  |
|
|
||||
|
||||||
| Journal of the New Zealand Medical Association,
24-September-2004, Vol 117 No 1202
|
||||||
|
Under-reporting of energy intake in the 1997 National
Nutrition Survey
Catherine Pikholz, Boyd Swinburn, Patricia Metcalf
Under-reporting of total energy intake (hereafter referred
to as ‘under-reporting’) is one of several potential sources of
measurement error in all types of dietary surveys, and is a common and
acknowledged problem. However, the extent of under-reporting in the 1997
National Nutrition Survey (NNS97)1 and ethnic
differences between European, Maori, and Pacific Islanders have not been
studied.
The gold standard method for assessing the validity of
reported total energy intake is through doubly-labelled water studies which can
accurately assess total energy expenditure. However, because of the high cost of
these studies, under-reporting of energy intake is most commonly measured by
comparing reported energy intake with an individual’s estimated basal
metabolic rate (BMR) or resting metabolic rate (RMR).
There is a strong positive relationship between BMR, RMR, or
total energy expenditure (all of which are very accurately measured) and weight
or body mass index (BMI). In other words, the higher the body weight, the more
energy is used to maintain that weight.2
However, the results of many dietary studies show either no relationship, or a
negative relationship, between self-reported energy intake and weight or
BMI.3,4 Since energy intake equals energy
expenditure (at weight maintenance), there must be significant under-reporting
of energy intake by people in the higher weight and BMI range.
Goldberg et al have suggested that while it may not be
possible to improve on the quality of food intake data in dietary studies, what
is important is that the possibility of bias (including bias due to
under-reporting) is acknowledged and quantified, and that the data are examined
and interpreted with this in mind.3,4 The aims
of this study were to estimate the levels of under-reporting by gender, age,
ethnicity, and body size in the NNS97 database.
MethodsThe NNS97 survey was conducted
by the University of Otago, using the primary methodology of multiple-pass
24-hour diet recall (24-HDR).1,5 For this
analysis, data from individual participants were excluded if key variables such
as height and weight were missing. Asians were excluded due to their small
numbers. After these exclusions, a total of 4,258 participants (1,808 men and
2,450 women) aged 15 years and over remained for the analysis.
Body mass index (BMI) was calculated as weight (kg)
divided by height squared
(m2).
A body size variable was created by grouping BMI into three categories: normal
weight, overweight, and obese. The BMI ranges used were those suggested for New
Zealanders of different ethnicities 6 as
follows: for Europeans: normal weight: BMI<25
kg/m2; overweight: 25
kg/m2≤BMI<30
kg/m2; obese:
BMI>30kg/m2;
for Maori and Pacific people: normal weight: BMI<26
kg/m2; overweight: 26
kg/m2≤BMI<32
kg/m2; obese:
BMI>32
kg/m2. Ethnicity was self
identified.5
Resting metabolic rate
(RMRest) was estimated using several steps. Fat
mass (FM, in kg) was calculated from BMI, using equations from Swinburn et
al.7 These equations were different for New
Zealand European, Maori, and Samoan males and females (Samoan equations were
used for the whole Pacific ethnic group). Fat free mass (FFM, in kg) was
calculated by subtracting fat mass from weight. Finally,
RMRest was calculated using an equation from
Bogardus et al 8 as follows:
RMRest (kilocalories per day) = (22.8 x FFM) +
489. RMRest in kilocalories per day was converted
to kilojoules per day by multiplying by the standard conversion factor of
4.184.
The ratio between energy intake (EI) and
RMRest (EI:
RMRest) was calculated by dividing EI by
RMRest.4 Cut-off
limits for identifying under-reporting were taken from the work done by Goldberg
et al 3 where they used basal metabolic rate
(BMR), which is virtually identical to RMR. Cut-off values for evaluating energy
intake using the ratio EI:BMR vary according to the sample size and the number
of days of diet intake records. The 95th
percentile lower cut-off values for EI:BMR based on 1 day of intake (as data
were from a 24-HDR) were used to define ‘definite’ under-reporting
in individuals and in population subgroups. Cut-off values for EI: BMR based on
1 day of intake ranged from 0.9 for one person, to 1.53 for a group of 2,000
people.3 The 0.9 cut-off value was used to
classify individuals, so that participants with an EI:
RMRest <0.9 were considered
‘definite’ under-reporters. The group with an
EI:RMRest
>0.9 clearly contains
a mixture of adequate reporters, under-reporters and over-reporters.
Because of unequal selection probabilities for
participants, all statistical analyses took into account the sampling weights
associated with the design of the study. Weighted means and standard errors of
the mean (SEM) were calculated either unadjusted or after adjusting for
potential confounders, using the statistical package STATA (StatCorp. 2001 Stata
Statistical Software: Release 7.0. College Station, TX: Stata Corporation). The
percentages of under-reporters calculated also took into account the unequal
selection probabilities.
