UK
2
Department of Nutrition and Dietetics, Faculty of Medicine and Health Sciences, Universiti Putra Malaysia,
Serdang, Malaysia
3
Diabetes and Nutritional Sciences Division, School of Medicine, King’s College London, London, UK
4
School of Exercise and Health Science, Edith Cowan University, Joondalup, WA, Australia
5
Telethon Institute for Child Health Research, University of Western Australia, Perth, WA, Australia
6
School of Population Health, University of Western Australia, Perth, WA, Australia
(Submitted 17 January 2014 – Final revision received 10 April 2014 – Accepted 17 April 2014 – First published online 22 May 2014)
British Journal of Nutrition

Abstract
Despite the increasing use of dietary patterns (DP) to study diet and health outcomes, relatively few studies have examined the reliability of
DP using different dietary assessment methods. Reduced-rank regression (RRR) is an emerging statistical method that incorporates a priori
information to characterise DP related to specific outcomes of interest. The aim of the present study was to compare DP identified using the
RRR method in a FFQ with those in a 3 d food record (FR). Participants were 783 adolescents from the Western Australian Pregnancy
(Raine) Cohort Study who completed both a FFQ and FR at 14 years of age. A similar ‘energy-dense, high-fat and low-fibre’ DP was
identified in the FFQ and FR that was characterised by high intakes of processed meat and sugar-sweetened beverages, and low intakes
of vegetables and fresh fruit. Nutrient profiles for this DP were consistent in the FFQ and FR. Pearson’s correlation coefficient between
participants’ z-scores for the DP identified in the FFQ and FR was 0·35 for girls and 0·49 for boys (P, 0·05). The mean difference between
DP z-scores derived from the FFQ and FR was 2 0·08 (95 % CI 2 0·21, 0·04) for girls and 20·05 (95 % CI 20·17, 0·07) for boys. The 95 %
limits of agreement were 22·55 to 2·39 for girls and 22·52 to 2·41 for boys. These findings suggest that very similar DP may be identified
and their z-scores show modest agreement when applying the RRR method to dietary intake data collected from adolescents using a
FFQ or FR.

Key words: Dietary patterns: Reliability: Adolescents: Raine Study

Dietary pattern analysis is a useful method for studying the role in studies examining dietary patterns in relation to various
of diet in relation to health outcomes or disease risk. Dietary outcomes, including diabetes, obesity and heart disease(2 – 4)
patterns have some advantages over the analysis of single nutri- Most studies investigating dietary patterns have employed
ents or foods as they consider the total diet and the cumulative a FFQ to estimate dietary intakes, owing to their lower cost
and interactive effects of foods and nutrients eaten together, and ease of administration compared with a more detailed
hence representing a more holistic perspective(1). Empirical food diary method. However, FFQ are often criticised for
dietary patterns are identified using statistical dimension- their lack of precision, particularly when applied in child or
reduction techniques, which can identify a small number of adolescent populations(5). Therefore, it is very important to
underlying constructs, or dietary patterns, from a large assess whether the limitations of a FFQ affect their ability
number of dietary variables. One such method is reduced- to describe dietary patterns. To date, no studies have com-
rank regression (RRR), which combines a priori information pared RRR-derived dietary patterns using different dietary
with exploratory statistics to identify dietary patterns related assessment methods in children or adolescents. Therefore,
to specific outcomes of interest. This method has been applied the present study set out to assess (1) whether the same

Abbreviations: CSIRO, Commonwealth Scientific and Industrial Research Organisation; DED, dietary energy density; EI:EE, ratio of energy intake:energy
expenditure; FR, food record; LOA, limits of agreement; RRR, reduced-rank regression.

