BMI Reporting and Accuracy of Child’s Weight Perception

The Fit Study (fall 2014 to spring 2017) was a cluster randomized clinical trial among California elementary and middle schools that assessed the impact of school-based BMI screening and reporting on student weight status.^11,15  Study details, including findings for the primary outcome analyses and trial protocol, are published elsewhere.¹¹  The UC Berkeley Committee for the Protection of Human Subjects approved the Fit Study. We use existing data from the study that included children in 27 treatment schools that were randomly assigned to offer screening and reporting (treatment) and 27 schools randomized to offer screening only (control).

The CONSORT flow diagram for the original cluster randomized trial that we leverage data from for this study is reported elsewhere.¹¹ Figure 1 displays the flow of eligible children with data available for secondary analysis for this study. This eligible sample includes 12 430 children who first entered the study either in fall 2014 or fall 2015 (treatment group: N = 5892; control group: N = 6538). Of this eligible sample, 2765 children (treatment group: n = 1202; control group: n = 1563), approximately 22.2%, were excluded because of missing BMI or child weight perception data. Thus, the final sample used in this study included 9665 children. Children excluded from the analysis were more likely to be Black (12% of excluded students were Black versus 6% of children who were included; P < .001) or white (17% vs 16%; P = .04) and less likely to be Asian (9% vs 16%; P < .001).

Children in the final sample were exposed to at least 1 round of BMI screening and reporting (treatment; n = 4690) or screening only (control; n = 4975). Among the 9665 children, 2795 were exposed to 2 rounds of screening and reporting (treatment; n = 1434) or screening only (control; n = 1361). Figure 2 displays a timeline depicting the sample of children who were exposed to at least 1 round and 2 rounds.

Data on children’s BMI comes from screenings conducted by school staff (PE teachers for 64% of students, nurses [26%], classroom teachers [7%], and office staff or principals or other adults [3%]) in the spring using research grade equipment¹⁵ ; staff height and weight measurements were found to be equivalent to those of trained researchers.²⁰  Parents of students in the treatment group were sent a BMI report in the fall, approximately 6 months after BMI was measured in spring. Reports were delivered after 6 months because of the time required to receive, process, and report on the data. A sampleBMI report is included in supplemental materials accompanying previously published findings of the Fit Study.¹¹  Focus groups with diverse groups of California parents informed the reports, which classified children as underweight (BMI <5^th percentile); healthy weight (BMI ≥5^th percentile and <85^th percentile); at risk for overweight (BMI ≥85^th percentile and <95^th percentile), or overweight (BMI ≥95^th percentile for sex and age) per Centers for Disease Control and Prevention standard cut points,²¹  but adopted an earlier naming convention to avoid labeling children as potentially having “obesity” given parents’ objection to this language.²² 

Data about weight perceptions comes from surveys administered to children in fall, approximately 6 to 9 months after BMI was assessed and 1 to 2 months after parents were sent BMI reports (for treatment students only). Children were asked if they perceived themselves to be very underweight, somewhat underweight, about the right weight, somewhat overweight, or very overweight.²³  “Very underweight” and “somewhat underweight” were collapsed to underweight for this analysis.

Using children’s objectively measured BMI from the spring and their self-reported weight perception from the fall, we created a 3-category measure to represent the accuracy of their weight perceptions for each study year. Because the BMI and weight perception accuracy categories used slightly different categorical labels, we aligned the BMI measurement categories to the weight perception categories as follows: underweight(BMI) = very underweight or somewhat underweight (weight perception); healthy weight (BMI) = about the right weight (weight perception); at-risk for overweight (BMI) = somewhat overweight (weight perception); and overweight (BMI) = very overweight (weight perception). Children were classified as (A) “underestimated” if their fall weight perception (eg, underweight) was lower than their subsequent spring BMI category (eg, healthy weight); (B) “accurately estimated” if their fall and spring categories matched each other; or (C) “overestimated” if their fall weight perception (eg, very overweight) was higher than their spring BMI category (eg, at-risk for overweight).

