QUICK SEARCH:   [advanced] Author: Keyword(s): Year:  Vol:  Page:

This Article Abstract Full Text (PDF) Purchase Article View Shopping Cart Alert me when this article is cited Alert me if a correction is posted Services Similar articles in this journal Similar articles in PubMed Alert me to new issues of the journal Download to citation manager Citing Articles Citing Articles via HighWire Citing Articles via Google Scholar Google Scholar Articles by Livingstone, M. B. E. Articles by Black, A. E. Search for Related Content PubMed PubMed Citation Articles by Livingstone, M. B. E. Articles by Black, A. E.

---

Supplement: Biomarkers of Nutritional Exposure and Nutritional Status

Markers of the Validity of Reported Energy Intake¹

School of Biomedical Sciences, University of Ulster at Coleraine, Coleraine BT52 1SA, Northern Ireland, U.K.

² To whom correspondence should be addressed. E-mail: mbe.livingstone{at}ulster.ac.uk.

	ABSTRACT

TOP ABSTRACT Energy intake for weight... Direct comparison of energy... Validation against presumed... Urinary nitrogen excretion in... Consequences of underreporting Some issues of interpretation... Can data be adjusted... Future directions LITERATURE CITED

 
Energy intake (EI) is the foundation of the diet, because all other nutrients must be provided within the quantity of food needed to fulfill the energy requirement. Thus if total EI is underestimated, it is probable that the intakes of other nutrients are also underestimated. Under conditions of weight stability, EI equals energy expenditure (EE). Because at the group level weight may be regarded as stable in the timescale of a dietary assessment, the validity of reported EI can be evaluated by comparing it with either measured EE or an estimate of the energy requirement of the population. This paper provides the first comprehensive review of studies in which EI was reported and EE was measured using the doubly labeled water technique. These conclusively demonstrate widespread bias to the underestimation of EI. Because energy requirements of populations or individuals can be conveniently expressed as multiples of the basal metabolic rate (BMR), EE:BMR, reported EI may also be expressed as EI:BMR for comparison. Values of EI:BMR falling below the 95% confidence limit of agreement between these two measures signify the presence of underreporting. A formula for calculating the lower 95% confidence limit was proposed by Goldberg et al. (the Goldberg cutoff). It has been used by numerous authors to identify individual underreporters in different dietary databases to explore the variables associated with underreporting. These studies are also comprehensively reviewed. They explore the characteristics of underreporters and the biases in estimating nutrient intake and in describing meal patterns associated with underreporting. This review also examines some of the problems for the interpretation of data introduced by underreporting and particularly by variable underreporting across subjects. Future directions for research are identified.

KEY WORDS: • biomarkers • diet assessment • energy intake • epidemiology • nutrition

Energy intake (EI)⁵ is the foundation of the diet, because all other nutrients must be provided within the quantity of food needed to fulfill the energy requirement. Reported EI is therefore a surrogate measure of the total quantity of food intake. If total EI is underestimated, then the intakes of nutrients correlated with EI (the macronutrients, most minerals and the B vitamins) are also likely to be underestimated. This may, for example, lead to overestimation of the proportion of the population with deficient intake or distortion of the associations between nutrient intake and disease outcome. Evaluating the validity of reported EI provides a valuable check on the general quality of the dietary data in any study.

Three concepts are fundamental to understanding the limitations of dietary assessment: habitual intake, validity and precision.

The habitual intake of an individual is the person's intake averaged over a prolonged period of time (weeks or months rather than days). For energy it is the intake that maintains weight stability. For other nutrients, it may be thought of as the intake required to produce a steady physiological state and hence to influence nutritional status and health in both the short and the long terms. Habitual intake is the value that studies of diet and health would ideally measure; however it is a largely hypothetical concept, because intake varies widely from day to day. Weekly or monthly variation can also be significant (1).

A valid (or accurate) report is one that measures the true intake during the period of study. A valid diet record is a complete and accurate record of all food consumed on specified days, and where the choice of food and drink consumed has not been influenced by the act of recording, i.e., a subject ate and drank exactly what he/she would have eaten and drunk if he/she had not been involved in a research study. A valid diet recall is a complete and accurate recall of all food and drink consumed on specified day(s). A valid diet history or food-frequency questionnaire (FFQ) accurately reflects typical food consumption over a designated period of time, undistorted by behavioral patterns or false memory. Poor validity derives from systematic errors (bias) in the reporting of food intake.

A precise technique is one that yields the same answer on repeat administrations. Precision may be expressed in various ways; for example, 1) mean absolute difference, 2) mean difference as a percentage of overall mean intake, 3) coefficient of variation of the differences within individuals, 4) correlation coefficient, or 5) percentage of individuals classified in the same quantile on both occasions. As noted above, food intake varies widely with time. Therefore precision of dietary assessment at the individual level is poor even when repeat surveys show good agreement for mean intake. Some published values for the coefficient of variation of the differences within individuals are 16.5% for 3-d food records (2), 18.6% for a dietary history (3) and 28.5% for an FFQ (4). Poor precision derives from large random errors in the techniques of dietary assessment. It reduces the sensitivity for identifying invalid reports (Fig. 1).

View larger version (23K): [in this window] [in a new window]   FIGURE 1  Visual representation of accuracy (validity) and precision (repeatability). From Black, 1999 (200). True average (solid lines), repeat measurements (•) and measured average (dashed lines) are shown.