* Corresponding author: Dr G. L. Ambrosini, email [email protected]

 610 G. Appannah et al.

dietary pattern, hypothesised to be associated with the risk of from the analysis. A total of 822 FR were classified as
obesity in children and adolescents, could be identified in representative of usual eating habits and included in the anal-
both a FFQ and 3 d food record (FR), and (2) whether dietary ysis. All FR were coded by a dietitian for nutrient analysis
pattern z-scores from each dietary assessment method are in using the Australian food composition database(8). A total of
agreement. 4400 foods were recorded in the 822 FR, and each food was
allocated to one of the forty-six predefined food groups
described above. Mixed dishes were disaggregated into main
Materials and methods constituents before food group coding.
Study population Using the ‘Goldberg cut-off’ method, dietary misreporting
was calculated based on the ratio of energy intake:energy
Participants in the present study were adolescent offspring expenditure (EI:EE)(11). A 95 % CI was calculated for EI:EE
in the Western Australian Pregnancy Cohort (Raine) Study(6). to take account of the potential variation in the estimates of
In brief, 2900 pregnant women were recruited into a trial at EI and EE. Participants were categorised as under-reporters
King Edward Memorial Hospital (Perth, Western Australia) to if their EI:EE was ,95 % CI, over-reporters if their EI:EE was
examine ultrasound imaging from 1989 to 1991. A total of . 95 % CI and plausible reporters if their EI:EE was within
2868 children born to 2804 mothers remained with the study the 95 % CI. Dietary misreporting was calculated only using
and subsequently formed the Raine cohort. These children the dietary intake data estimated from the FFQ.
were followed up at regular intervals, i.e. 1, 2, 3, 5, 8, 10, 14
and 17 years of age. The present analysis used data collected
at the 14-year follow-up, when both a FFQ and FR were
British Journal of Nutrition

administered. Of the 2868 baseline sample, 2337 (82 %) adoles- Dietary pattern analysis
cents were eligible for a follow-up at 14 years of age, while 152 The partial least-squares procedure with a RRR option in
(5 %) were lost to follow-up, 348 (12 %) had withdrawn from the SAS (SAS Institute, Inc.) was applied to derive the dietary pat-
study and thirty-one (1 %) were deceased. Ethical approval for terns. In brief, the RRR method identifies linear combinations
the study was obtained from the ethics committees of King of weighted food intakes, or patterns in food intakes, that
Edward Memorial Hospital and Princess Margaret Hospital explain the maximum variation in a set of response variables,
for Children. Adolescents and their parent or guardian gave which are hypothesised to be on the pathway between food
informed written consent. intake and a health outcome of interest(12). In the present
study, the RRR model included intakes of the forty-six prede-
fined food groups (g/d) as predictor variables and intakes of
Dietary assessments
three dietary variables (dietary energy density (DED), percen-
A 227-item semi-quantitative FFQ developed by the Common- tage of energy from total fat intake and fibre density) as
wealth Scientific and Industrial Research Organisation response variables. These response variables were linked
(CSIRO) was used to estimate habitual dietary intakes over the with obesity and were of interest in relation to the risk of
previous year. Parents of adolescents completed the FFQ with obesity in the Raine Study(3,9,13). In 2003, the WHO classified
their child at the 14-year follow-up(7). For each food item, the foods with high DED and low fibre density to be important
average frequency of consumption over the past year was predictors of obesity in adults(13). Furthermore, a dietary pat-
recorded as ‘never’, ‘rarely’, ‘times a month’, ‘times a week’ or tern characterised by high DED and percentage of energy
‘times a day’. The selected frequency category for each food from total fat intake and low intakes of fibre in childhood
item was then converted to a daily intake (g) and linked by and adolescence has been shown to be prospectively associ-
the CSIRO with the Australian food composition database to ated with greater adiposity in a UK pregnancy cohort(3). DED
estimate daily nutrient intakes(8). This FFQ was evaluated; was calculated by dividing total food energy (kJ) by total
relative to a 3 d FR, the FFQ correctly ranked most individuals food weight (g) excluding beverages because they may
according to their nutrient intakes at 14 years of age(9). All disproportionately influence total DED values(14,15). Fibre
food and beverage intakes (n 227 items) in the FFQ were density was expressed as absolute intake of fibre (g/d)
assigned to forty-six predefined food groups based on nutrient divided by total daily energy intake (MJ). Percentage of
profiles or culinary usage, and their hypothesised contribution energy from total fat intake was calculated by dividing total
to diet – disease relationships(10). A total of 1631 adolescents energy intake from fat (kJ) by total energy intake (kJ) and
completed the FFQ at the 14-year follow-up. Of these, 1611 then multiplying by 100.
reported plausible energy intakes (. 3000 and ,20 000 kJ/d) Separate RRR analyses were applied to the FFQ and FR data.
and were included in the analysis. Each study participant received a z-score for the dietary pattern
Participants were also requested to complete a 3 d FR with identified in the FFQ and FR, discriminating how strongly
parental assistance at the 14-year follow-up. Participants their dietary intakes corresponded with the dietary patterns.
were not required to complete their FR on a certain number These z-scores were then categorised into quartiles to enable
of week days or weekend days. Of the 1286 adolescents comparisons between the two dietary assessment methods.
who agreed, 962 returned a completed FR. Where any of Separate RRR analyses were conducted for boys and girls,
the 3 d recorded was noted by the respondent as not typical although their dietary patterns were similar, and therefore
or unrepresentative of their usual diet, the FR was excluded dietary patterns for the whole sample are reported here.
 Reliability of an adolescent dietary pattern 611

