IJE Advance Access originally published online on August 27, 2004
International Journal of Epidemiology 2004 33(6):1382-1386; doi:10.1093/ije/dyh181
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
IJE vol.33 no.6 © International Epidemiological Association 2004; all rights reserved.
Article |
Adjusting for energy intake—what measure to use in nutritional epidemiological studies?
1 MRC Epidemiology Unit, Strangeways Research Laboratory, Cambridge, CB1 8RN, UK
2 Department of Public Health and Primary Care, University of Cambridge, Strangeways Research Laboratory, Cambridge, CB1 8RN, UK
3 MRC Dunn Human Nutrition Unit, Cambridge, CB2 2XY, UK
Correspondence: Prof. NE Day, Strangeways Research Laboratory, Institute of Public Health, University of Cambridge, CB1 8RN, UK. E-mail: nick.day{at}srl.cam.ac.uk
 |
Abstract |
|---|
 Top Abstract MethodsResultsDiscussionReferences |
|---|
Background The measurement of energy intake in epidemiological studies is difficult. However, it is important that energy intake is assessed if epidemiological analyses are to correspond to isocaloric experiments. The aim of this study was to compare self-reported energy intake, physical activity, and body weight with energy expenditure measured by 4 days of heart rate monitoring with individual calibration of the relationship between heart rate and oxygen consumption.
Methods Volunteer sub-study of 97 men and women (mean ages 54 and 51 years respectively) within the European Investigation into Cancer (EPIC) study in Norfolk (UK). Dietary assessment of energy intake and physical activity was by self-report and weight was measured using standard techniques. Energy expenditure was assessed objectively by recording heart rate for 4 days following a calibration of the relationship between heart rate and oxygen consumption.
Results Self-reported energy intake by 7-day diary (mean 8.5 MJ/day) and food frequency questionnaire (FFQ) (mean 8.8 MJ/day) were significantly lower than objectively measured total energy expenditure (mean 11.2 MJ/day). The deattenuated partial correlations between total energy expenditure were 0.33 (7-day diary), 0.34 (FFQ), 0.50 (physical activity), and 0.56 (weight). Weight accounted for 31% (deattenuated) of the sum of squares about the mean of true energy intake after adjusting for age and sex. With the addition of self-reported physical activity, the model was significantly improved (R2 = 0.57). Adding energy either assessed by the diary or FFQ did not improve the model.
Conclusions The results presented here indicate that to adjust for energy intake, for the purpose of replicating an isocaloric experiment in an observational epidemiological study, one would do considerably better adjusting for weight and physical activity, than adjusting for energy intake estimated from an FFQ.
Keywords Energy, physical activity, food diary, food frequency questionnaire, weight, validation, heart-rate monitoring
Accepted 23 February 2004
There is continuing high interest in epidemiology in identifying specific components of diet which may be related to specific health endpoints. The diet–disease relationships may be with the level of consumption of specific micronutrients, macronutrients, food groups, or particular foods. In each case, it will be important to demonstrate that the relationship is specific, and not a reflection of a more general association with total quantity of food consumed. It is generally accepted that associations between nutrients and disease should only be considered primary if the effects are independent of energy intake resulting from differences in body size, physical activity, and metabolic efficiency.1 That is, epidemiology should attempt to mimic the isocaloric requirement of experimental or metabolic studies. This has led to the practice of adjusting for total energy intake, as assessed by the dietary instrument used to generate the epidemiological data. The problem, however, is that the dietary instruments used in large-scale epidemiological studies may estimate energy intake rather badly. The most commonly used instrument in cohort studies of diet has been the food frequency questionnaire (FFQ) of Willett2 or adaptations of it, although diet record or diet diary methods are now being introduced.3 A recent large study4 compared energy intake assessed by the FFQ with energy expenditure measured by the doubly labelled water technique.5 The correlation between the FFQ estimate of energy intake and the ‘gold standard’ estimate of energy expenditure was 0.1 in women and 0.2 in men. Thus as a method of adjusting for true energy intake, adjusting for the FFQ estimate of energy intake is clearly inadequate. Energy intake has not been adjusted for to any appreciable extent. The results of an analysis adjusting for the FFQ estimate of energy intake cannot be considered as replicating an isocaloric experiment.