Understanding the limitations of dietary assessment techniques and the quantification of the errors involved has been handicapped for decades by a lack of independent methods for validation. Only for the 24-h recall technique or for individual meals (where direct observations are possible) have any true studies of validity been undertaken. In early studies in which a subject's intake was covertly observed and then assessed by recall, intake was usually underestimated, the assumption being that this was primarily due to faulty memory (6). Nevertheless until recently, dietary intakes were reported as if valid, and the interpretation of links between intake and health were based, often erroneously, on the assumption of validity. Until the advent of biomarkers, the so-called validation studies simply compared the results of one method with another. The weighed dietary record was often assumed to be the gold standard, and the validity of other methods was evaluated by comparison. These studies are actually studies of "relative validity." Recently the term "calibration" studies was introduced to describe studies of relative validity and to distinguish them from studies of validity that use external markers of intake. Only the advent of external markers of intake has made it possible to test assumptions about validity.

In contrast to the micronutrients, there are no biochemical biomarkers of EI. All three methods of validation rest on the assumption that EI must equal energy expenditure (EE) when weight is stable. These methods are as follows: 1) comparison of self-reported EI with the EI required to maintain weight; 2) direct comparison of reported EI and measured EE; and 3) comparison of reported EI with presumed energy requirements, both expressed as multiples of basal metabolic rate (BMR).

The technique for the validation of protein intake, 24-h urinary nitrogen excretion, also has some limited value as a marker of EI.

	Energy intake for weight maintenance

TOP ABSTRACT Energy intake for weight... Direct comparison of energy... Validation against presumed... Urinary nitrogen excretion in... Consequences of underreporting Some issues of interpretation... Can data be adjusted... Future directions LITERATURE CITED

 
The earliest indicator of bias to the underestimation in self-reported EI came from an 18-wk metabolic study in which the EI reported by the subjects before entry into the study was 27% less than that required to maintain weight during the study (9). The finding has been confirmed in other similar studies (10–12). Clearly this technique cannot be used to validate reported EI in studies of free-living subjects, although combined residential and free-living studies have been used to examine characteristics of underreporters (12), psychological and behavioral correlates of underreporting (13) and the accuracy of the multiple-pass 24-h recall (14). However, because in the time scale of a diet record the group mean weight should remain unchanged, the monitoring of body weight at the beginning and end of a diet record can indicate whether, on average, undereating has occurred. Substantial undereating such as temporary dieting may be detectable at the individual level.

	Direct comparison of energy intake and measured energy expenditure

TOP ABSTRACT Energy intake for weight... Direct comparison of energy... Validation against presumed... Urinary nitrogen excretion in... Consequences of underreporting Some issues of interpretation... Can data be adjusted... Future directions LITERATURE CITED

 
The doubly labeled water technique.

The doubly labeled water (DLW) technique is the gold standard for measuring EE under free-living conditions. The subject is given a dose of water enriched with the stable isotopes deuterium (²H) and oxygen 18 (¹⁸O). Urine samples are collected at baseline before administration of the dose and subsequently either daily [multipoint method (15)] or at the beginning and end of the measurement period [two-point method (16)]. The urine samples are analyzed by isotope ratio mass spectrometry to determine the rate of disappearance of each isotope from the body. Deuterium is lost in water only, whereas oxygen 18 is lost in both water and carbon dioxide. The rates of disappearance measure the body's water and water-plus-carbon dioxide turnover rates, from which carbon dioxide production can be calculated by difference. The total EE is calculated from carbon dioxide production by applying the classical indirect calorimetric equations. The measurement period is most usually 14 d in adults, but periods from 7 to 21 d have been used. The principle of the method, experimental protocol, details of mass spectrometric analysis, methods of calculation, fractionation and respiratory quotient assumptions and sources of errors have been fully documented elsewhere (17,18).

The DLW technique has been validated against concurrent measurements of EE by respiratory gas exchange in a wide variety of subjects and metabolic circumstances including sedentary adults, adults exercising to exhaustion and subjects in states of energy balance and imbalance (16,19–25). Under well-controlled experimental conditions, accuracy is on the order of 1–3% and precision is 2–8%. A review of 14 studies with repeat DLW measurements in the field (26) estimated analytical variation in the method to be 4% and within-subject physiological variation to be 7%.

Because the DLW measurement is integrated over 10–14 d, it accounts for daily and weekly fluctuations in EE. It does not account for monthly or seasonal fluctuations and is not necessarily a measure of habitual EE. In 32 studies with repeat DLW measurements, the mean within-subject coefficient of variation in EE (CV_wEE), including analytical and physiological variation and that due to changed activity, ranged from 6.5 to 22.6%. Multiple regression showed a positive association between mean CV_wEE and the time span covered by the DLW measurements. There were small associations with weight change and reproduction, but neither accounted for additional variance above that accounted for by time span. There were no associations with age, sex, mean age or mean total EE. From the regression equation, CV_wEE ranged from 8.2% at zero time span to 25.4% at 52 wk (27).

Doubly labeled water provides an independent and objective measure of EE and it is easy to use in the field because it places minimal burden on the subjects and no restriction on the subjects' activities. Unfortunately it requires sophisticated laboratory and analytical backup and is extremely expensive; therefore it cannot be a routine tool for validating EI data.

Limitations of validating energy intake using doubly labeled water energy expenditure.