Statistical methods intake, but negatively correlated with fibre density. This
dietary pattern explained the most variation in all response
Dietary patterns identified in the FFQ and FR were compared
variables in the FFQ and FR (53 and 46 %, respectively). The
using three methods. First, the product of the dietary patterns,
subsequent two dietary patterns explained little additional vari-
their nutrient profiles, was compared. Pearson’s correlation
ation; 15 and 12 % for the second dietary pattern and only 7 and
coefficient was used to compare nutrient intakes estimated
8 % for the third dietary pattern, in the FFQ and FR, respectively.
from the FR (as reference) with dietary pattern z-scores
These two dietary patterns were not as easily interpretable and
derived from the FFQ and FR. The nutrient intakes estimated
were not hypothesised to be associated with the risk of obesity.
from the FR were adjusted for total energy intake (kJ) using
Therefore, only the first dietary pattern was taken forward for
the residual method(16). Key nutrients of interest included
further analysis.
carbohydrate, protein, total fat, fibre, sugars, saturated fat,
Fig. 1 shows the factor loadings for the ‘energy-dense, high-
monounsaturated fat, polyunsaturated fat, cholesterol, Na,
fat and low-fibre’ dietary pattern identified using the FFQ and
Ca, K, Mg, folate, Fe, thiamin, niacin, riboflavin, Zn, retinol,
FR. Intakes of foods with a positive factor loading increased
vitamin A and vitamin C.
the dietary pattern z-score, while intakes of foods with a
Second, study participants’ z-scores for the dietary pattern
negative factor loading decreased the z-score. The ‘energy-
identified in the FFQ and FR were compared using the partial
dense, high-fat and low-fibre’ dietary pattern identified in
Pearson’s correlation coefficient, adjusted for dietary
the FFQ was strongly characterised by high intakes of pro-
misreporting. Although correlations have been widely used in
cessed meat, chocolate and confectionery, low-fibre bread,
many studies, they only measure the strength of a relationship
sugar-sweetened beverages, full-fat milk, and crisps and
between two variables and not the agreement between
British Journal of Nutrition

