Accuracy of Self-Reported Energy Intakes in Low-Income Urban 4th Grade Minority Children

Abstract

We examined the accuracy of self-reported energy intake (rEI) in low-income, urban minority school-aged children at risk for obesity and associated diabetes utilizing a relatively new, simple previously published prediction equation for identifying inaccurate reports of dietary energy intake. Participants included 614 nine-year-old boys (51%) and girls (49%). Three 24-h dietary recalls were collected. Children’s height, weight (used to calculate BMI), and percent body fat (%BF) were measured. Physical fitness, reported family history of diabetes, and ethnicity were also collected. A previously published prediction equation was used to determine the validity of rEIs in these children to identify under-, plausible-, and over-reporters. Additionally, we examined the question of whether there is a difference in reporting by sex, ethnicity, BMI, and %BF. On average, 18% of the children were at risk of being overweight, 43% were already overweight at baseline, yet these children reported consuming fewer calories on average than recommended guidelines. Additionally, reported caloric intake in this cohort was negatively associated with BMI and %BF. Using the previously described methods, 49% of participants were identified as under-reporters, whereas 39 and 12% were identified as plausible- and over-reporters, respectively. On average, children reported caloric intakes that were almost 100% of predicted energy requirement (pER) when the sedentary category was assigned. Inactivity and excessive energy intake are important contributors to obesity. With the rising rates of obesity and diabetes in children, accurate measures of energy intake are needed for better understanding of the relationship between energy intake and health outcomes.

INTRODUCTION

It is widely recognized that self-reported energy intakes (rEIs) in dietary surveys (i.e., dietary recalls) tend to underestimate actual energy intake, where individuals of different age, sex, race/ethnicity, and socioeconomic backgrounds typically underreport by 10–54% (1–11). Most of the evidence about under-reporting comes from measurements of total energy expenditure (TEE) by the doubly labeled water method (4,12). The doubly labeled water method is a technique that measures TEE in free-living individuals (12). In weight-stable individuals, energy intake must equal TEE, and therefore, accuracy of food intake reporting can be assessed by comparing reported energy intake to TEE (13). Although doubly labeled water is the standard approach for comparing reported energy intake to TEE, this particular method is costly and not feasible for large samples (7,12).

Despite the limitations of the 24-h dietary recall, it is widely used in studies of both free-living adults (7,8,13) and children (6,10,11,14–16) because it is relatively inexpensive and easy to administer in many settings (i.e., school, home, workplace, church). Methods have therefore been developed to “screen” for implausible reporters of energy intake in free-living persons (2,7,8,13). These methods allow for the accuracy of dietary recall data to be assessed. These methods replace direct measures of energy expenditure with predictions of energy expenditures. Briefly, the approach involves comparing reported energy intake to predicted energy expenditure, which is calculated from age, weight, height, and sex (7). Expected associations between reported energy intake and health outcomes can emerge when reports identified as probably inaccurate are excluded by this method (7–9). This approach has been used in adults’ diet recall data (7,8), but its use in children remains limited (9–11,17,18). One publication reported the accuracy of two dietary recalls from the Continuing Surveys of Food Intakes by Individuals 1994–1996 and 1998 in 1,995 US boys and girls (without any race or ethnicity identified) (9). The main results of this epidemiological study showed that implausible dietary reports were prevalent in childhood and adolescence (55% of total sample). In the second study, data were examined from three dietary recalls of 176 eleven-year-old white, non-Hispanic, and middle-class girls of relatively well-educated and affluent families (10). As in Huang et al. (9), this particular study reported 50% of children’s energy intakes as implausible. The third study, which examined energy under-reporting from a 3-day average of dietary recalls of 284 eight-to ten-year-old African-American (AA) girls (11), showed that more than half (54.8%) of the AA girls under-reported.

Until the present study, specific information concerning the reporting bias in dietary recall data of low-income Mexican-American (MA) children has not been available in the literature. Therefore, the present study applies the methods developed by Huang et al. (9,10) to better understand the accuracy of reported energy intake in low-income minority children, and particularly MA children.