The main thrust of this paper is to examine whether there are better measures of energy intake that could be used. The paper reports a study nested within the European Investigation into Cancer (EPIC) study in Norfolk (UK) cohort6 in which heart rate monitoring was used as an objective method of estimating energy expenditure. The use of heart rate monitoring (HRM), with individual calibration of the relationship between oxygen consumption and heart rate, to estimate energy expenditure has been demonstrated to have good correlation (0.93) with the gold standard methods of doubly labelled water and whole body calorimetry.7,8 The technique was used in this study in order to compare self-reported energy intake by 7-day diary and FFQ, self-reported physical activity and weight. It is assumed that the errors in measurement of energy expenditure associated with HRM are independent from the errors in measurement of energy associated with the FFQ and diary, reported physical activity, and weight.
There is, however, another reason, usually considered secondary,9 for adjusting by the FFQ estimate of energy intake, related to the error structure of the FFQ. The errors of estimation of the intake of many dietary components, energy included, are likely to be highly correlated. Few studies have attempted to estimate these correlations rigorously, but those that have4,10 suggest that values in the region of 0.8 to 0.9 may not be uncommon. The arithmetic of adjustment could then lead to errors in the estimated intake of different nutrients at least partially cancelling out if these nutrients were included together when regressing disease risk on diet. This effect has been demonstrated for protein.4 Since perhaps the major problem in nutritional epidemiology is the large degree of error attached to estimating dietary intake, there would be clear value in exploiting the error structure to reduce the effect of measurement error on the estimate of diet–disease relationships. It has been proposed that this reduction could be achieved by routinely adjusting for FFQ energy intake. In a companion paper we explore this issue in detail for the bivariate case.11 The conclusions of that paper suggest that routine adjustment for estimated energy intake is not the uniformly optimum approach, an issue which we expand on later in this paper.
|
Methods |
|---|
|
TopAbstractMethodsResultsDiscussionReferences |
|---|
Setting and study participants
The EPIC study is based on a cohort of 500 000 volunteers from nine European countries with the aim of relating biomarkers and lifestyle to cancer and other chronic disease endpoints. The study reported here is a calibration sub-study embedded in the EPIC-Norfolk cohort and details of recruitment and measurement protocol have been previously reported.6 Dietary assessment of energy intake by the 7-day food diary (diary) and the semi-quantitative FFQ in the Norfolk study has been described in detail previously.12,13 Self-reported physical activity has been assessed throughout the EPIC study by the short EPIC physical activity questionnaire14 and individuals are classified according to a four-point physical activity index; inactive, moderately inactive, moderately active, and active.15 Over a period of 18 months 179 members of the Norfolk cohort took part in a validation study of dietary assessment by the FFQ and diary compared with urinary markers of nitrogen, potassium, and sodium.10,16 In addition to the collection of urine and the completion of questionnaires, energy expenditure was estimated by 4 days of heart rate monitoring (HRM) with individual calibration in 97 of the 179 participants. The diary, FFQ, and self-reported physical activity were completed, on average, 1.8 (SD 1.3) years before the objective assessment of energy expenditure.
Assessment of energy expenditure
Participants visited the clinic in order to undergo a standard protocol for individually calibrating heart rate against energy expenditure by indirect calorimetry.17,18 Following the individual calibration, volunteers wore the Polar heart rate monitor (Polar Electro, Finland) continuously during the waking hours for a period of 4 days. Heart rate readings were downloaded onto a database via a serial interface and the individual calibration data were used to predict minute-by-minute energy expenditure for each participant. Sleeping energy expenditure was calculated as 95% of basal metabolic rate (BMR) where this was derived from published prediction equations.19 Physical activity level (PAL) was computed for the 4 days as the ratio of total energy expenditure (TEE) to BMR.