savoury snacks (Fig. 1). Similarly, the ‘energy-dense, high-

them(17). To examine the exact agreement between dietary
fat, low-fibre’ dietary pattern identified in the FR was also
pattern z-scores derived from the FFQ and FR, Bland – Altman
strongly characterised by high intakes of processed meat,
plots were used(17). The plots showed the difference between
sugar-sweetened beverages, full-fat milk, and crisps and
each individual’s z-scores derived from FFQ and FR against
savoury snacks, in addition to fried/roasted potatoes (chips)
their averages(17). The mean difference and 95 % limits of
and coated/breaded meat and fish. In both the FFQ and FR,
agreement (LOA, calculated as mean differences and ^2 SD )
the dietary pattern was strongly negatively associated with
were used to summarise agreement at the population level.
intakes of vegetables, fresh fruit, high-fibre bread, legumes
For a normally distributed variable, the LOA describes the
and high-fibre breakfast cereals. Notably, factor loadings for
range containing 95 % of all individual’s differences between
fresh fruit and vegetables were the strongest of all the food
their FFQ and FR z-scores; the wider the LOA, the weaker the
groups (almost double the loading for processed meat) in
agreement(17). The Bland – Altman plots also illustrate whether
both the FFQ and FR (Fig. 1).
agreement between the FFQ and FR varies with the magnitude
Factor loadings in the FR were generally lower than those in
of z-scores by showing the fitted regression line (slope)
the FFQ. Some notable differences in factor loadings were
between differences and averages.
observed for rice, pasta and other grains, high-energy-dense
sauces and low-fibre bread (Fig. 1). However, despite the
Results different factor loadings observed for some food groups, the
majority were similar in their ranking and associations (Fig. 1).
Characteristics of the study population
A total of 783 adolescents completed both the FFQ and 3 d FR
Nutrient profiles
at 14 years of age. Participation was similar for girls (49 %) and
boys (51 %). Of 783 adolescents, the proportion of dietary The participants’ z-scores for the ‘energy-dense, high-fat and
under-reporters, plausible reporters and over-reporters was low-fibre’ dietary pattern identified in both the FFQ and FR
36, 57 and 7 %, respectively. Adolescents who were were positively correlated with intakes of saturated fat, total
overweight or obese were more likely to be dietary under- fat, monounsaturated fat, retinol, cholesterol and negatively
reporters compared with other misreporting categories (57 v. correlated with carbohydrate, protein, fibre, Mg, K, folate,
43 %). Other characteristics of cohort members who vitamin C, Fe, thiamin, vitamin A, Ca, niacin, riboflavin and
completed both a FFQ and FR at the 14-year follow-up were Zn, as estimated from the FR (Fig. 2). As expected, correlations
previously compared(9). In summary, those included in the were weaker between the FFQ and FR; however, the
present analysis had higher levels of maternal education, directions of associations were consistent.
and were slightly less likely to be overweight or from
low-income families(9).
Correlations and agreement between
dietary pattern z-scores
Dietary patterns identified in the FFQ and 3 d food record
Pearson’s correlation coefficient between the dietary
As three response variables were included in the RRR pattern z-scores derived from the FFQ and FR was 0·35 for
analyses, three dietary patterns were identified. The first girls and 0·49 for boys (P,0·05) after adjusting for dietary
dietary pattern identified in the FFQ and FR was positively misreporting. Moderate agreement was observed between
correlated with DED and percentage of energy from total fat the dietary pattern z-scores derived from the FFQ and FR
 612 G. Appannah et al.

Processed meat
Fried or roast potatoes
Sugar-sweetened beverages
Coated and breaded meat and fish
High-fat milk and cream
Chocolate and confectionery
Crisps and savoury snacks
Pizza
Biscuits and cakes
Low-fibre bread
Ice creams
Sauces (high energy dense)
Other fruit
Low-fat milk
Sauces (low energy dense)
Yogurts
Boiled or baked potatoes
Soups
British Journal of Nutrition

Water
High-fibre breakfast cereals
Legumes
High-fibre bread
Rice, pasta and other grains
Vegetables
Fresh fruit

–0·6 –0·4 –0·2 0 0·2 0·4

Factor loading
Fig. 1. Factor loadings for an ‘energy-dense, high-fat and low-fibre’ dietary pattern identified using the FFQ ( ) and 3 d food record ( ) (Raine Study). Food
groups with very small factor loadings (, j0·10j) were excluded from the graph for brevity. These included butter and animal fats, margarine and vegetable oils,
eggs and egg dishes, other bread products, other breakfast cereals, cereal-based mixed meals, puddings, spreads, meat and poultry, mixed meat dishes, fish,
meat substitutes, fried vegetables, mixed vegetable dishes, nuts and seeds, low-energy beverages, fruit juice, hot and powdered drinks.