The validation of reported EI against measured EE rests on the fundamental physiological equation

At the group level and in the time scale of a dietary assessment, body weight can be regarded as constant, and therefore mean EI must equal mean EE. Validation is by direct comparison of EI with EE. Various modes of expressing this comparison have been used, including the ratio EI:EE (as a proportion of 1.0 or as a percentage), the percentage difference (EI - EE/EE) x 100 (i.e., the reciprocal of EI:EE) and the difference EI - EE. Alternatively a number of authors have presented data as a Bland-Altman plot [EE - EI against the means of both measurements (28)] and shown ±2 SD of the difference. However these define the confidence limits of the data including any invalid (biased) data. To identify invalid reports, the confidence limits of a valid data set must be defined.

For individual subjects in energy balance, habitual EI must equal habitual EE. The expected ratio of EI:EE is 1.00, but variation in both measurements means that absolute agreement cannot necessarily be expected even for valid data. The limitations of EI:EE for identifying individual underreporters were explored in data from 22 studies (249 subjects) with measurements of both EI and DLW-EE (27). Values falling above or below the 95% confidence limits of the ratio indicate over- or underreporting, respectively. The 95% confidence limits of the ratio define the range within which the differences between EI and EE could have arisen by chance in a valid data set. The 95% confidence limits were calculated as

where CV_wEI is the within-subject coefficient of variation for daily EI, d is the number of days of diet records, CV_wEE is the within-subject coefficient of variation for repeat DLW-EE and r is the correlation between EI and EE. Within-subject CV_wEI for daily EI ranges from 15 to 45% with an average of 23% (6). For the purpose of assigning confidence limits to the diet history and food-frequency questionnaire, CV_wEI based on weighed records assuming a 28-d record was used. The CV_wEE derived from analysis of studies with repeat DLW measurements is 8.2% assuming EE was measured concurrently with EI (27) and the correlation between EI and EE from accumulated individual DLW data was r = 0.425 (29). Substituting these values into the equation and assuming a 7-d record yields a figure of ±18%. Thus subjects with EI:EE < 0.82 or > 1.18 would be deemed under- or overreporters, respectively. A value of 15% is the lowest that could be expected, and shorter records have substantially wider confidence limits of approximately ±40% for a 1-d 24-h recall (27). When individual data from accumulated DLW studies were examined and values between ±18% were excluded, mean EI:EE was 0.97. Thus not all invalid records could be identified, and even under the best conditions a bias of -3% in mean intake remained.

Other direct measures of energy expenditure.

Reported EI can also be compared directly with estimates of EE derived from heart-rate monitoring, accelerometers or physical activity questionnaires. Four validation studies have done so, deriving EE from leisure-time activity and BMR using an equation derived from earlier DLW studies (30), an activity diary (31), the EI for weight maintenance (12) and heart-rate monitoring (32). These other methods have their own sources of errors and bias. Their precision and validity is certainly poorer and the sensitivity and specificity for detecting invalid reports are worse than when DLW-EE is used. However, if the characteristics of underreporters are to be explored, techniques such as these must be used (see EI:BMR: The Goldberg cutoff technique). A detailed discussion of other methods for measuring EE are beyond the scope of this review.

Doubly labeled water validation studies of reported energy intake.

Since 1986, many studies have been conducted in which DLW-EE and reported EI have been measured concurrently in the same individuals. Figure 2 shows EI:EE from 43 studies of adults comprising 77 subgroups (men and women separately). Mean ± SD EI:EE was 0.83 ± 0.14. In 22 (29%) subgroups, EI and EE agreed to within ±10%, but 53 (69%) subgroups had a reported mean EI > 10% below mean EE, whereas only two groups had a mean EI > 10% above mean EE.

View larger version (49K): [in this window] [in a new window]   FIGURE 2  Frequency distribution of the ratio of energy intake to energy expenditure (EI:EE) by sex in 43 doubly labeled water (DLW) energy-expenditure (DLW-EE) studies of adults comprising 77 subgroups (men and women separately).

    Subject-specific bias to dietary assessment. Dietary surveys usually report a range of EI that at the extremes of the distribution cannot represent habitual intake. It was customary to assume that these extreme values were obtained by chance due to day-to-day variation in food intake, and that repeat measurements would eventually obtain a measure of the habitual intake of individuals. However, this assumption does not necessarily hold. An analysis of seven studies conducted in Cambridge and Belfast, U.K. provides evidence for subject-specific bias in repeated weighed records and also in assessment by different methods (34). Two studies are shown in Figure 3 in which expressing misreporting as the ratio EI:EE controls for real differences in EE. Figure 3A (adults) demonstrates variable bias across individuals but similar bias within individuals in weighed records obtained 2 y apart. Figure 3B demonstrates differential mean bias between weighed records and diet history and variable bias across individuals with a tendency to rank similarly by both methods. Further evidence of subject-specific bias comes from three national surveys. Price et al. (35) found that the strongest indicator of underreporting in 1989 was being an underreporter in 1982. Kroke et al. (36) reported a high correlation (=0.74; n = 28) between the degree of underreporting (EI - EE) from two different assessments (the German European Prospective Study on Nutrition, Cancer and Health (EPIC) FFQ versus 12 x 24-h recalls). In the third National Health and Nutrition Examination Survey (NHANES III) (37), a subsample of 311 men and 312 women provided 2 x 24-h recalls 1 mo apart: Fifty-five percent of men and 58% of women who underreported on the first occasion also underreported on the second occasion.