ResultsThe baseline characteristics of the
participants, 2450 women and 1808 men, are shown in Table 1. European men and
women were 8–10 years older than the Maori and Pacific participants. The
patterns of higher mean BMI and greater prevalence rates of obesity in Maori and
Pacific people compared to European as shown here have previously been
reported.1
Mean values of EI:RMRest (and
SEM) for all participants, and for various subgroups, are presented in Table 2.
Mean EI:RMRest for all the participants was 1.40.
Overall, females had a significantly lower
EI:RMRest than men (1.30 versus 1.51, p<0.001,
adjusted for age, ethnicity, and body size).
There were no significant differences in mean
EI:RMRest between different ethnic groups in men
(adjusted for age and body size) but Maori women had a significantly higher mean
EI:RMRest compared to European and Pacific women
(p<0.01). Mean EI:RMRest decreased with age
(adjusted for ethnicity and body size). After adjustment for age and ethnicity,
obese men and women had significantly lower
EI:RMRest compared to overweight and normal
weight groups (p<0.001), and in women the
EI:RMRest of the overweight group was also lower
than that of the normal weight group (p<0.001).
The mean EI:RMRest values
from Table 2 were compared to the cut-off values from Goldberg et al, as
discussed in the methods. 3 Cut-off values for
one day of intake were used, and varied from a cut-off value of 1.47 for 104
Pacific men, to a cut-off of 1.53 for all 4258 participants. In all subgroups
shown in Table 2, the mean EI:RMRest was below
the suggested appropriate cut-off value from Goldberg et al, except for the
15–29 year and 30–39 year age groups of men. The prevalence of
‘definite’ under-reporters, defined as a cut-off value for an
individual’s EI:RMRest of
<0.93,9 was 12% in men and 21% in women.
Figure 1 shows the percentage of ‘definite’
under-reporters in the different ethnic groups. In contrast to the mean
EI:RMRest data, the ethnic differences from this
analysis were statistically significant for men (p=0.0004) but not for women.
Figure 2 shows the percentage of ‘definite’ under-reporters in the
different age groups. Differences between age groups in men, women and the total
group were statistically significant (p<0.0005). The percentage of
‘definite’ under-reporters in normal weight, overweight and obese
groups is presented in Figure 3, with significant differences found across the
groups in men and women (p<0.0001).
DiscussionThis study examined the extent of
under-reporting in the 1997 National Nutrition Survey. We found a substantial
level of under-reporting across most subgroups analysed. Overall, 12% of men and
21% of women reported energy intakes of less than 90% of their estimated resting
metabolic rate (RMRest) and were considered
‘definite’ under-reporters. In addition, under-reporting was
significantly higher in older age groups, and those classified as overweight
(women only) or obese. These patterns have been well described in other
studies.9–12
There is some evidence in the literature suggesting that
under-reporting may be more common in members of ethnic minority
groups.9–11 However, in the present
study, the results from two different analyses of under-reporting by ethnic
group gave mixed results. Using the mean
EI:RMRest data, European and Pacific women seemed
to have more under-reporting than Maori women. However using
EI:RMRest <0.9 cut-off value, under-reporting
seemed most prevalent in Pacific, lower in Maori and least prevalent in European
women. Low numbers in the Maori and Pacific groups may be contributing to this
uncertainty.
A re-examination of the distribution of
EI:RMRest by gender and ethnicity showed that
there were several very high individual values for
EI:RMRest in Maori women, which caused a very
positively skewed distribution, and this may explain the higher mean value of
EI:RMRest. The range of
EI:RMRest values was also wider in Maori women.
The median EI:RMRest for all three ethnic groups
(in men and women) were slightly lower than the mean
EI:RMRest values, but in Maori women the median
EI:RMRest was much lower than the mean.
How does the level of under-reporting in the NNS97 compare
with other surveys internationally? The NHANES III survey in the US also used
24-hour dietary recall methods. Briefel et al carried out an analysis of
under-reporting in that survey.9 Their data
analysis methods were similar to those used here, using the same cut-off value
of 0.9 for EI:BMRest (or
EI:RMRest as used here), derived from Goldberg et
al.3 Mean values of
EI:BMRest in their analysis (1.47 and 1.26 in men
and women respectively) were lower than in the NNS97 and the percentages of
‘definite’ under-reporters were correspondingly higher (18% of men
and 28% of women).
Comparisons with other large studies in the literature are
more difficult to make, as the methods of assessment of dietary intake vary and
include 7-day diet diaries, 3-day diet diaries, and food frequency
questionnaires. In a meta-analysis by Black et al, a mean value for EI:BMR of
1.43 (for men and women combined) was calculated for all the studies analysed
(all methods), while the mean value for studies using the 24-HDR method was
1.31.4
The NNS97 investigators1
felt that estimating under-reporting using the Schofield
equations13 (which use body weight rather than
fat free mass for estimating BMR) might not be appropriate for use in the New
Zealand population. As the Schofield equations were developed in a normal weight
population (up to 84 kg), but more than 25% of the NNS97 survey population had a
weight exceeding 84 kg, the equations could not be assumed to be valid in this
group. We have addressed this issue by using fat free mass to calculate RMR.