with a non-significant mean difference of 2 0·08 (95 % CI and 0·38 for girls, respectively(18). These correlation coeffi-
2 0·21, 0·04) for girls and 20·05 (95 % CI 2 0·17, 0·07) for cients compare well with those in the present study despite
boys (Fig. 3). The 95 % LOA were similar for girls (22·55, the fact that different statistical methods were used to identify
2·39) and boys (2 2·52, 2·41) (Fig. 3). However, the significant the dietary patterns. In adults, correlations between principal
slope between the averages and differences in dietary pattern components analysis-derived dietary pattern z-scores ident-
z-scores (r 0·49 for boys, r 0·35 for girls) indicated that agree- ified using a FFQ and 1-week diet records ranged from 0·34
ment between the FFQ and 3 d FR decreased as the dietary to 0·73(19,20). Similarly, dietary patterns identified using princi-
pattern z-scores increased. pal components analysis among pregnant women in the UK
also suggested correlation coefficients to be ranging between
0·35 and 0·67 for the dietary patterns derived from a FFQ and
Discussion 4 d food diaries(21). Furthermore, differences in factor loadings
In the present study of a large sample of adolescents, a similar for a few food groups have also been reported by some of
‘energy-dense, high-fat and low-fibre’ dietary pattern was these studies(5,18,19).
identified using the RRR method in a FFQ and FR, and there Although the key foods and their factor loadings for the
was modest agreement between the dietary pattern z-scores. ‘energy-dense, high-fat and low-fibre’ dietary pattern were
No published studies have previously reported comparisons similar for the FFQ and FR, some differences were noted, in
between RRR-derived dietary patterns in a FFQ and FR in particular, for high energy-dense sauces, low-fibre bread, rice,
adolescents. However, exploratory principal components pasta and other grains. Some variations might be expected
analysis was previously applied in the Raine Study to identify given the differences in dietary assessment methods used in
‘Healthy’ and ‘Western’ dietary patterns in the FFQ, which the present study. The FFQ was designed to capture habitual
were compared with those in the FR at 14 years of age(18). dietary intakes for the past 12 months, whereas the FR only
The correlations for the ‘Healthy’ and ‘Western’ dietary pat- captured foods eaten (and recorded) over a 3 d period. There-
terns were observed to be 0·47 and 0·34 for boys and 0·42 fore, a FR may not capture food items consumed infrequently.
 Reliability of an adolescent dietary pattern 613

Saturated fat*
Total fat*
Monounsaturated fat*
Retinol*
Cholesterol
Na
Sugars
Zn
Polyunsaturated fat
Protein
Riboflavin*
CHO*
Niacin*
Ca*
Vitamin A*
Thiamin*
British Journal of Nutrition

Fe*
Vitamin C*
Folate*
K*
Mg*
Fibre*

–1 –0·8 –0·6 –0·4 –0·2 0 0·2 0·4 0·6

Correlations
Fig. 2. Correlations between nutrient intakes estimated from the 3 d food record (FR, ) and dietary pattern z-scores derived from the FFQ ( ) and 3 d FR (Raine
Study). Nutrient intakes were estimated from 3 d FR and adjusted for total energy intake using the residual method(14). * P, 0·05 for the FFQ. CHO, carbohydrate.

Furthermore, variations in the disaggregation of mixed dishes in the FR were generally weaker than those observed in the FFQ at
the FR may have contributed to the differences in factor loadings 14 years of age(18). Similar findings were also observed among
for the ‘rice and pasta’ food group. With the use of principal com- American men in a study by Hu et al.(19). While the RRR
ponents analysis-derived dietary patterns, an earlier study of this method incorporates a priori information in the form of response
cohort has also reported that factor loadings for all food groups in variables, it is also partly an exploratory statistical method that

(a) 7 (b) 7
6 6
5 5
Difference between FFQ and 3 d

4 4
+2.41 +2.39
3 3
FR DP z-scores

2 2
1 1
0 0
–1 –1
–2 –2 –2.55
–3 –2.52 –3
–4 –4
–5 –5
–6 –6
–4 –2 0 2 4 –4 –2 0 2 4

Average of FFQ and 3 d FR DP z-scores

Fig. 3. Bland – Altman plots of dietary pattern (DP) z-scores derived from the FFQ and 3 d food record (FR) for (a) boys and (b) girls (Raine Study). Represents
95 % limits of agreement between DP z-scores derived from the FFQ and 3 d FR. A regression line (slope) was fitted by the regression of differences in DP z-scores
against their averages: (a) r 0·49, P, 0·01; (b) r 0·35, P, 0·01. (A colour version of this figure can be found online at http://www.journals.cambridge.org/bjn).
 614 G. Appannah et al.