METHODS AND PROCEDURES

Participants

Participants were 711 nine-year-old, low-income boys and girls, living in San Antonio, TX and enrolled in Bienestar, a School-based Diabetes Prevention Program (15). Bienestar was a randomized, controlled trial, and the children used in this study were enrolled in the control arm from (start year to end year). Children were excluded from the analysis if they did not complete three nonconsecutive 24-h dietary recalls (n = 44) or did not have records of both weight and height data (n = 53), resulting in a total sample size of 614. Only the baseline Bienestar data in 2001 were used in this analysis. Demographic data were obtained from parents and the schools. All procedures were reviewed and approved by the University of Texas Health Science Center Institutional Review Board.

Procedures

Collection of food intake data.

Diet assessments were collected at the beginning of the 4th grade school year. Data collection occurred from 20 August 2001 through 31 September 2001. Three 24-h dietary recalls were performed face-to-face by trained staff in English, or Spanish when needed, during school hours on the day after the day of interest. Bienestar staff training focused on interviewing and quantifying techniques. The multiple-pass interviewing methodology consisted of a scripted dialogue, food probing guides, and recording methods. Food models and measuring utensils were used to improve portion size estimation. Although parents were not present at the dietary interviews, parents were available to the interviewers when further clarification of the child’s dietary food intake recall was necessary, that is, when the child did not remember what he/she ate or requested the parent’s presence during the interview. Less than 5% of parents were contacted for those reasons. Because 97% of participants were on federally assisted free- or reduced-cost meal program, school meal menus were also used to verify dietary intake of children reporting eating breakfast and/or lunch at school.

Dietary intake data were entered from written records of the food recall data into a food and nutrient calculation software program (Nutrition Data System for Research (NDSR), version 4.02; Nutrition Coordinating Center, University of Minnesota, Minneapolis, MN). The NDSR database was customized in collaboration with the University of Minnesota to incorporate ethnic food items commonly consumed by the Bienestar participants but not found in the NDSR food database. A list of the common foods was obtained by Bienestar staff from focus group feedback provided by a separate, yet similar group of 9-year-old children and food cafeteria staff who were independent of the Bienestar study, that is, they worked for a district outside of the targeted schools. Bienestar staff piloted the NDSR customized database with another independent group and concluded that the database was appropriate and culturally sensitive for the target population prior to using it in this study.

Measures

Anthropometrics.

A registered nurse or trained medical assistants collected all anthropometric data: height, weight, and percent body fat (%BF). All children wore indoor clothing at the time of testing and were asked to remove their shoes and socks for these measurements. Height was measured to the nearest 0.5 cm with a stadiometer (Seca Body-meter 206; Seca, Hanover, MD). Weight was measured to the nearest 0.1 kg with an electric scale; and %BF was measured using bioelectric impedance analysis (Tanita, Arlington Heights, IL). BMI was calculated as weight in kilograms divided by the square of height in meters using the Quetelet index measure (9).

Physical fitness. Physical fitness was measured using a modified Harvard Step Test (15,19). Physical activity (PA) level was not assessed.

Identification of under-, plausible-, and over-reporters

In this study, the method developed by Huang et al. (9,10) was used to identify under-, plausible-, and over-reporters as used by Ventura et al. (10). Specifically, sex- and age group–specific ±1 s.d. cut points were created for rEI as a percentage of predicted energy requirement (pER): (rEI/pER) × 100. These methods and cut points are explained in more detail elsewhere (7–9). Briefly, the equation for pER was obtained from the 2002 Dietary Reference Intakes (12) and used for both girls and boys: pER = gender constant − 30.8 × age (years) + PA × (10 × weight (kg) + 934 × height (m)) + (kilocalories for energy deposition). This equation includes age, weight, height, and PA levels (PALs; sedentary, low active, moderately active, and highly active) and a constant representing energy deposition for growth. Bienestar did not measure PA but rather assessed physical fitness using a modified Harvard Step Test. Therefore, due to the lack of an objective, validated PA measure in our sample, the Huang et al. (9) study suggested PAL value of 1.5 was used. That is, we assigned all boys and girls to the low-active category of PAL (i.e., PAL between 1.4 and 1.6; a PAL of 1.5 is the equivalent of walking 3 miles a day for an individual weighing 44 kg) (12).