View larger version (23K): [in this window] [in a new window]   FIGURE 3  Subject-specific bias in dietary assessment demonstrated by repeat measurements of EI validated by DLW-EE. Thirty-one adults in whom EI was measured by 7-d weighed record in 1987 and 1989 and DLW-EE was measured in 1989 [from (34), data of Livingstone et al. 1990 (33)] (upper panel). Fifty-eight children and adolescents aged 7–18 y in whom EI was measured by 7-d weighed records and by diet history within a 4-wk period and DLW-EE was concurrently measured with the weighed records [from (34), data of Livingstone et al. 1992 (48)] (lower panel).

    Men versus women. Table 1 summarizes mean EI:EE values from studies that included both sexes. None of the tested differences between men and women were significant. The overall mean ± SD values for men and women, respectively, were 0.87 ± 0.09 and 0.85 ± 0.09. In an alternative analysis of individual data from 21 studies comprising 429 adults (29), underreporters, acceptable reporters and overreporters were defined as having EI:EE values < 0.76, 0.76–1.24 and > 1.24, respectively. For men and women, respectively, the proportion of underreporters, valid reporters and overreporters were 28, 67 and 5% and 38, 59 and 4%. These differences between men and women were not significant (² = 4.25; df = 2; P = 0.2).

Figure 4 summarizes EI:EE from 43 studies comprising 77 groups of men and women by dietary assessment method (33,50–60).

View larger version (49K): [in this window] [in a new window]   FIGURE 4  Frequency distribution of EI:EE according to dietary assessment method in 43 studies of adults comprising 77 subgroups (men and women separately).

View larger version (23K): [in this window] [in a new window]   FIGURE 6  Ratio of EI to estimated basal metabolic rate (EI:BMRest) against physical activity level (PAL) (ratio of EE to measured basal metabolic rate, BMRmeas) in female subjects (n = 264) from 21 studies with measured EI, BMR and DLW-EE. Symbols designate acceptable reporters, over- or underreporters (AR, OR and UR, respectively) by the direct comparison of EI:DLW-EE. Lower Goldberg cutoff for PAL = 1.55 or 1.95, d = 7 and n = 1 are shown (horizontal lines) [from Black, 2000 (29)]. Underreporters, EI:DLW-EE < 0.76; acceptable reporters, EI:DLW-EE = 0.76–1.24; overreporters, EI:DLW-EE > 1.24.

    Undereating versus underrecording. Underrecording is the discrepancy between EI and measured EE where there is no change in body mass. Undereating, on the other hand, is accompanied by a decline in body mass over the food-recording interval (74). When reported EI is adjusted to account for changes in body mass, some of the bias can be attributed to undereating (44,54,74,75). Underrecording is then inferred from the difference between total bias and the amount attributable to undereating. However, underrecording was ingeniously quantified using water balance (76,77). Total water intake was calculated from reported food and water intakes and the calculated amount of metabolic water. Water loss was determined from deuterium elimination over the period of study. Because true water intake must equal water loss, underrecording was defined as [(water intake - water loss)/(water loss)] x 100%. Undereating was defined as [(body mass change in recording week x 30 MJ/7 d)/(EE)] x 100%. In lean women, the water intake and loss values were equivalent, but body weight declined and the totality of the reporting bias (-16%) could be attributed to undereating (76). In contrast, in obese men, calculated underrecording bias was -26%, calculated undereating bias was -12% and together they accounted for the total reporting bias of -37% (77). In both studies, weight was monitored during nonrecording and recording weeks. The respective mean ± SD weight changes were 0.07 ± 0.59 and -0.57 ± 0.77 kg for lean women and 0.0 ± 1.0 and -1.0 ± 1.3 kg for obese men. These figures are uncomfortable reminders of the effect that recording food intake has on modifying behavior in the direction of reduced intake.

    Body weight status. Profound underreporting was found in obese subjects recruited in response to advertising or through obesity clinics (54,75,78). A negative association between the extent of underreporting and measures of weight status (body weight, percentage body fat or BMI) has also been found in studies that have encompassed a range of body sizes (30,79). The fact that this association is found in samples that include lean, overweight and obese subjects is understandable. However, this association has also been found in groups of subjects not including obese persons (13,45), although the association has not always been significant (80). The probability of underreporting increases as BMI increases, but not all obese persons underreport, and not all normal-weight persons provide valid reports. Figure 5 shows the association between EI:EE and BMI in subjects of DLW studies recruited from the community (i.e., not specifically obese subjects). Both under- and overreporters appear across the full range of BMI values from lean to grossly obese.

View larger version (22K): [in this window] [in a new window]   FIGURE 5  Relationship between EI:EE and body mass index (BMI) in studies of adult volunteers (n = 243 women,165 men) recruited from the community (excluding obese subjects recruited specifically for studies of obesity).

Overall, the DLW validation studies provide substantial evidence of bias to underestimation by diet records. However, because the sample sizes were often small and the subjects were generally highly selected, they cannot be regarded as representative of randomly selected populations. Thus the true size of the bias remains unquantified. Unfortunately there have been relatively few DLW studies that validate the diet history, recall or FFQ, and the results are even less conclusive. Nevertheless, DLW-EE could be extremely valuable in validating the ability of different methods for ranking subjects as well as measuring mean intake in a randomly selected subsample from a large-population study, particularly if it were used alongside other biomarkers. Such a definitive validation study remains to be done.

	Validation against presumed energy requirements

TOP ABSTRACT Energy intake for weight... Direct comparison of energy... Validation against presumed... Urinary nitrogen excretion in... Consequences of underreporting Some issues of interpretation... Can data be adjusted... Future directions LITERATURE CITED

 
EI:BMR: The Goldberg cutoff technique.