depends on the sample correlation matrix. Hence, some The authors also acknowledge the National Health and Medi-
variations would be expected in factor loadings regardless of cal Research Council of Australia for long-term funding contri-
the differences in dietary assessment methods. bution to the Raine Study over the last 20 years, the Telethon
The correlations between the dietary patterns derived from Institute for Child Health Research for long-term support of
the FFQ and FR were observed to be higher in boys than in the study and the CSIRO for the use of the FFQ.
girls. These results were not unforeseen as studies among ado- The present study was supported by a programme grant
lescents have shown greater inconsistency in the recording of from the Medical Research Council (grant no. U105960389).
food intakes among adolescent girls than among boys(5,18). G. A. was supported by a PhD studentship from the Ministry
However, there may be differences between the two variables of Higher Education, Malaysia and Universiti Putra Malaysia.
across their range of values even if these variables are highly Management funding for the Raine Study was provided by
correlated. In contrast, mean agreement and 95 % LOA are the University of Western Australia, the Telethon Institute for
better indicators of exact agreement between two different Child Health Research, the Raine Medical Research Foun-
dietary assessment methods(17,18). In the present study, the dation, the Faculty of Medicine, Dentistry and Health Sciences
95 % LOA between the dietary pattern z-scores derived from of the University of Western Australia, the Women’s and
the FFQ and FR were comparable with the earlier analysis in Infants Research Foundation and Curtin University. Research
the Raine Study in which minor differences were observed funding was provided by the National Health and Medical
using the 95 % LOA between boys and girls in ‘Healthy’ and Research Council of Australia, the National Heart Foundation
‘Western’ dietary patterns identified from the FFQ and FR(18). of Australia, and Beyond Blue Cardiovascular Disease and
There are some limitations in the present study. First, the Depression Strategic Research Program.
British Journal of Nutrition

analysis was confined to those adolescents who completed The authors’ contributions were as follows: G. A. was
both the FFQ and FR (n 783). The observed differences responsible for FFQ food group coding, data analyses and
between these participants and the remainder of the cohort the first draft of the manuscript; G. K. P. assisted with data
may limit the generalisability of these findings. Nonetheless, interpretation and manuscript writing; T. A. O. handled and
the same ‘energy-dense, high-fat and low-fibre’ dietary pattern coded the 3 d FR data and contributed to data interpretation
was identified in the whole cohort (n 1611) using the FFQ at and manuscript preparation; W. H. O. was a principal investi-
14 years of age. Furthermore, while FFQ are often criticised for gator for the collection of dietary data from which the data for
their measurement error, it is important to note that a FR is not the present study were taken, and assisted with manuscript
error-free. For example, they may be prone to dietary under- preparation; S. A. J. provided critical review of the manuscript;
reporting and may not reflect habitual dietary intake, as foods G. L. A. provided significant advice on the data analysis, data
eaten infrequently may not be captured(5). We have, however, interpretation and the writing of the manuscript. All authors
taken steps to minimise the impact of dietary under-reporting read and approved the final manuscript.
and unrepresentative diary recording days. The authors have no conflicts of interest to declare.
The strengths of the present study include a sample size which
is larger than that included in most reliability studies and good
response rates, with 70 and 75 % of adolescents who were eligible
for follow-up, completing a FFQ and FR, respectively, at 14 years References
of age. Adolescents tend to show a greater variation in their 1. Schulze MB & Hoffman K (2006) Methodological approaches
dietary intakes than younger children and adults; therefore, to study dietary patterns in relation to risk of coronary heart
assessing their usual food intakes is challenging(5). However, disease and stroke. Br J Nutr 95, 860–869.
we attempted to control for dietary misreporting, and this was 2. Nettleton JA, Steffen LM, Schulze MB, et al. (2007) Associ-
ations between markers of subclinical atherosclerosis and
considered as a potential confounder in the present analysis.
dietary patterns derived by principal components analysis
Additionally, the FFQ has been evaluated in this cohort and and reduced rank regression in the Multi-Ethnic Study of
been shown to correctly rank a reasonable proportion of nutrient Atherosclerosis (MESA). Am J Clin Nutr 85, 1615 –1625.
intakes relative to this FR at 14 years of age(9). 3. Ambrosini GL, Emmett PM, Northstone K, et al. (2012)
Identification of a dietary pattern prospectively associated
with increased adiposity during childhood and adolescence.
Conclusions Int J Obes 36, 1299– 1305.
4. Liese AD, Schulz M, Moore CG, et al. (2004) Dietary patterns,
In this adolescent cohort, a comparable ‘energy-dense, high-
insulin sensitivity and adiposity in the multi-ethnic Insulin
fat and low-fibre’ dietary pattern was identified using the Resistance Atherosclerosis Study population. Br J Nutr 92,
RRR method in a FFQ and FR. The present study supports 973– 984.
the use of a FFQ to study the relationships between a RRR- 5. Livingstone MBE, Robson PJ & Wallace JMW (2004) Issues in
derived dietary pattern and health outcomes in Australian dietary intake assessment of children and adolescents. Br J
adolescents. Nutr 92, Suppl. 2, S213 –S222.
6. Newnham JP, Evans SF, Michael CA, et al. (1993) Effects
of frequent ultrasound during pregnancy: a randomised
Acknowledgements controlled trial. Lancet 342, 887– 891.
7. Baghurst K & Record S (1984) A computerised dietary
The authors are extremely grateful to all the Raine Study analysis system for use with diaries or food frequency ques-
participants and their families and the Raine Study team. tionnaires. Community Health Stud 8, 11 –18.
 Reliability of an adolescent dietary pattern 615