Schools provided data on parent-reported sex (male or female), race and ethnicity, and grade. Schools collected race and ethnicity information directly from parents during school enrollment and provided it in categories, including Black, Asian, white, or Hispanic. Children whose parents designated another race or ethnicity, or multiple races, were included in an “all other races and ethnicities” category. We used a “declined to state” category if parents declined to provide race or ethnicity data. Finally, we use school-level data on free and reduced-price meal eligibility from the California Department of Education.²⁴ 

Where m indexes the number of accuracy categories minus 1 omitted reference category. We used underestimated as the reference category (= 0). The outcome, $ln(P(Accuracy Categoryis = m)P(Accuracy Categoryis = 0))$⁠, is the log odds of being in category m relative to the reference category. $γqm$ captures the effects of the baseline controls $xis$⁠, whereas $β1m$ is the effect of the treatment on the outcome. Based on results of this model, we calculated the average marginal effect of BMI reporting on the probability that children accurately estimated their weight category. The average marginal effect is estimated by predicting how each child’s probability of being accurate changed if they were in the treatment versus control group, holding all other variables at their observed values, and then averaging over the changes in predicted probabilities. These predicted probabilities are interpreted as the percentage point change in the probability that children accurately reported their weight category. Models included baseline controls for child sex, race and ethnicity, grade level, and school district, as well as school-level proportion of students eligible for free or reduced-price meals. We clustered standard errors by school.

We first fit our model to data on the overall sample. Then, in a series of exploratory analyses, we fit the model stratified by sex, race and ethnicity, grade, and baseline BMI category subgroups. Our analyses by race and ethnicity were based on samples with a reported race or ethnicity; thus, we excluded in those analyses children whose race or ethnicity was unavailable (n = 24). We also fit models separately based on whether the sample received 1 or 2 rounds of BMI reporting. For our overall analysis focusing on all sample children, we adopted a standard level of significance (α = .05) with which to test the null that there was no impact of BMI reporting on the accuracy of children’s weight perceptions. Given the exploratory nature of our subgroup analyses, which typically do not require adjustments for multiple hypothesis testing since the findings are preliminary and used to generate hypotheses rather than confirm them,²⁵  we also adopted a conventional level of significance (α = .05).

For the sample used to estimate the impact of 1 round of BMI reporting (Table 1), slightly more children at baseline were female (51%) and in middle school grades (6^th or 7^th) (52%). The majority of children were Hispanic (60%) and had a baseline BMI between the 5^th and 85^th percentile (56%). Similar demographic patterns existed in the analytic sample used to estimate the impact of 2 rounds of BMI reporting. Of note is that the sample used to estimate 2 rounds versus 1 of reporting had a higher proportion of Asian students (29% vs 16%; P < .001) and 6^th graders (49% vs 31%; P < .001). Finally, by accuracy category, in the sample of treatment and control schools used to estimate the impact of 1 round of reporting, 48% underestimated, 43% accurately estimated, and 9% overestimated their weight.

Table 2 displays the differences between children in the treatment and control schools for the samples used to estimate the impact of 1 and 2 rounds of BMI reporting. Among the sample used to estimate the impact of 1 round, children in treatment schools were more likely to be Asian or white, and less likely to be Black or Hispanic, compared with children in control schools (P < .001). Slightly more students in treatment schools were in fourth and slightly fewer were in the 6^th grade, compared with children in control schools (P < .001). Among the sample used to estimate the impact of 2 rounds, similar differences also existed, with a few notable exceptions. Children in treatment schools were more likely to be healthy weight or overweight (P = .04), although there was no difference in the proportion of white students and fourth graders between treatment and control groups.

We found no impact of BMI reporting on children’s self-reported accuracy in the overall sample after 1 round (average marginal effect [AME]: 0.2; 95% confidence interval [CI]: −1.3 to 1.7, P = .82) or 2 rounds (AME: 3.9; 95% CI: −0.7 to 8.5, P = .09) of BMI reporting. Through our exploratory subgroup analyses, we detected statistically significant associations within subgroups by race, grade levels, and BMI categories. As shown in Table 3, for Black children, exposure to 1 round of BMI reporting was related to a 10.0 percentage point (95% CI: 2.6 to 17.4, P = .008) increase in accuracy. However, there was no association after exposure to 2 consecutive rounds of BMI reporting, each a year apart (AME: 9.4; 95% CI: −9.1 to 27.9, P = .32). On the other hand, for Asian children, exposure to 2 consecutive rounds of BMI reporting was associated with a 6.6 percentage point (95% CI: 0.4 to 12.8, P = .04) increase in accuracy. Finally, 5^th graders had a higher probability of accurate weight estimation after 2 rounds of BMI reporting (AME: 11.1; 95% CI: 1.6 to 20.5, P = .02) as well as those with BMI <5^th percentile (AME: 17.1; 95% CI: 2.7 to 31.6, P = .02).