Statistical methods
The reliability coefficients20 for PAL and TEE, using the HRM method, were 0.5 and 0.74 respectively as calculated from a previous study in which individually calibrated HRM was employed on four separate occasions over a 12-month period in a group with similar age, sex, and geographical structure.21
Comparisons between TEE and PAL with self-report measures and body weight are represented by partial correlation coefficients (adjusted for age and sex) and deattenuated using reliability coefficients. We focused attention on correlation coefficients between various measures of energy intake and true energy intake, the reason being that this correlation reflects the extent to which confounding by true energy is accounted for by the measured variable in a linear regression. The relative contributions of the self-reported measures and body weight in explaining TEE are illustrated by the effect on R2 in regression models, again deattenuated using reliability coefficients. All statistical computations were undertaken using Stata.22 Analyses were repeated using the minute-by-minute energy expenditure obtained during the time the heart rate monitor was worn and the results remained substantially unchanged. Similarly, analyses were performed with and without transformation of energy expenditure, weight, and energy intake from both the FFQ and diary; there was no marked difference between the results. Results reported are those for the untransformed variables.
|
Results |
|---|
|
TopAbstractMethodsResultsDiscussionReferences |
|---|
The characteristics of the volunteers in this study are described in Table 1, stratified by sex. Self-reported energy intake was significantly lower for both diary and FFQ when compared with TEE. For men, estimates of energy intake from the FFQ were lower than for the diary but for women energy intake was higher from the FFQ when compared with the diary. The ratio of energy expenditure in men to the energy expenditure in women was 1.28; the ratios of energy intakes in men to intakes in women were 1.25 for the diary estimates and 1.04 for the FFQ estimates. PAL was 1.72 (0.3) for both men and women.
|
Deattenuated partial correlation coefficients, adjusted for age and sex, between TEE and energy intake were 0.33 (P < 0.01) and 0.34 (P < 0.01) for diary and FFQ respectively (Table 2). The equivalent correlations between TEE and self-reported activity and body weight were stronger than for the dietary markers; 0.50 (P < 0.001) and 0.56 (P < 0.01) respectively. Partial correlations between PAL and self-reported behaviours were similar to those for TEE. However, PAL and weight were weakly correlated.
|
The partial R2 for TEE with weight, physical activity, diary, and FFQ are reported in Table 3. This shows that weight accounts for 31% (deattenuated) of the sum of squares about the mean of true energy intake after adjusting for age and sex. Adding physical activity to the model improved the model significantly, producing an R2 of 0.57. However, adding either the FFQ or the diary to the model including weight alone did not substantially improve the R2 (0.38 and 0.39 respectively) or when either was added to the model including weight and physical activity (0.64 for both diary and FFQ).
|
Figure 1 illustrates mean values of TEE from heart rate by quintile of body weight, energy intake from FFQ and diary. The FFQ estimate of energy intake is almost independent of TEE, however, the diary does better, and weight is the most strongly associated with energy expenditure.
|
Figure 2 illustrates the positive association, adjusted for age and sex, between the self-reported physical activity index and PAL (test for monotonic trend, P < 0.001). The adjusted mean PAL in the inactive group is 1.58 (95% CI: 1.43, 1.73) and in the active group 1.94 (95% CI: 1.81, 2.07). The relationship between TEE and the physical activity index was also strong (monotonic trend P < 0.001) across the four categories (inactive group 9.93 MJ/day [95% CI: 8.8, 11.1], active group 12.8 MJ/day [95% CI: 11.9, 13.8]), adjusted for age and sex.
|
|
Discussion |
|---|
|
TopAbstractMethodsResultsDiscussionReferences |
|---|
In this sub-study within the EPIC-Norfolk cohort we have demonstrated a positive association between weight, self-reported energy intake, and self-reported physical activity with objectively assessed energy expenditure. The key findings are that both weight and self-reported physical activity demonstrate a stronger association with energy expenditure than energy intake estimated with either a 7-day diary or a FFQ. The value reported here of the correlation of the FFQ estimate of energy intake with true intake (deattenuated correlation = 0.34 after adjustment for age and sex) is higher than the values reported from the recently published US OPEN study (0.1 in women, 0.2 in men) although the differences between the two studies are not statistically significant.4 The OPEN study clearly confirms our conclusion that the FFQ provides a poor estimate of energy intake. Under an assumption of stable weight, a simple self-assessed physical activity index appears to provide an appropriate alternative to dietary intake in ranking participants according to energy expenditure. Similarly, body weight is an independent and objective marker that has a strong association with energy expenditure. In this study a combination of weight and physical activity contributed substantially to the explained variance in energy expenditure (57%).