Unfortunately, DLW is too expensive and technically challenging to be used for routine validation of EI. However, reported EI can also be evaluated against presumed energy requirements (83,84). In this technique, mean EI is expressed as a multiple of the mean BMR estimated from equations (85) and is compared with the presumed mean EE of the population, which is also expressed as a multiple of the BMR. The ratio EE:BMR is here referred to as the physical activity level (PAL), although other authors refer to it as the average daily metabolic rate (ADMR). The equation of Goldberg et al. (83) calculates the lower 95% confidence limit of EI:BMR assuming a given PAL requirement, below which it is unlikely that the mean intake represents either habitual intake for weight maintenance or a random low intake (the Goldberg cutoff). It makes allowance for the errors associated with the number of subjects (n), the length of the dietary assessment (d days) and variation in each of food intake, BMR and physical activity. In the original publication, the cutoff was calculated assuming an energy requirement of 1.55 x BMR, and it was demonstrated conclusively that underreporting was widespread (86). In two-thirds of 37 community-based and epidemiological surveys from 10 countries, the mean EI:BMR was below the study-specific Goldberg cutoff (86).

A PAL of 1.55 x BMR was selected as the basis for comparison because it is the value defined by FAO/WHO/UNU (87) as that which represents a sedentary level of energy expenditure. However, subsequent analysis of nearly 600 DLW measurements has shown that this is a conservative figure (38). In 16 age-sex groups, except those over 75-y-old, the mean PAL in free-living people was > 1.55. Thus the extent of underreporting based on this figure was almost certainly underestimated.

The Goldberg equation was devised to evaluate the overall bias to underreporting at the group level. In theory, the cutoff value calculated for a sample size of n = 1 can be used to identify underreporters at the individual level. Since publication of the Goldberg equation, numerous investigators have used a cutoff value based on a PAL of 1.55 and n = 1 to identify a group of "low-energy reporters" (LER) (31,35,37,39,88–95). These studies are summarized in Table 5 and discussed below (see Factors associated with low-energy reporting). It has been demonstrated subsequently that this approach identifies only 50% of underreporters (29). It excludes those who have underreported from a higher intake such that EI:BMR does not fall below the cutoff for a PAL of 1.55 x BMR.

Undoubtedly, the Goldberg cutoff has been useful in raising awareness of the issue of misreporting, but it has considerable limitations, and the concepts have not always been either fully understood or applied correctly. Common misinterpretations of the concepts include applying the cutoff calculated for groups to individuals, confusing the cutoff for habitual intake with that for low intake obtained by chance and interpreting the exemplar tables of cutoffs based on a PAL value of 1.55 as recommendations for universal application. The effect of the latter misapplication on estimates of the proportion of underreporters has been examined in young adults (90), where true PAL values were higher (1.92 and 1.77), and in children aged 1–18 y (96), where age-specific cutoff values based on sedentary PAL values in children (1.45–1.6) (97) were compared with the blanket cutoff based on 1.55.

The principles of the technique have recently been restated and the use and limitations of the technique fully discussed (84). At the group level, the Goldberg cutoff can be used to assess the overall bias in a study provided that a PAL value appropriate to the population is used for comparison. At the individual level, the cutoff is limited by low sensitivity and poor specificity. To improve sensitivity, methods must be found to account for differing levels of physical activity. Simple questionnaires have their errors and raise the problem of choosing appropriate PAL values for differing levels of activity. If more sophisticated methods of assessing EE such as heart-rate monitoring are used, then EE can be assessed in absolute terms and EI can be compared directly with EE, in which case the Goldberg cutoff becomes redundant.

Five studies exemplify these approaches. Four have compared EI directly with EE, the latter being derived respectively from leisure-time activity and BMR using an equation based on earlier DLW studies (30), an activity diary (31), the EI for weight maintenance (12) and heart-rate monitoring (32). Samaras et al. (98) used a physical activity questionnaire to assign subjects to low, medium and high activity categories, but did not calculate Goldberg cutoffs; they merely defined acceptable reporters and underreporters as those above or below the FAO/WHO/UNU (87) recommended activity level. None of these five studies compared the technique used with any other, and because individual data were not presented, it is impossible to determine how these techniques might compare with a blanket EI:BMR cutoff in identifying misreporters or whether the conclusions about association between misreporting and subject characteristics would have been different. The identification of individual misreporters is clearly complex, but until this is achieved, the characteristics of the bias cannot be fully explored or its effects minimized in the analysis of dietary data.

Factors associated with low-energy reporting

The use of EI:BMR to check the validity of EI data has increased awareness of the prevalence of underreporting in dietary surveys. Since 1991 a substantial but by no means definitive body of evidence has been generated examining the factors associated with underreporting. Results from 10 studies based on national dietary surveys have been reviewed and extensively discussed by Macdiarmid & Blundell (99). This section briefly summarizes the results from a larger body of data.

