Childhood Overweight Prevalence in the United States: The Impact of Parent-reported Height and Weight

Authors

  • Lara J. Akinbami,

    Corresponding author
    1. Infant, Child and Women's Health Statistics Branch, National Center for Health Statistics, Centers for Disease Control and Prevention, Hyattsville, Maryland, USA
    2. US Public Health Service, Rockville, Maryland, USA
    Search for more papers by this author
  • Cynthia L. Ogden

    1. Division of Health and Nutrition Examination Surveys, National Center for Health Statistics, Centers for Disease Control and Prevention, Hyattsville, Maryland, USA
    Search for more papers by this author

(lakinbami@cdc.gov)

Abstract

Parent-reported height and weight are often used to estimate BMI and overweight status among children. The quality of parent-reported data has not been compared to measured data on a national scale for all race/ethnic groups in the United States. Parent-reported height and weight for 2–17-year-old children in two national health interview surveys—the 1999–2004 National Health Interview Survey (NHIS) and the 2003–2004 National Survey of Children's Health (NSCH)—were compared to measured values from a national examination survey—the 1999–2004 National Health and Nutrition Examination Survey (NHANES). Compared to measured data, parent-reported data overestimated childhood overweight in both interview surveys. For example, overweight prevalence among 2–17-year-olds was 25% (s.e. 0.2) using parent-reported NHIS data vs. 16% (s.e. 0.6) using measured NHANES data. Parent-reported data overestimated overweight among younger children, but underestimated overweight among older children. The discrepancy between reported and measured estimates arose mainly from reported height among very young children. For children aged 2–11 years, the mean reported height from NHIS was 3–6 cm less than mean measured height from NHANES (P < 0.001) vs. no difference among children aged 16–17 years. Measured data remains the gold standard for surveillance of childhood overweight. Although this analysis compared mean values from survey populations rather than parent-reported and measured data for individuals, the results from nationally representative data reinforce previous recommendations based on small samples that parent-reported data should not be used to estimate overweight prevalence among preschool and elementary school–aged children.

Introduction

Increasing prevalence of childhood overweight (BMI ≥95th percentile for age and gender) has focused attention on monitoring local trends (1). Periodic surveys of measured weight and height in the United States show that between 1971–1974 and 2003–2006, overweight increased from 5 to 12.4% for 2–5-year-olds, from 4 to 17% for 6–11-year-olds, and from 6 to 17.6% for 12–19-year-olds (2,3).

Height and weight data are needed to calculate BMI (weight in kilograms divided by height in meters squared) to assess weight status according to the Centers for Disease Control and Prevention growth charts (4). Measured height and weight are the gold standard for surveillance, but measurements are expensive to obtain and rarely available for local populations or annually. Parent-reported data is an attractive alternative. However, among adults and adolescents, self-reported height and weight underestimate BMI and obesity rates compared to measured data (5,6,7,8,9,10,11,12,13,14,15). The validity of overweight prevalence estimates based on parent-reported data for younger children is less clear because findings vary (16,17,18). Nevertheless, the imperative to assess overweight prevalence among younger children has increased with research demonstrating that being overweight early in life increases the likelihood of being an overweight adolescent (19).

To assess the quality of parent-reported data for childhood overweight surveillance, we compared data from two national surveys that collect parent-reported height and weight data for children to a third survey that collects measured data during physical examinations. All three surveys are routinely used in national health and disease surveillance and provide an opportunity to compare the performance of reported vs. measured data collected in standardized protocols among nationally representative samples selected using similar sampling methodology.

Methods and Procedures

The National Health Interview Survey (NHIS) is a continuous national survey of a representative sample of the civilian noninstitutional population in the 50 States and the District of Columbia using a complex multistage probability design (20). Interviews in selected households are conducted for a randomly selected sample adult and, if children are present in the family, a sample child. For children <18 years of age, a knowledgeable adult (usually a parent) reports health conditions. When reporting child weight and height, the parent is not instructed to measure the child or consult records. Parent-reported weight and height are available on restricted use files, but are not publicly released. The response rate for the sample child component (the completion rate among eligible children multiplied by the overall family response rate) was 79–81% during 1999–2004 (21).

The National Survey of Children's Health (NSCH) is a module of the State and Local Area Integrated Telephone Survey conducted in 2003–2004 with future periodic fielding planned. The survey produces national and state-level estimates for a variety of health indicators and measures of children's health care experiences. A random-digit-dial sample of households with children <18 years of age was selected from each State and the District of Columbia. A knowledgeable adult (usually a parent) responded for a randomly selected child in each household and, similar to NHIS procedures, no prompt was given to measure the child or consult records before reporting weight or height. The Council of American Survey Research Organizations response rate (the product of the resolution rate, the screener completion rate, and the interview completion rate) was 55.3% (22).