Methodological limitations associated with this study include the long gap between the dietary assessment and the energy expenditure assessment. The mean time between dietary assessment and heart rate monitoring was 1.8 years. However, although noise introduced by true changes in behaviour over this time period is likely to dilute the association between self-reported diet intake and energy expenditure, this should not bias the association differentially. The relatively small sample prohibited stratification by sex and hence further investigation of the differences of the relationship with energy according to sex. Although the HRM method of measuring energy expenditure is free of reporting bias, it is possible that participants altered their behaviour during the time the monitor was being worn. Comparison with published values for energy expenditure and PAL indicate that our results are within the expected range for a population in this age range.23
The precision and accuracy of the measurement of energy intake in nutritional epidemiology is critical to studying the role of energy as an exposure for disease risk and also for adjusting for its role as a confounding factor. In adjusting for energy intake in cohort studies, we are attempting to mimic an isocaloric experiment. However, dietary assessment of energy is not only difficult but the effects of measurement error in its assessment can have a significant impact on the results of a study.24 A poor estimate of energy intake will lead to distorted estimates of associations of other nutritional factors with health endpoints when adjusting for energy. Adjusting for strong confounding with a poorly measured confounding factor can be worse than not adjusting at all. With the relatively recent increase in media coverage and interest in the role that physical activity and inactivity may have with regard to the rising prevalence of overweight and obesity one should be cautious that differential reporting, which has plagued dietary assessment, might become evident in self-reported level of physical activity. The use of an objective method for estimating energy expenditure, which is less likely to be susceptible to the same differential reporting biases as dietary assessment, could prove to be a useful alternative for energy adjustment in nutritional epidemiology. The further development of the heart rate technique, to allow for less-intensive individual calibration, might make this method a viable option in an epidemiological setting.
This study suggests that both FFQ and diary are poor in predicting energy. However, body weight and a simple index of physical activity appear to correlate much better with energy. The results presented here indicate that to adjust for energy for the purpose of replicating an isocaloric experiment, one would do considerably better adjusting for weight, and perhaps additionally a physical activity index, than adjusting for an FFQ estimate of energy intake. The argument in favour of energy adjustment is that it may achieve a number of aims simultaneously, one of which being the partial control of measurement error. The inclusion in a diet–disease regression of two or more dietary factors, the errors in the estimated intakes of which are highly correlated, may substantially reduce the effect of measurement error on the regression coefficients. However, as shown in the companion paper, the effect on the regression coefficients may be highly unpredictable.11 There would not appear to be a single approach which most effectively exploits the multivariate error structure to reduce the impact of measurement error. Adjusting for an estimate of total energy intake may on occasion be ineffective.
In conclusion, adjusting for energy intake is important in analyses of nutritional epidemiological data to control the confounding effects of energy intake and of body size. However, such adjustment may be better achieved by using weight, together with an index of physical activity if available, than by using an FFQ-derived estimate of energy intake. Adjusting for energy to correct partially for measurement error may often not be optimum, or even effective.
|
Acknowledgments |
|---|
The EPIC-Norfolk Study is funded by the Cancer Research Campaign, British Heart Foundation, Department of Health, Europe Against Cancer Programme of Commission of European Communities, Medical Research Council, Ministry of Agriculture, Fisheries and Food, Wellcome Trust and the World Health Organisation. We are grateful to the participants of this study.
|
References |
|---|
|
TopAbstractMethodsResultsDiscussionReferences |
|---|
1 Willett WC. Nutritional Epidemiology. New York: Oxford University Press, 1998.
2 Willett W. Invited commentary: OPEN questions. Am J Epidemiol 2003;158:22–24.
3 Bingham SA, Gill C, Welch A et al. Validation of dietary assessment methods in the UK arm of EPIC using weighed records, and 24-hour urinary nitrogen and potassium and serum vitamin C and carotenoids as biomarkers. Int J Epidemiol 1997;26(Suppl.1):S137–51.