Table 5 summarizes 25 adult studies that have examined the characteristics of LER. The majority used a single Goldberg cutoff for EI:BMR to define LER and compared these with either the rest of the study population or a subgroup from the top end of the distribution or across quintiles. The cutoff values for EI:BMR ranged from < 0.9 to < 1.28 depending on the criteria set by the authors. A limited number of studies compared EI directly with EE or assigned subjects to low-, medium- or high-activity levels. One study assigned age-sex-specific PAL values and calculated subject-specific cutoffs. Two studies used 24-h urinary nitrogen excretion as the validation tool. Sample size ranged from 50 to 11,000; age from 18 to 64 y; mean study BMI from 22.1 to 27.8 kg/m² and mean study EI:BMR from 1.09 to 1.57. The number of variables examined for their association with low-energy reporting varied from two or three anthropometric measures to a comprehensive range of anthropometric, sociodemographic and sociopsychological factors. Consequently it is highly probable that the many differences between studies and the variations in statistical analysis may have influenced the conclusions drawn from each.

    Weight status. The most robust finding in 22 out of 25 studies was a positive association between low-energy reporting and a high BMI. Furthermore, weight status was the single most significant variable associated with underreporting in those studies that examined a range of variables. However, the association between obesity and low-energy reporting is not absolute. The probability that a subject will underreport increases as BMI increases (35,101), but there are obese subjects who do not underreport and nonobese subjects who do underreport. Johansson et al. (101) noted that 52% of underreporters had a BMI < 25 kg/m² and, although the proportion of underreporters was highest among obese subjects (BMI > 30), only 5% of the total sample was obese. In the study by Samaras et al. (98), the proportion of underreporters increased from 28% using a blanket cutoff of 1.35 x BMR to 48% when three cutoffs based on three activity levels were used. However, the conclusions about associations between underreporting and parameters of body size and composition remained unaltered.

    Age-sex effects. Most studies found a higher proportion of LER among women and older subjects. It is unclear, however, whether this is a true finding or an artifact of the application of a single cutoff for EI:BMR. Doubly labeled water data suggest that men and younger subjects have higher EE (38) and thus higher EI values. If they underreport to the same degree as women and older subjects, then a single cutoff applied to all would inevitably identify more women and older persons as LER. In those studies with information on EE, Adams (100) found an association of underreporting with female sex and younger age. On the other hand, Johnson et al. (30) and de Vries et al. (12) found an association with female sex but none with age in relatively young groups. The DLW studies reviewed (see Men versus women) did not find significant differences in reporting between men and women. The inconsistencies in age-sex associations require further investigation in representative population samples that have identified underreporters at all levels of energy requirement.

    Socioeconomic effects. The effects of education and/or socioeconomic (SE) status on reporting accuracy are less predictable. On the one hand, poor literacy skills in the less well educated might be expected to result in underreporting. On the other hand, health or diet consciousness in the better educated or those of higher SE status might prompt the same response. Of the 11 studies in which this variable was examined, the findings were inconsistent. For example, one study (89) observed that LER in men was associated with living in a low-SE area but having a higher occupation. In low-income women, poor literacy scores along with percentage body fat were the best predictors of misreporting of EI (79). On the other hand, in two studies, LER was associated with higher SE status (93,102).

    Health consciousness. Although various aspects of perceived health status may be associated with reporting accuracy, only smoking and facets of dieting behavior have been significantly associated with LER. Significant associations have been found between LER and not smoking (37,41,94). The positive association between underreporting and obesity, weight consciousness and dieting is the most securely based. Factors that have been associated with misreporting include reporting of fewer foods and trying to lose weight (37), dieting in the recent past and efforts to maintain weight stability (41), dieting during the survey period (39), weight change over 5 y (103), self perceptions of feeling too heavy and having dieted at least once (93).

    Cultural effects. Just as there are cultural differences in attitudes toward food, so it is likely that there are cultural differences in dietary reporting behaviors. The majority of studies in Table 5 found an association between LER and low education or SE status. Exceptionally, in France, LER was associated with high SE status (93). In the U.S.A., an association with non-Caucasian compared with Caucasian ethnic groups has been found (91). The less pronounced degree of misreporting that has been observed in older African Americans has been attributed to their more relaxed attitudes toward body image and body weight (104). In a collaborative study in five European cities (Cambridge, U.K.; Maastricht, Netherlands; Potsdam, Germany; Copenhagen, Denmark; Barcelona, Spain), the mean reported EI was lowest in Cambridge and relatively high in Potsdam (our unpublished data). Cultural patterns and attitudes regarding food and body weight in each country may account for the reporting differences. However differences in study samples, recruitment procedures and field workers' attitudes and behaviors cannot be ruled out as contributory factors. Nevertheless, Table 5 includes data from 10 different countries, and the inescapable conclusion is that underreporting seems to be a universal phenomenon in Western cultures.

    Behavioral effects. Subject response to recording food intakes has been little explored but merits much more attention in the future. Subjects questioned hypothetically about what they might do if asked to keep a dietary record openly admitted the possibility of misreporting (105). Postsurvey focus groups have reported their food records to be an accurate account of food eaten but admitted to simplifying food choice to ease the reporting burden. In another postsurvey interview, 46% of subjects admitted altering their eating pattern during a 7-d record. Those indicating inconvenience as the reason for changed behavior had a mean EI:BMR of 1.53; those indicating embarrassment or guilt had a mean EI:BMR of 1.1. However those who claimed not to have changed their eating pattern had an EI:BMR of 1.23 (106). These findings highlight some key issues for future research. Are people too guilty to admit to changed eating habits, or do they truly not perceive that they have altered their eating behavior? A person may eat a series of meals that is within their normal dietary pattern and perceived as normal, but is not what they would have eaten had the survey not intervened. Exhortations by researchers such as "please don't change what you eat" may be falling on deaf ears.