Cut point at ±1 s.d.

Cut points were also created using the established equations developed by McCrory et al. (7,9). Specifically, after calculating pER for each participant, rEI was divided by pER and multiplied by 100 to provide an estimate of plausibility for rEI as a percentage of pER (%rEI/pER).

Propagation of error variance was utilized to create the ±1 s.d. cut point for %rEI/pER. This was calculated with the Goldberg cut-point equation that was then adopted by Huang et al. (8,9):

$$1\text{s}.\text{d}.=\sqrt{\frac{{\text{CV}}_{{\text{rEI}}^2}}{\text{d}}+{\text{CV}}_{{\text{pER}}^2}+{\text{CV}}_{{\text{mTEE}}^2}}$$

where ${\text{CV}}_{\text{rEI}}$ is the average, expressed as %, of within-subject coefficients of variation for reported energy intake across the three reported days; d is the number of days of intake data; ${\text{CV}}_{\text{pER}}$ is the coefficient of variation for pER; and ${\text{CV}}_{\text{TEE}}$ is the day-to-day biological variation in TEE. In this study, d was equal to 3 days of dietary intake data, CV_rEI was 33%, and ${\text{CV}}_{\text{pER}}$ was sex specific, 4.2% for boys and 4.8% for girls. The calculations for each component of the equation listed above and used to determine the ±1 s.d. cut point are the same as used by others and described in more detail elsewhere (8–10). The ${\text{CV}}_{\text{mTEE}}$ was measured in previous studies with doubly labeled water techniques (2,7) and we used the value of 8.2%.

Data analysis

SPSS 16.0 (SPSS, Chicago, IL) was used to analyze the quantitative data. P < 0.05 was used to indicate significant effects. ANOVA was used to compare mean differences of reporting (rEI/pER) by sex and race/ethnicity. Linear regression modeling compared differences in reporting by weight status (BMI) and adiposity (%BF); and post hoc pairwise comparisons of significant differences were computed with a Tukey’s honestly significance test to correct for error rates at P < 0.05. Bivariate correlation for weight status (BMI), adiposity (%BF), and reported total kilocalories were also performed.

RESULTS

Background characteristics

Participant characteristics are presented in Table 1. Of note, these were primarily MA minority children. Based on questions asked of parents, the participants were from low-income families residing in inner-city neighborhoods. Nearly three in four parents (72%) reported that they had less than a high school diploma and 28% of households reported that none of their residents were employed. According to the information provided by the school district about the participating elementary schools, socioeconomic status was confirmed with families reporting on average 3.5 persons per household and over 95% participating in US Department of Agriculture (USDA) food assistance programs.

Table 1.

Characteristics of participating 9-year-old boys and girls (N = 614)

Age (years) (mean ± s.d.)	9.7 ± 0.45
Sex (%)
Boys (n = 308)	51%
Girls (n = 306)	49%
Ethnicity (%)
Mexican American (n = 478)	78%
African American (n = 79)	13%
Asian (n = 36)	6%
Other ethnic group (n = 21)	3%
Physical fitness (%)
Acceptable	13%
Marginal or unacceptable	87%
Reported energy intake a (total kcal) (mean ± s.d.)
All	1,630 ± 610
Boys	1,700 ± 650
Girls	1,550 ± 570
Reported energy intake by ethnicity (mean ± s.d.)
Boysb
Hispanic	1,700 ± 660
African American	1,750 ± 660
Asian	1,690 ± 670
Other ethnic group	1,650 ± 630
Girlsc
Hispanic	1,490 ± 520
African American	1,920 ± 740
Asian	1,390 ± 430

US guidelines/Dietary Reference Intakes recommended:

^a1,850 (sedentary) – 1,950 (active), Source: ref. 12;

^b1,800 (sedentary) – 2,200 (active), Source: ref. 12;

^c1,850 (sedentary) – 1,950 (active).