4 Kipnis V, Subar AF, Midthune D et al. Structure of dietary measurement error: results of the OPEN Biomarker Study. Am J Epidemiol 2003;158:14–21.
5 Prentice AM (ed.). The Doubly-Labeled Water Method for Measuring Energy Expenditure, Technical Recommendations for Use in Humans. Vienna: Atomic Energy Agency, 1990.
6 Day N, Oakes S, Luben R et al. EPIC in Norfolk: study design and characteristics of the cohort. Br J Cancer 1999;80(Suppl.1):95–103.
7 Ceesay SM, Prentice AM, Day KC, Murgatroyd PR, Goldberg GR, Scott W. The use of heart rate monitoring in the estimation of energy expenditure: a validation study using indirect whole-body calorimetry. Br J Nutr 1989;61:175–86.[CrossRef][Web of Science][Medline]
8 Spurr GB, Prentice AM, Murgatroyd PR, Goldberg GR, Reina JC, Christman NT. Energy expenditure from minute-by-minute heart-rate recording: comparison with indirect calorimetry. Am J Clin Nutr 1988;48:552–59.
9 Willett W. Commentary: dietary diaries versus food frequency questionnaires—a case of undigestible data. Int J Epidemiol 2001;30:317–19.
10 Day NE, McKeown N, Wong MY, Welch A, Bingham S. Epidemiological assessment of diet: a comparison of a 7-day diary with a food frequency questionniare using urinary markers of nitrogen, potassium and sodium. Int J Epidemiol 2001;30:309–17.
11 Day NE, Wong MY, Bingham S et al. Correlated measuement error—implications for nutritional epidemiology. Int J Epidemiol 2004;33:2004;33:1373–81.
12 Bingham S, Welch A, McTaggart A et al. Nutritional methods in the European Prospective Investigation of Cancer in Norfolk. Public Health Nutr 2001;4:847–58.[Web of Science][Medline]
13 Welch AA, McTaggart A, Mulligan AA et al. DINER (Data into Nutrients for Epidemiological Research) a new data entry program for nutritional analysis in the EPIC-Norfolk cohort. Public Health Nutr 2002;4:1253–65.
14 Pols MA, Peeters PHM, Ocke MC, Slimani N, Bueno-De-Mesquita HB, Collette HJA. Estimation of reproducibility and relative validity of the questions included in the EPIC physical activity questionnaire. Int J Epidemiol 1997;26(Suppl.1):S181–89.
15 Wareham NJ, Jakes RW, Rennie KL et al. Validity and repeatability of a simple index derived from the short physical activity questionnaire used in the European Prospective Investigation into Cancer (EPIC) study. Public Health Nutr 2003;6:407–13.[CrossRef][Web of Science][Medline]
16 McKeown NM, Welch A, Runswick SA et al. The use of biological markers to validate self reported dietary intake in a random sample of the European Prospective Investigation into Cancer (EPIC) UK Norfolk Cohort. Am J Clin Nutr 2001;74:188–96.
17 Wareham NJ, Hennings SJ, Prentice AM, Day NE. Feasibility of heart rate monitoring to estimate total level and pattern of energy expenditure in a population-based epidemiological study: the Ely young cohort feasibility study 1994–5. Br J Nutr 1997;78:889–900.[CrossRef][Web of Science][Medline]
18 Wareham NJ, Hennings SJ, Byrne CD, Hales CN, Prentice AM, Day NE. A quantitative analysis of the relationship between habitual energy expenditure, fitness and the metabolic cardiovascular syndrome. Br J Nutr 1998;80:235–41.[Web of Science][Medline]
19 James WPT, Schofield EC. Human Energy Requirements. Oxford: Oxford University Press, 1990.
20 Armstrong BK, White E, Saracci R. Principles of Exposure Measurement in Epidemiology. New York: Oxford University Press, 1994.