A plausible influence on reporting behavior may be the need to achieve a self-presentation goal through socially desirable responding. Worsley et al. (107) found a significant positive correlation between a 12-item scale for social desirability specifically related to food and the intake of vegetables and fresh fruit, but negative correlation with snack foods such as cakes, cookies, chocolates, candies, pies and pastries. On the other hand, no such association was observed with the nonspecific Marlow-Crowne score for social desirability responding (108). This suggests that it is social desirability behaviors specifically related to food intake that need to be studied in the context of misreporting. Hebert et al. (109,110) distinguished between social desirability—the defensive tendency of individuals to respond in a manner consistent with societal norms and to avoid criticism, and social approval—the tendency for an individual to seek a positive response (praise) in testing situations. They hypothesized that the FFQ, because it involves complex cognitive tasks, may be more prone to both biases than are short-term records that are focused on specific foods eaten on specific days. For men, the social approval score was significantly associated with greater fat and energy intake reported by FFQ, whereas in women a higher social desirability score was associated with lower reported fat and energy intake by FFQ.

    Psychological effects. Various psychological instruments have been used to identify traits associated with underreporting, but the results have been conflicting and have made only a limited contribution to understanding underreporting behavior. Questionnaires derived from the area of eating disorders, including the Dutch Eating Behavior Questionnaire (DEBQ) (111) with scales for restraint, emotional and external eating, and the Three Factor Eating Questionnaire (TFEQ) (112) with scales for restraint, disinhibition and hunger, have been used. Bingham et al. (113) found underreporters to be significantly different from others on both restraint scales but not on disinhibition or hunger scales. One study (64) found a modest degree of underreporting in restrained compared to unrestrained eaters, and two studies (59,114) found no significant difference. Lindroos et al. (115) observed no associations between EI and any TFEQ score in normal weight women but identified a positive association between EI and both disinhibition and hunger in the obese group. There was no significant association between EI and restraint across the obese group as a whole, but those in the highest quartile for restraint score did report the lowest EI.

Other cognitive, perceptual and emotional components of dietary reporting behavior have been explored. Significant associations have been found between LER at age 43 y, extroversion scores at age 26 y and recent emotional problems (35). No associations were found with neuroticism scores at age 26 y, other depressive symptom scores relating to the past year or symptoms of a clinically recognizable mental disorder (35). Kretsch et al. (13) reported an association between underreporting and the Beck depression inventory in obese but not lean women. Taren et al. (116) explored the most comprehensive battery of tests. They examined the associations between reporting accuracy and social desirability (108), two restraint items (117,118), eight items from the Eating Disorder Inventory (119,120), two items from the Weinberger Adjustment Inventory (121,122) and two elements of the Sorenson-Stunkard Silhouettes. However, they found that age and percentage body fat were the most important predictors of underreporting. After correcting for these variables, there were associations between underreporting and only three variables: social desirability score, body dissatisfaction and perception that a smaller body image is healthy. Overall it appears that questionnaires designed for other purposes may not help substantially in investigating the issue of misreporting. These limited findings suggest that research needs to be focused very specifically on behaviors connected with food and the process of dietary assessment.

The foregoing discussion highlights the complexity of the phenomenon of misreporting and the difficulty of drawing meaningful conclusions from existing work. In an analysis of the most comprehensive set of variables (anthropometric, social and psychological) studied in the UK 1946 birth cohort (35), it is notable how few factors were significantly associated with LER apart from low energy reporting at an earlier phase of the study (1982) and parameters of body weight. The first association was the stronger (see Subject-specific bias to dietary assessment). When low EI:BMR in 1982 was included in the logistic regression model, only previous underreporting, ln(BMI) and, in women, social class at age 4 y and "currently in paid work" remained significant predictors of underreporting. In NHANES III, which also investigated a wide range of variables (37), the five main factors, which accounted for 30% of the variance in EI:BMR, were the total number of foods reported, weight status, age, trying to lose weight and smoking.

	Urinary nitrogen excretion in relation to reported energy intake

TOP ABSTRACT Energy intake for weight... Direct comparison of energy... Validation against presumed... Urinary nitrogen excretion in... Consequences of underreporting Some issues of interpretation... Can data be adjusted... Future directions LITERATURE CITED

 
Because validation by urinary nitrogen excretion (123,124) indicates underreporting of protein intake, by definition there will also be underreporting of EI. However, the converse may not be true. There is evidence that protein is better reported than total EI (discussed in Bias in estimating nutrient intake). In double validations that used urinary nitrogen excretion and DLW-EE (57,58,62), mean reporting bias was -2% for protein and -14% for energy. For individuals there was a significant correlation between the validation ratios EI:EE and the urinary nitrogen/dietary nitrogen ratio, but the proportion of subjects identified as underreporters was greater by DLW than urinary nitrogen excretion. It was unclear whether this was a real phenomenon or instead dependent on the criteria for each validation. In another double validation, however, Kroke et al. (36) found similar mean underreporting of protein (23%; n = 134) and energy (22%; n = 34). If protein or nutrients highly correlated with protein intake are of primary interest, then validation by urinary nitrogen excretion must be the method of choice. If energy or macronutrients are of primary interest, then validation of EI together with critical examination of the macronutrient intake are essential.

Several studies have used urinary nitrogen excretion to validate the diet-history technique. These have been reviewed recently (62,201). Urinary nitrogen has also been used to validate the FFQ as discussed by Bingham (201).