For both the NHIS and NSCH, respondents were asked: “How tall is [child name] now?” and “How much does [child name] weigh now?” However, the NHIS respondents were also prompted to report weight without shoes. The NHIS height and weight data were not edited before release of restricted files. Children whose parents did not know or refused to provide height and/or weight were excluded (14%). We used the World Health Organization exclusion criteria (z-score for gender-specific BMI-for-age was <−4.0 or >+5.0) (23) to exclude records with implausible parent-reported height and weight (6.3%) from the main analysis for a final sample size of 54,325 children and teens 2–17 years of age. In a sensitivity analysis, however, we included records with implausible height and weight values to assess the actual data collected by professional interviewers using a well-tested survey instrument. In contrast, the NSCH height and weight data were edited before public release (22) and extreme values were suppressed to decrease disclosure risk. The NSCH file includes a precalculated BMI for respondents with nonmissing and unsuppressed weight and height, and nonmissing sex. Of the NSCH sample aged 2–17 years, 8.5% was excluded for missing values for height and/or weight (respondent did not know or refused to answer), and 0.2% for suppression of extreme values so that overall, 8.7% had missing values for precalculated BMI. The sample size for the 2003–2004 NSCH was 82,390.

These data were compared to measured data for children 2–17 years of age from the National Health and Nutrition Examination Survey (NHANES). The NHANES collects measured height and weight data through standardized physical examinations in mobile examination centers. A nationally representative sample of respondents are selected from the United States civilian noninstitutional population in the 50 States and the District of Columbia using a complex multistage probability design (24). For 1999–2004, unweighted response rates for the examination portion of the survey for children 1–19 years of age ranged between 81 and 88% (25). Pregnant girls and those with missing measured weight and height data were excluded (<4%), a routine exclusion that was not possible with the NHIS or NSCH because pregnancy status was not assessed in those surveys. The final NHANES sample size was 12,261.

Weight status was defined according to BMI percentile for age and gender from the 2000 Centers for Disease Control and Prevention growth charts (4,26). Overweight was defined as BMI ≥ 95th percentile for age and gender. To calculate BMI percentiles for the NHIS sample, we used an online SAS program provided by Centers for Disease Control and Prevention which defines percentiles based on months of age (23). Because age was reported in whole years for the NHIS sample, we categorized children according to the midpoint of their age-year (e.g., 2-year-olds were categorized as 30-month-olds). For the NSCH, we used the precalculated BMI percentiles that were also estimated using this method (22). That is, BMI percentile was determined using the same methodology for the NHIS and NSCH samples.

The childhood overweight prevalence estimates and mean height and weight values from NHIS and NSCH parent-reported data were compared to estimates from NHANES measured data. Differences were analyzed by age group, gender, and race/ethnicity. Two-year age groups produced sufficient sample size for stable estimates. Race/ethnicity groups included were non-Hispanic white, non-Hispanic black, and Mexican American from NHANES and NHIS. Estimates for the overall Hispanic population are not available from NHANES. The NSCH, however, does not collect data on Hispanic subgroups. Therefore, race/ethnicity estimates from NSCH are for the entire Hispanic population. Estimates for the total sample from each survey also include race/ethnic groups not shown separately.

National estimates were calculated using sample weights. Data were analyzed using PC SAS version 9.1. Because of the complex sampling design of the surveys, we used SUDAAN 9.0 to calculate standard errors. Differences in means were tested at the 0.05 level using t-tests.

Results

Using the gold standard of measured height and weight (NHANES 1999–2004), 15.6% (95% confidence interval 14.4, 16.8) of children aged 2–17 years were estimated to be overweight. Parent-reported data yielded much higher estimates: 25.3% (95% confidence interval 24.8, 25.8) using NHIS data (1999–2004) and 24.5% (95% confidence interval 24.0, 25.1) using NSCH data (2003–2004) (Figure 1). Keep in mind that these results compare different samples of children and not parent-reported and measured data for individual children. However, all three samples are nationally representative. The NHANES 1999–2004, the only national source of measured data, does not contain parent-reported data for children. Overestimation of childhood overweight prevalence using parent-reported data was observed for both genders and across race/ethnic groups. When finer age groups were examined, it became clear that the results for younger age groups drove this pattern. Among 2–3-year-olds, NHANES measured data yielded a prevalence estimate of 9% overweight vs. 42% using NHIS parent-reported data and 44% using NSCH parent-reported data. For 12–13-year-olds, in contrast, measured and parent-reported data yielded similar estimates (Figure 2). For 16–17-year-olds, the comparison of measured and parent-reported data reflected the pattern seen among adults: underestimation of overweight prevalence using reported data compared to measured data, although the confidence intervals from all three surveys overlap (6). A sensitivity analysis of NHIS data included records with implausible BMI values as defined by the World Health Organization exclusion criteria. The pattern was similar, but with even more exaggerated overestimation of overweight prevalence among very young children (53 vs. 42% with implausible records removed) with the difference narrowing as age increased (data not shown).