21 Wareham NJ, Wong M-Y, Day NE. Glucose intolerance and physical inactivity: the relative importance of low habitual energy expenditure and cardiorespiratory fitness. Am J Epidemiol 2000;152:132–39.
22 Stata Statistical Software [program]. Release 7.0 version. College Station, TX: Stata Corporation, 2001.
23 Black AE, Coward WA, Cole TJ, Prentice AM. Human energy expenditure in affluent societies: an analysis of 574 doubly-labelled water measurements. Eur J Clin Nutr 1996;50:72–92.[Web of Science][Medline]
24 Kipnis V, Freedman LS, Brown CC, Hartman AM, Schatzkin A, Wacholder S. Effect of measurement error on energy-adjustment models in nutritional epidemiology. Am J Epidemiol 1997;146:842–55.
CiteULike 
Connotea 
Del.icio.us What's this?
This article has been cited by other articles:
|
|
 |
|
  H. Besson, S. Brage, R. W Jakes, U. Ekelund, and N. J Wareham Estimating physical activity energy expenditure, sedentary time, and physical activity intensity by self-report in adults Am. J. Clinical Nutrition, January 1, 2010; 91(1): 106 - 114. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 T. K. Lam, A. J. Cross, D. Consonni, G. Randi, V. Bagnardi, P. A. Bertazzi, N. E. Caporaso, R. Sinha, A. F. Subar, and M. T. Landi Intakes of Red Meat, Processed Meat, and Meat Mutagens Increase Lung Cancer Risk Cancer Res., February 1, 2009; 69(3): 932 - 939. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 S. Bingham, R. Luben, A. Welch, Y. L. Low, K. T. Khaw, N. Wareham, and N. Day Associations between dietary methods and biomarkers, and between fruits and vegetables and risk of ischaemic heart disease, in the EPIC Norfolk Cohort Study Int. J. Epidemiol., October 1, 2008; 37(5): 978 - 987. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
P. Ferrari, N. E Day, H. C Boshuizen, A. Roddam, K. Hoffmann, A. Thiebaut, G. Pera, K. Overvad, E. Lund, A. Trichopoulou, et al. The evaluation of the diet/disease relation in the EPIC study: considerations for the calibration and the disease models Int. J. Epidemiol., April 1, 2008; 37(2): 368 - 378. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 A. E. Cust, N. Slimani, R. Kaaks, M. van Bakel, C. Biessy, P. Ferrari, M. Laville, A. Tjonneland, A. Olsen, K. Overvad, et al. Dietary Carbohydrates, Glycemic Index, Glycemic Load, and Endometrial Cancer Risk within the European Prospective Investigation into Cancer and Nutrition Cohort Am. J. Epidemiol., October 15, 2007; 166(8): 912 - 923. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
K.-T. Khaw, R. Jakes, S. Bingham, A. Welch, R. Luben, N. Day, and N. Wareham Work and leisure time physical activity assessed using a simple, pragmatic, validated questionnaire and incident cardiovascular disease and all-cause mortality in men and women: The European Prospective Investigation into Cancer in Norfolk prospective population study Int. J. Epidemiol., August 1, 2006; 35(4): 1034 - 1043. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 T. Norat, S. Bingham, P. Ferrari, N. Slimani, M. Jenab, M. Mazuir, K. Overvad, A. Olsen, A. Tjonneland, F. Clavel, et al. Meat, Fish, and Colorectal Cancer Risk: The European Prospective Investigation into Cancer and Nutrition J Natl Cancer Inst, June 15, 2005; 97(12): 906 - 916. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
N. Day, M. Wong, S Bingham, K. Khaw, R Luben, K. Michels, A Welch, and N. Wareham Correlated measurement error--implications for nutritional epidemiology Int. J. Epidemiol., December 1, 2004; 33(6): 1373 - 1381. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
D. Spiegelman Commentary: Correlated errors and energy adjustment--where are the data? Int. J. Epidemiol., December 1, 2004; 33(6): 1387 - 1388. [Full Text] [PDF]
|
|
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||