Urinary nitrogen excretion has scope for the validation of ranking of individuals. However, it should be noted that eight collections are needed to obtain a precise measure of urinary nitrogen excretion in individuals (123), and that validation of the collection itself is also essential. As many as 25% (Bingham, personal communication) and 17% (125) of collections have been either not received or rejected as incomplete as judged by para-aminobenzoic acid recovery (126).

	Consequences of underreporting

TOP ABSTRACT Energy intake for weight... Direct comparison of energy... Validation against presumed... Urinary nitrogen excretion in... Consequences of underreporting Some issues of interpretation... Can data be adjusted... Future directions LITERATURE CITED

 
Bias in estimating nutrient intake.

Validation against EE identifies only the bias in reporting EI. Tables 6 and 7 list studies that have examined whether this reflects underreporting of the diet as a whole or whether there is bias in estimating nutrient intakes through altered food choices and/or selective reporting of foods. With the caveat that these data were derived from studies using different dietary methods and different criteria were used to define LER, the summary (Table 7) shows that the percentage of energy from protein was significantly higher (P < 0.000) in the LER, whereas fat energy (percent) tended to be lower (P < 0.004) and total carbohydrate energy (percent) was variable (not significant). In six studies (nine subgroups) where data on the percentage of energy derived from starches and sugars were available, starch energy tended to be higher in LER (P < 0.002) but sugars energy was lower (P < 0.03). It is evident that bias in reporting total EI is associated with variable bias in estimated macronutrient intake.

The significant differences in micronutrient density between LER and non-LER as tested in the original papers are also shown. In 52 (58%) out of 89 tests, the differences were significant. Out of 119 ratios, 105 (88%) were > 1.0, which indicates that the diet of LER had a higher nutrient density than that of the non-LER. The mean ± SD percentage difference (excluding the exceptionally high value for ascorbic acid in Study 6) was 8.8 ± 12.1%. Thus bias in reporting total EI is associated with variable bias in reporting nutrient intake.

Bias in reporting meal patterns and foods eaten.

Differential reporting of nutrients must be a consequence of differential reporting of foods, but fewer studies have examined this issue. Table 8 summarizes these studies (35,101,102,113,127–129). There was a general tendency for LER to report more "good" foods such as meat, fish, vegetables, salad and fruit and less "bad" foods such as cakes, cookies, sugar, candies and fats. However, this could simply reflect the fact that main food items tend to be remembered better than ancillary items as demonstrated by Poppitt et al. (114) in a direct validation of food intakes reported by 24-h recall.

In conclusion, underreporting of food intake is a selective rather than a general phenomenon. The evidence points to differences in reporting for macro- and micronutrients, foods and meal patterns. The reasons for such misreporting are clearly complex and operate in different ways in different people. Possible factors include the general climate of knowledge about food and health, perceived reasons for the study, personal image management and the unconscious messages conveyed by the researchers themselves. The latter has not been researched at present. Undoubtedly a major advance has been the growing recognition that cultural, behavioral and psychological factors underpin dietary reporting behavior. If nutritionists are to fully understand the impact of these factors on the dietary reporting process, collaboration with behavioral scientists is essential in the future.

	Some issues of interpretation raised by misreporting

TOP ABSTRACT Energy intake for weight... Direct comparison of energy... Validation against presumed... Urinary nitrogen excretion in... Consequences of underreporting Some issues of interpretation... Can data be adjusted... Future directions LITERATURE CITED

 
Obesity, meal patterns and macronutrient intake: an example of interpretative problems.

Obesity-related issues provide good illustrations of the interpretative problems associated with misreporting. The hypothesis that obese people were energy thrifty derived from studies in the 1960s that found the EI of obese subjects to be similar to or lower than their lean counterparts. This led to nearly two decades of research to elucidate the hypothesized defect in the EE of obese persons. None was found. With the advent of DLW, it became clear that the hypothesis had been based on invalid reports of food intake by obese persons recruited for studies of obesity (44,54,75).

The study of links between obesity and diet composition has run into similar problems. For example, the Scottish Heart and Health Study (133) reported the highest prevalence of obesity in subjects in the top fifth of fat energy (by percentage) and the lowest prevalence in the top fifth of sugars energy (by percentage). However, this conclusion is confounded by three facts. First, because the percentage energy derived from protein and starch is relatively constant, there is a reciprocal relationship between fat and sugars when expressed as percentage energy (134). Second, there is an association between obesity and low energy reporting (see EI:BMR: The Goldberg cutoff technique) and third, there is a possible association between LER and underreporting of sugars (see Bias in estimating nutrient intake). Flynn et al. (134) found a positive association between BMI and fat intake as assessed by diet history when dieters were excluded from the analysis. On the other hand, Macdiarmid et al. (135) found their conclusions differed depending on which subjects were included or excluded from the analysis.

Conclusions about obesity and meal patterns may also be confounded by associations between obesity/high BMI, low meal frequency and underreporting (136). Bellisle et al. (137) used data from Kant et al. (138) to demonstrate the probable artifactual nature of the association between low meal frequency and BMI. As reported meal frequency decreased, the gap between EI and a presumed energy requirement of 1.4 x BMR increased. In adolescents (139), a negative association between BMI and feeding frequency also disappeared when dieters, underreporters and normal-weight subjects who considered themselves to be overweight were excluded.

Estimating the proportion of a population that has deficient nutrient intake.

Extremes of the nutrient-intake distribution, particularly those derived from short records or 24-h recalls, do not represent habitual intake. Consequently, even a valid data