Since New Zealand equations for estimating fat mass were
available,7 fat free mass and then RMR could
therefore be estimated.
Fat free mass has been shown to have a much tighter
relationship than body weight with RMR, and is the best available determinant of
energy expenditure, explaining about 80% of the variance observed between
individuals.14–16 Several authors have
strongly supported using prediction equations for BMR which incorporate fat free
mass rather than body weight, as these would allow more accurate estimation of
BMR, especially in population groups of varying body size and
composition.17,18 As already discussed, RMR and
BMR are virtually equivalent and may be substituted for one another. Other
studies have used fat free mass in prediction equations for resting metabolic
rate (RMR) and for 24-hour energy
expenditure.8,14–16
The limitations of this analysis can be considered in two
broad groups, concerning firstly the methodology and the data collection in the
NNS97 survey itself and secondly the methods used here to analyse the data. The
NNS971 was linked to the concurrent New Zealand
Health Survey.19 Of approximately 9,000 people
who participated in the New Zealand Health Survey, and who were invited to
participate in the linked NNS97, only 4,636 completed the 24-HDR in the
NNS97—an overall response rate of only
50.1%.1 The possibility of selection bias
should therefore be borne in mind when interpreting the data.
An analysis of non-responders (people who took part in the
New Zealand Health Survey but not in the NNS97) suggested that the NNS97 sample
had similar characteristics to the New Zealand Health Survey sample.
1
The primary methodology used in the NNS97 was a 24-hour diet
recall (24-HDR).1,5 The data from the 24-HDR
have been analysed in this study. Other dietary assessment methods used in
dietary studies include retrospective questionnaires of typical diet and
prospective diet records (usually for 3 to 7 days, either weighed or quantified
in some other way). All dietary assessment methods are subject to bias, usually
towards underestimation of habitual energy intake, but the 24-HDR method tends
to give lower intakes than other methods.4
Other issues for the 24-HDR method include intra- and inter-individual
variability, day-to-day variability in food intake, weekday and weekend
variability, and seasonal variability.
The gold standard method for measuring energy expenditure is
to use the doubly-labelled water (DLW)
technique.4,20–22 The rationale for doing
a DLW validation study as part of a nutrition survey is enable assessment of the
accuracy (or level of inaccuracy) of the dietary intake data with a greater
degree of certainty than merely by calculating the ratio of
EI:BMRest or
EI:RMRest. Energy expenditure can then be
compared with self-reported energy intake, and an assessment of the degree of
under-reporting can be made. The DLW method has been used in several small New
Zealand studies 23,24 but is expensive. No DLW
validation study was performed in the NNS97; however the recently completed New
Zealand Child Nutrition Survey included a DLW study.
Since the early 1990s, various authors have recommended that
the data collected in dietary surveys should include information regarding
physical activity level (PAL), dieting and weight-consciousness
4,20,25–29 as well as DLW studies, in
order to be able to assess the validity of the survey results and the level of
under-reporting. Black has recently concluded that in order to assign the
correct Goldberg cut-off values3 to subjects in
dietary surveys, sufficient information on their level of activity is
essential.27, 28
ConclusionsThis study highlights the
difficulties of accurately measuring dietary intake through self-reporting
methodologies such as the 24-hour diet recall, and the need to acknowledge and
to attempt to measure the bias inherent in dietary assessment methods. Analyses
such as the present one estimate the level of under-reporting and identify the
subgroups with potentially greater levels of under-reporting—namely women,
older people, overweight, and obese people. For more accurate estimates of
under-reporting, validation studies using doubly-labelled water
methodologies4,20–22 in those subgroups
would be needed, as well as collection of information regarding physical
activity level and dieting.25–29
Author information:
Catherine Pikholz, Public Health Medicine Registrar; Boyd Swinburn,
Associate Professor; Patricia A. Metcalf, Senior Lecturer, University of
Auckland, Auckland
Acknowledgements: We
thank Dr Noela Wilson (University of Otago) for her insightful input to this
study. We also thank Dr Robert Scragg (University of Auckland) who provided
technical guidance in the early stages of Catherine Pikholz’s dissertation
that preceded this study.
Correspondence: Boyd
Swinburn, 221 Burwood Highway, Melbourne 3125, Australia. Fax: +61 3 9244 6017;
email: swinburn@deakin.edu.au
References:
|
|
||||
| | Current
issue | Search journal |
Archived issues | Classifieds
| Hotline (free ads) Subscribe | Contribute | Advertise | Contact Us | Copyright | Other Journals |
|