Figure 1.

Percent overweight (BMI ≥ 95 percentile for age and gender): measured vs. proxy-reported data for children 2–17 years, United States. aMexican American estimates are not available from NSCH—the NSCH category for this estimate is for all Hispanics. NSCH, National Survey of Children's Health.

Figure 2.

Measured vs. proxy-reported data: percent overweight (BMI ≥ 95 percentile for age and gender), children 2–17 years by age group, United States.

Discrepancies between measured and parent-reported data were also examined by gender and race/ethnic groups (Table 1). Although there was variation in the specific patterns observed by gender and race/ethnicity, the same general pattern described earlier was observed: a gradual transition from overestimation of overweight prevalence using parent-reported data to underestimation over the 2–17-year-age span. Overestimation of overweight prevalence using parent-reported data was greater for boys and persisted until older ages compared with girls. Conversely, compared with boys, parent-reported data underestimated overweight prevalence beginning at younger ages among girls.

Table 1.  Percent overweighta (s.e.): measured vs. parent-reported data for children 2–17 years, by age group, race/ethnicity, and gender, United States, 1999–2004
inline image

Next, differences between measured and parent-reported weight and height were assessed. This analysis revealed that the means of parent-reported height among young children were lower than measured means, and this difference drove the observed discrepancies in overweight prevalence estimates. For brevity, only the differences between NHANES and NHIS are shown in Table 2 because the survey years match and the overweight estimates for both parent-reported surveys (NHIS and NSCH) were similar. For both boys and girls between ages 2 and 11 years, the mean NHIS reported height was 3–6 cm less than mean NHANES measured height (P < 0.05). Among children 16–17 years of age, there was little difference between measured and parent-reported mean height. The relative difference between NHANES and NHIS was 3–5% for younger children and near 0% for older children. The absolute differences in the mean values for weight between NHANES measured data and NHIS parent-reported data were smaller than for height (Table 3), and diverged between boys and girls. For older boys, there was no difference between the measured and the parent-reported mean weight values. Among 8–17-year-old girls, mean weight was significantly lower in the sample with parental report compared to that with measured data (P < 0.05). This was also true for boys 8–9 years of age. However, as for boys, mean parent-reported weight was significantly higher than mean measured weight for 2–3-year-old girls (P < 0.05).

Table 2.  Measured and parent-reported mean height in centimeters (s.e.) and differences, children 2–17 years, by age group and gender, United States
inline image
Table 3.  Measured and parent-reported mean weight in kilograms (s.e.) and differences, children 2–17 years, by age group and gender, United States
inline image

Discussion

The rise in childhood overweight prevalence (2), the association between high BMI percentile in childhood and obesity in adulthood (27), and the health consequences throughout the lifespan of being overweight in childhood (28,29) make monitoring childhood overweight prevalence a public health priority. For groups pressed to produce local and/or real-time data, low-cost options for surveillance are attractive, and reliance on reported height and weight data is common. Furthermore, the justification for using reported data is bolstered by studies among adolescents and adults demonstrating a high correlation between self-reported and measured height and weight (8,9,10,11,12,13,14,30,31). However, some investigators warn that BMI calculated from self-reported data may result in misclassified weight status with up to 50% of overweight adolescents being missed (8,9,10,11,12,13,14).

What are the implications of using parent-reported data for surveillance and research of overweight children?

Use of parent-reported data overestimated childhood overweight prevalence overall, specifically among the youngest children when compared to measured data. The results are similar when measured data were compared to each of two national interview surveys with nationally representative samples. Although we compared sample populations and did not directly compare measured and reported data for individual children, this analysis supports previous studies demonstrating the potential for misclassification of weight status when using reported data (8,9,10,11,12,13,14,16,18). We recommend that parent-reported data for preschool- and elementary school–aged children not be used to estimate overweight prevalence.

The overestimation of childhood overweight prevalence using parent-reported data is driven primarily by the lower mean height for the youngest children. Although the discrepancy in mean height between reported and measured data is not very large, the calculation of BMI compounds any difference. Neither the NHIS nor the NSCH ask parents to measure children or to consult records before reporting height and weight. The consistently lower reported height we observed could arise from difficulty in estimating the height of much shorter children, or from parents not tracking height as closely as weight. It is likely that weight is measured more frequently using household scales.

Our results differ from studies analyzing discrepancies between measured and reported data for individual preschool children. Two studies, in which parents were not aware that reports would subsequently be validated with measurements, found little difference between means of reported and measured height (16,17). However, both groups of investigators concluded that studies of childhood overweight should rely on measured data because of significant misclassification bias.