On average, the study’s 9-year-old children had higher BMIs than the 85th percentile of the CDC (Centers for Disease Control and Prevention) normative growth charts (see Table 2). Approximately one in five (18%; 112/614) of children were at risk of being overweight (BMI ≥85th–95th percentile) and 38% (233/614) were already overweight (BMI ≥95th percentile). These children also had higher %BF than US guidelines. There were no statistically significant differences in BMI and %BF by ethnic group.

Table 2.

Anthropometric data (N = 614)

	Mean ± s.d.	US guidelinesa
BMI
Boys	20.6 ± 4.8	17.2
Girls	20.3 ± 4.9	17.4
%BF
Boys	26.0 ± 11.5	14
Girls	27.4 ± 10.6	20
By ethnicity
BMI
Hispanic	20.6 ± 4.9
African American	19.9 ± 4.5
Asian	20.9 ± 4.9
Other ethnic group	19.8 ± 4.9
Total	20.3 ± 4.6
%BF
Hispanic	27.0 ± 11.1
African American	24.9 ± 10.6
Asian	27.4 ± 11.3
Other ethnic group	24.9 ± 12.3
Total	26.7 ± 12.3
Exceeding 85th percentile BMI
	BMI ≥85th–95th percentile	BMI ≥95th percentile
Boys	16% (49/308)	41% (127/308)
Girls	21% (63/306)	35% (106/306)
Total	18% (112/614)	38% (233/614)
Hispanic	14% (86/614)	31% (189/614)
African American	2% (14/614)	4% (23/614)
Asian	2% (10/614)	2% (13/614)
Other ethnic group	0.3% (2/614)	1% (8/614)

^aSource: ref. 28.

Identification of under-, plausible-, and over-reporters

As indicated in Table 1, these children reported consuming fewer calories on average than the recommended guidelines regardless of ethnic group affiliation, already suggesting under-reporting on average. The total sample mean for %rEI/pER, which was calculated using PAL 1.5 for low-active PAL category, was 86.0%. That is, on average, the reported caloric intake was 14% below the predicted requirement. Figure 1 illustrates the frequency distribution of %rEI/pER values in this group. Using the ±1 s.d. cut point, in this group of 9-year-old boys and girls, 49% (n = 300) of participants were identified as under-reporters, whereas 39% (n = 241) and 12% (n = 73) were identified as plausible- and over-reporters, respectively. Using the more inclusive ±2 s.d. cut point, 18% of these children (n = 110) were identified as under-reporters, whereas 77% (n = 473) and 5% (n = 31) were identified as plausible- and over-reporters, respectively.

Figure 1.

Reporting accuracy (ratio of rEI to pER) with PAL 1.5 (low active). PAL, physical activity level; pER, predicted energy requirement; rEI, self-reported energy intake.

Differences in rEI/pER by sex and ethnicity

Results for rEI/pER ratio (at ±1 s.d.).

In these children, there were no statistical significant differences in %rEI/pER by sex (P = 0.369). The mean %rEI/pER was 85 and 88% for boys and girls, respectively. In contrast, there were statistically significant differences in reporting by ethnicity (P = 0.004). The MA group was significantly different from the AA group (P = 0.003), but not from the Asian-American (P = 0.845) or other group (P = 0.997). Specifically, the mean for MA %rEI/pER was 84% compared to 99% for AA. The Asian-American group mean of 79% approached significant differences with the AA group (P = 0.026), but not with the MA (P = 0.845) or other group (P = 0.986).

Differences in rEI/pER by BMI and %BF

In the study, there were statistically significant differences in reporting by BMI (P = 0.001), such that energy intake and BMI were inversely related. That is, children with the highest BMI reported consuming the fewest calories and vice versa. Similarly, there were significant differences in reporting by mean %BF (P = 0.001), again illustrating an inverse relationship: children with the highest %BF reported consuming the fewest calories and vice versa. Correlation was significant at the 0.01 level (two-tailed) (Figures 2 and 3).