8. Food Standards Australia New Zealand. NUTTAB (2006) 15. Johnson L, Wilks DC, Lindroos AK, et al. (2009) Reflections
Australian Food Composition Database. http://www.food- from a systematic review of dietary energy density and
standards.gov.au/science/monitoringnutrients/nutrientables/ weight gain: is the inclusion of drinks valid? Obes Rev 10,
Pages/default.aspx (accessed February 2009). 681– 692.
9. Ambrosini GL, de Klerk NH, O’Sullivan TA, et al. (2009) The 16. Willett W & Stampfer MJ (1986) Total energy intake: impli-
reliability of a food frequency questionnaire for use among cations for epidemiologic analyses. Am J Epidemiol 124,
adolescents. Eur J Clin Nutr 63, 1251 –1259. 17– 27.
10. Johnson L, Mander AP, Jones LR, et al. (2008) Energy-dense, 17. Bland J & Altman DG (1986) Statistical methods for assessing
low fiber, high-fat dietary pattern is associated with agreement between two methods of clinical measurement.
increased fatness in childhood. Am J Clin Nutr 87, 846– 854. Lancet 1, 307–310.
11. Goldberg GR, Black AE, Jebb SA, et al. (1991) Critical evalu- 18. Ambrosini GL, O’Sullivan TA, de Klerk NH, et al. (2011) Rela-
ation of energy intake data using fundamental principles of tive validity of adolescent dietary patterns: a comparison of a
energy physiology: 1. Derivation of cut-off limits to identify FFQ and 3 d food record. Br J Nutr 105, 625– 633.
under-recording. Eur J Clin Nutr 45, 569 – 581. 19. Hu FB, Rimm E, Smith-Warner SA, et al. (1999) Reproduci-
12. Hoffman K, Schulze MB, Schienkiewitz A, et al. (2004) bility and validity of dietary patterns assessed with a food-
Application of a new statistical method to derive dietary pat- frequency questionnaire. Am J Clin Nutr 69, 243–249.
terns in nutritional epidemiology. Am J Epidemiol 159, 20. Khani BR, Ye W, Terry P, et al. (2004) Reproducibility and
935–944. validity of major dietary patterns among Swedish women
13. WHO (2003) Diet, Nutrition and the Prevention of Chronic assessed with a food-frequency questionnaire. J Nutr 134,
Diseases. Technical Report Series no. 916. Geneva: WHO. 1541 –1545.
14. Ledikwe JH, Blanck HM, Khan LK, et al. (2005) Dietary 21. Crozier SR, Inskip HM, Godfrey KM, et al. (2008) Dietary
energy density determined by eight calculation methods in patterns in pregnant women: a comparison of food-
British Journal of Nutrition

a nationally representative United States population. J Nutr frequency questionnaires and 4 d prospective diaries. Br J
135, 273– 278. Nutr 99, 869– 875.