Are reported data more valid for older children?

Determining overweight prevalence among adolescents using parent-reported weight and height begins to reflect the pattern seen for adult self-reported data—lower mean reported weight with larger differences between reported vs. measured mean weights for females compared to males (6). This pattern has been well documented for self- and parent-reported data for adolescents (5,8,9,11,12,13,14,30,31), although not in all studies (10). In general, studies found high correlation between measured and reported weight and lower correlation between measured and reported height and BMI (8,9,10,12,13,30). Some studies also demonstrated that reported data have high specificity (92–99.6%) but low sensitivity (48–76%) in identifying overweight status (8,9,10,12,13). We also observed that differences between measured and reported data may be accentuated among subgroups. For example, among non-Hispanic white females aged 14–15 years, overweight prevalence was 14.7% using measured data vs. 8.2% using reported NHIS data. Our results support the recommendations of other investigators to avoid use of parent-reported data for overweight surveillance in adolescents if possible (9,10,11,13).

Can a correction factor be used for reported data for adolescents?

Analyses of correction factors have focused on self-reported rather than parent-reported data, but the lessons are applicable to our findings among adolescents given the similarity between parent- and self-reporting patterns for older adolescents. Some researchers have argued that high correlation between reported and measured data justifies use of a correction factor for self-reported data for adolescents to assess overweight prevalence (8,14,32). But there are reasons to avoid correction factors. First, errors may not be homogenous. That is, discrepancies between reported and measured values vary by other factors in addition to age, gender, and race/ethnicity (5), including pubertal stage, time since menarche, exercise levels (6,14), and socioeconomic status (15). Although correction factors based solely on reported height and weight among adults have been shown as accurate as more complicated methods (33), some investigators concluded that the pattern of error in self-reported data among adults is difficult or impossible to correct using linear regression (15), and similar results have not yet been demonstrated for adolescents for whom pubertal stage may be an important modifier. Others advise that studies using reported data should collect measured data from a subset of the sample to calculate a study-specific measure of bias (8). Second, correlation does account for the magnitude or direction of bias. If reported height and weight are biased in opposite directions, the bias for calculated BMI can become very high (17). Where only reported data are available, such as state-level surveillance from NSCH, estimates of overweight among older children (starting as early as age 10 years among girls) must be interpreted with the knowledge that the overweight prevalence is likely to be underestimated, and that no valid correction factor to adjust for misclassification has been evaluated for children. Parent-reported data for younger children are so poor that no correction is possible.

Should reported data for weight and height (and the derived BMI) be used as covariates in analyses?

Some researchers argue that the high correlation between reported and measured data may justify the use of continuous reported data for height, weight and BMI as covariates (5,6,13,18,31). However, because reporting among persons with high measured values of weight tend to be least accurate (14,15,30), high BMI categories calculated from reported values have low sensitivity. Additional issues suggest caution when interpreting analyses using continuous values for reported weight, height, and BMI: (i) many studies showing high correlation have been based on surveys where respondents report height and weight with the knowledge of subsequent validation (e.g., persons ≥16 years in NHANES)—correlation between reported and measured values may be lower for surveys in which reported data are not validated (e.g., NHIS); (ii) because patterns of misclassification may vary by age (8,9,11,12,13,30), gender (8,14,32), race (13,30,31), and overweight status (34), using reported data increases the likelihood of bias. For example, misclassification has been observed to attenuate associations between weight status and outcomes in a study of asthma among adults (34); and (iii) the group with the least accurate reporting of weight—persons with high body weight—has increased in prevalence. Again, parent-reported data for small children yield unreliable values for BMI and should not be used, even as continuous data as covariates.

The main limitation of this study is the inability to compare parent-reported and measured data for individual children. By directly comparing the NHIS sample to the NHANES sample, we assume that these nationally representative surveys contain comparable samples of children in terms of weight status. Finally, our results are based on samples of children from the United States. To the extent that reporting of weight and height may differ by race/ethnicity, cultural norms, and other factors, results may not be generalizable to populations of other countries.

Use of parent-reported data for children from each of two national health surveys overestimates childhood overweight prevalence compared to measured data from a national examination survey. Weight status estimated using measured data remains the gold standard for childhood overweight surveillance. Although this analysis could not compare parent-related and measured data for individuals, the results reinforce previous recommendations that parent-reported data not be used to estimate prevalence of overweight among preschool- and elementary school–aged children. Given that there is a need for simple, cost-effective methods to obtain reported height and weight in young children, more research is needed to find either appropriate correction factors or techniques that generate more accurate reports.

Acknowledgments

We thank Sylvia Tan and Margaret Carroll for their programming expertise with NHANES. We also thank Randy Curtin and Ken Schoendorf for their careful review and comments.

Disclosure

The authors declared no conflict of interest.

Ancillary