The accuracy of children’s weight perceptions is 1 factor among a set of complex and interdependent influences at the individual, family, community, and societal levels that has been shown to be linked to children’s weight control behaviors, particularly among children with overweight or obesity.^4,5,26  In this follow-up study to the Fit Study, a large-scale randomized controlled trial to evaluate BMI measurement and reporting, we found that BMI reporting did not improve the accuracy of children’s weight perceptions, including those with overweight or obesity. Our exploratory subgroup analyses show that several subgroups became more accurate in their weight self-perceptions: Black children (after exposure to 1 round of BMI reporting) as well as Asian children, fifth graders, and those with BMI <5^th percentile (after exposure to 2 rounds of BMI reporting). Given that these subgroup findings are exploratory, future prospective studies that are theoretically grounded can be designed to confirm these results and to also establish whether accuracy influences their weight and weight-related outcomes. Further, studies that focus on specific racial and ethnic groups can assess whether there is an interplay between BMI notification and socially constructed ideals and norms about weight experienced by certain racial and ethnic groups that helps drive specific groups to have greater perception accuracy. Finally, additional effects that could be further examined include whether the direction of the shift in weight perceptions of children in particular weight categories influence dietary practices. For example, analyses can investigate whether underweight children in fifth grade who believe they were normal weight and became more accurate in their weight perceptions adopt less healthy dietary practices to “bulk up.”

Although school-based BMI screening is an important tool for surveillance purposes,²⁷  BMI screening and reporting have been shown to have no impact on BMI^11,28  and the present results may help elucidate 1 mechanism for why no effects have been found. In theory, weight perception can serve as a potential antecedent to behavioral changes in diet and exercise; yet, without increasing perception accuracy, BMI reporting may not have changed children’s behaviors sufficiently enough to lead to lower BMI.

Despite evidence that BMI reporting has no effect on either BMI or weight self-perception accuracy, schools are likely to continue to measure children’s BMI and share such information with parents. Given this, the CDC promotes safeguards to ensure that notifications are nonstigmatizing, actionable, and accessible, both in terms of reading level and language.²⁹  Notifications should also be designed in the context of cultural norms and values.³⁰  Lastly, given concerns about widening existing health disparities over the course of the coronavirus disease 2019 pandemic, including increases to BMI percentile,^31–33  schools and districts that already conduct BMI measurement programs may find it valuable to continue collecting height and weight data to better understand how coronavirus disease 2019 disrupted childhood obesity prevalence.

These results are specific to California schools in the Fit Study and may not hold in other state contexts. Additionally, our subgroup findings are exploratory rather than confirmatory and because of multiple hypothesis testing, these findings are subject to an inflated Type I error rate. Also, students were surveyed 6 to 9 months after they participated in BMI assessment, and we assumed that their measured BMI category remained stable between the assessment and survey. Finally, those receiving 1 versus 2 rounds of BMI reporting represent different subsamples that are demographically distinct. Notably, a higher proportion of Asian children received 2 rounds (29%) versus 1 (16%), and a lower proportion of Black and Hispanic children received 2 rounds versus 1 (3% versus 6% and 54% versus 60%, respectively). These differences limit our ability to directly compare the efficacy of 1 versus 2 rounds of BMI reporting within the same group of children.

In this follow-up study to the Fit Study, we found that BMI reporting did not improve the accuracy of children’s perceived weight status, a finding that offers additional insight into why BMI reports sent home to parents, as a behavior change strategy, may have had no impact on children’s BMI. Despite the lack of impact on both BMI and the accuracy of weight self-perceptions, BMI information gathered and reported at the aggregate level can be used to understand patterns and trends in overweight and obesity and to also help target interventions to schools who can most benefit from evidence-based³⁴  multipronged approaches.

We thank reviewers at the Centers for Disease Control and Prevention for comments on previous versions of this manuscript.