Figure 2.

Correlation between rEI/pER and BMI. pER, predicted energy requirement; rEI, self-reported energy intake

Figure 3.

Correlation between rEI/pER and percent body fat (%BF). pER, predicted energy requirement; rEI, self-reported energy intake.

Differences in rEI/pER based on PA assumptions

Less than one in eight (13%) of these children had physical fitness at an acceptable level based on the physical fitness assessment reported in detail elsewhere (15). Although PA was not directly assessed, one possible explanation for the high frequency of apparent under-reporting of caloric intake by these children is that they were in fact so inactive—as suggested by the very poor fitness levels—that they needed even less energy intake than the pER calculated here. Therefore, the analyses were repeated using a PAL value of 1.0 (PAL value ≥1.0 and ≤1.4 = sedentary category), which would imply a resting, bed-ridden status. For PAL 1.0 (PA coefficient used of 1.0 for both sexes) and ±1 s.d. cut point, the plausible rEI/pER range was 79–121%. In this analysis, one in three children (32%) would still be categorized as under-reporters. The total sample reporting mean was 100.7% (±41.8). That is, on average, reported caloric intake was almost 100% of pER. In the total sample, the lowest reporting was 19.10% and the highest was 428.50%.

DISCUSSION

In this study, we examined the accuracy of rEIs in low-income, urban 4th grade MA minority children at risk for obesity and associated type 2 diabetes mellitus utilizing a previously published prediction equation method for identifying inaccurate reports of energy intake. Using this method and the same cut points as others (6,10), nearly half of the children (49%) were identified as likely under-reporters, whereas 39 and 12% were identified as plausible- and over-reporters, respectively. Our study supports the findings of others (6,10) that children under-report caloric intake.

We also found that, in this group of children, there were statistically significant differences in reporting by ethnicity (P = 0.004). There appears to be racial differences in under-reporting where MA and Asian children on average under-report, but AA children do not. These findings suggest energy intakes of MA and Asian children should be interpreted with caution especially when associations between diet and related factor are examined in these groups. In this study, we did not examine factors contributing to these differences nor whether better dietary recall instruments or improved techniques would promote more accurate reporting in these subgroups. An explanation of these differences may help explain why our findings differ regarding under-reporting among AA from those reported by Lanctot et al. (11) who find that 54.8% of AA girls under-reported energy intake. To our knowledge, there are no other published findings from studies examining these relationships, and therefore, racial differences in under-reporting of energy intakes in children warrant further study.

We also found statistically significant differences in reporting by mean BMI (P = 0.001). That is, children with the highest BMI reported consuming the fewest calories and vice versa. Similarly, there were significant differences in reporting by mean %BF (P = 0.001) where children with the highest %BF reported consuming the fewest calories and vice versa. Although the majority of these children were already overweight (BMI >95th percentile) at age nine, they reported consuming fewer calories, on average, than the recommended guidelines. One possible explanation for why the heavier children tended to under-report their caloric intake could be more attributable to physical inactivity than energy intake. Recent studies used data on individuals age ≥2 years from four nationally representative surveys of the US populations: (i) Nationwide Food Consumption Survey (NFCS77); (ii and iii) Continuing Survey of Food Intake by Individuals (CSFII 89 and CSFII 96), and (iv) USDA. Measuring intake largely through an in-home, interviewer-administered 24-h dietary recall and/or self-report, these studies report that US children have decreased their total energy intake over the past decade (20); yet, the prevalence of childhood obesity keeps rising. Physical inactivity may help to explain this discrepancy in the rising rate of obesity and the lower reported energy intake (21). The data in the present study may support, in part, this relationship. With the ±1 s.d., using the sedentary category (PAL 1.0), the mean reported caloric intake was almost 100% of the pER compared to the reported caloric intake of 13.6% below the pER when using the low-active (PAL 1.5) category. When the sedentary category for PAL was used, the mean reporting plausibility in our study (100%) is the same as that for children in the same age range reported by Huang et al. in the US national survey data from 1994 to 1995 (~15 years ago) who used the low-active category for PAL (9). Although it may not be unreasonable to assume these children are in the same age range have a reporting plausibility of 100%, it is unlikely assigning all children to a sedentary category is justifiable. When the sedentary (PAL 1.0) category was assigned to these children, approximately one in three (32%) children under-reported, compared to approximately one in two (47%) children in the low-active category. The exercise of calculating reporting plausibility using PALs consistent with either sedentary or low-active category illustrates the importance of measuring PAL in future studies so that dietary intake plausibility can be assessed with greater certainty. Even though we had an objective measure of physical fitness, only 13% of these children were considered to be physically fit, and there are no methods for using fitness level to assign PAL categories.

There are several considerations that may be worth noting that could, in part, help explain the higher degree of underreporting seen in this study compared to some others (7,8,10). First, in this study, the children were predominantly from low-income families. Low-income children (and their families) have been documented to be at increased risk of experiencing food insecurity and food insufficiency (22–24), so that a child may not accurately represent his/her usual intake. Food insufficiency could be dependent on food assistance or other assistance such that food intakes could be higher or lower depending on food availability. Capturing these differences in intake might be difficult within the context of 3 days of intake. There was considerable variability among the days of intakes shown by the high coefficients of variation. Whether more days of intake in this population would reduce the within-person variability and be more reflective of usual intake warrants further investigation. Highs and lows in energy intake may promote weight gain or weight loss and possibly affect the weight stability of a child; as a result, the fundamental principles of these equations would not be applicable to these children. It is, however, unlikely that weight instability would be an issue within the short data collection period in this study.

Most studies examining energy intake in school-aged children interview in the presence of the parent (6,10). Even though attempts were made to obtain additional details from parents after the interviews, it might have been impossible for parents to remember those details. Compared to the face-to-face method, a telephone method (10) may be less threatening to participants, so that they may answer more freely because they generally do not know the interviewers. This method may result in less response bias, that is, participants may be less prone to provide “socially acceptable” responses. Second, this study used a multiple-pass method with three passes through a day to improve memory or recall of foods eaten on the previous day (25). Although there are no published studies showing that one method is “more” valid than the other, the USDA five-step multiple-pass method has been shown to assess intake within 10% of actual intake (25).

Another important methodological distinction between this study and others is that dietary recall data were collected on paper forms and then entered into the NDSR software program to generate the nutrient and energy intakes. This is an important distinction because NDSR provides a guided, structured, interactive interface that contains standard probes and cues to increase the level of detail about each food item reported. Even with extensive training and probing guides, it would be impossible for an interviewer to obtain the amount of detail needed to enter a particular food and to guess at all the potential probes for the detail that the software provides. In case of missing details or amounts, the NDSR software defaults to the most common detail for that food. This important feature standardizes difficult food entry decision but could also result in an under- or overestimation of the actual food consumed. These methodological differences between studies highlight the need for further research to identify important contributors to the under- or overestimation of energy intake.

To our knowledge, this is the first study examining the accuracy of rEI in low-income MA children utilizing a comparison to pER. Thus, the finding that one in two minority children may be inaccurately reporting their food intake is alarming. Because these children already have one or more of the risk factors associated with obesity and associated type 2 diabetes mellitus at the age of nine (14,15,24,26,27), methods to improve accurate reporting are needed to study the current epidemics of obesity and diabetes in children (27).

Conclusion

In this study, we examined the accuracy of rEI in low-income, urban, MA, school-aged children at risk for obesity and associated type 2 diabetes mellitus and found that a significant proportion of these children under-report. The observation in this study that obesity is associated with lower reported caloric intake is consistent with previous findings that weight status is an important consideration in understanding the degree of under-reporting of energy intake. Additionally, this study suggests that inactivity and/or excessive energy intake are important contributors to obesity. The differences in underreporting by assigning different PALs emphasize the need for accurate measures of PA. With the rising rates of obesity and diabetes in children, accurate measures of energy intake and expenditure are needed for a more thorough understanding of the relationship between dietary intake and health outcomes.