Parent‐reported compared with researcher‐measured child height and weight: impact on body mass index classification in Australian pre‐school aged children

Abstract Issue Addressed Parent‐reported data may provide a practical and cheap way for estimating young children's weight status. This study aims to compare the validity and reliability of parent‐reported height and weight to researcher‐measured data for pre‐school aged children (aged 2‐6 years). Methods This was a nested study within a cluster randomised controlled trial (October 2016‐April 2017), conducted within 32 Early Childhood Education and Care (ECEC) services across New South Wales, Australia. Parents of children reported on demographics and child height and weight via a survey. For the same child, height and weight data were objectively collected by trained research staff at the service. We calculated mean differences, intra‐class correlations, Bland‐Altman plots, percentage agreement and Cohen's kappa coefficient (>0.8 = “excellent”; 0.61‐0.8 = “good”; 0.41‐0.60 = “moderate”; 0.21 and 0.4 = “fair [weak]”; <0.2 = “poor”). Results Overall, 89 children were included (mean age: 4.7 years; 59.5% female). The mean difference between parent‐reported and researcher‐measured data were small (BMI z‐score: mean difference −0.01 [95% CI: −0.45 to 0.44]). There was “fair/weak” agreement between parent‐categorised child BMI compared with researcher‐measured data (Cohen's Kappa 0.24 [95% CI: 0.06 to 0.42]). Agreement was poor (Cohen's kappa <0.2) for female children, when reported by fathers or by parents with a BMI > 25 kg/m2. Conclusion There was “fair/weak” agreement between parent‐reported and measured estimates of child weight status. So What? Parent's report of weight and height may be a weak indicator of adiposity at the level of individuals however it may be useful for aggregate estimates.

Childhood obesity is an excess in body fatness, frequently defined according to body mass index scores, adjusted for child sex and age.
As childhood obesity can track throughout the lifespan 1 and is associated with a higher risk of chronic diseases in childhood and later life 2 early childhood is considered a critical period for directing public health obesity prevention efforts. 3[6] To support monitoring and understanding of the impact of public health interventions and policies, pragmatic and valid measures of child height and weight are needed to provide population estimates.
While objectively measured height and weight, collected by trained personnel is the gold standard for determining BMI, this approach is often not feasible at a population level.As such, many studies utilise parent reported weight and height to assess child BMI, as it is a simple and cost-effective method for collecting this information. 7Previous research internationally with pre-school aged children found that carers could reasonably estimate children's weight and height however over 75% of obese children would be missed using parent selfreport. 8,9To our knowledge, there have been no studies that have assessed this specifically in Australian preschool-aged children (aged <6 years).
Therefore, the aim of this study was to, in children aged between 2 and 6 years old, (i) compare parent reported to researcher measured child height and weight; and (ii) explore validity of parent-reported child height and weight classifications by child sex, or parent characteristics.

| Data source
This is a nested study using baseline data (collected between October 2016 and April 2017) from a cluster randomised controlled trial that aimed to investigate the impact of a web-based menu-planning intervention in Early Childhood Education and Care (ECEC) services.1][12]

| Participants
1][12] A total of 54 services consented to larger trial and a nested sample of parents and children attending 35 ECEC services in New South Wales (NSW), Australia participated in this study (see Figure 1).All participating ECEC services distributed information and consent forms to parents of children in the room with the highest number of children aged 2-6 years at baseline as a range of other data was also collected on the day.Consistent with previous approaches employed by the research team, 13,14 research assistants also approached parents at drop-off to provide additional information about the study and enable return of consent forms on the day to increase participation.Eligible children were: (i) aged 2-6 years; (ii) present at the centre on days of data collection; (iii) had no dietary requirements preventing consumption of foods while in care; and (iv) had parental consent.

| Data collection
Consenting parents were invited to participate in an online or telephone survey.The survey collected: (i) parent demographic information including parent age, gender, height, weight, Aboriginal and Torres Strait Islander status, country of birth, language spoken at home, and highest education attainment and (ii) parent report of their child's height (without shoes, in cm/feet and inches) and weight (without clothes or shoes, in kg, pounds or ounces).
Objective measures of child height and weight were also collected from the same children on the day by trained research staff following a standardised protocol. 10Specifically, weight was measured using calibrated digital scales (NUWEIGH LOF842) and height was measured using a standing portable stadiometer (Charder HM 200P) on a hard, flat surface.Trained staff collected two measurements, from which the average of each measurement was used.The difference between measures was required to be 3 cm or less for height and 0.4 kg or less for weight, otherwise a third measure was taken.A third measurement for height was required in 13.5% of children, and none required a third measurement for weight.
For both parent-report and researcher-measured data, BMI was calculated as: weight (in kilograms) divided by height (in meters 2 ).BMI z-scores were calculated according to child age and sex and grouped according to cut-offs defined by the World Health Organization. 15

| Statistical analysis
All data analysis was conducted using SAS software, version 9.3 (SAS Institute).Descriptive statistics, including means, frequencies and proportions were used to describe the demographic data.Parent and child data were only included if the child had both researchermeasured and parent-reported height and weight data.This resulted in data from 32 of the 35 participating services being included in the analysis.
T A B L E 1 Differences between parent reported and researcher measured child height and weight and BMI.values from each method. 16,17 assess reliability related to parent-reported BMI categorisation (Underweight; Healthy; Overweight/Obese), we calculated percentage agreement and weighted Cohen kappa coefficient (ranging from 0 to 1), were considered "excellent" if >0.Overall, the mean difference between parent-reported and researcher measured height, weight, BMI (kg/m 2 ) and BMI z scores are small and not statistically significant (see Table 1).The lCC for was lowest for BMI (kg/m 2 ) and BMI z scores (ICC = .16,indicating poor reliability), and highest for weight (ICC = .72,indicating moderate reliability).There were statistically significant differences in mean differences between measures for male child height (À2.00 [95% CI: À3.98 Percentage agreement between parent-reported and research-measured data on BMI classifications was 58%.Subgroup analyses revealed that percentage agreement between the measures was better for male children (67%); when reported by mothers (64%); and parents with a healthy BMI (≤25 kg/m 2 ) (67%).Cohen's Kappa indicated fair agreement between the two measures overall (0.24 [95% CI: 0.06 to 0.42]), with "poor" agreement for female children (0.11 [95% CI: À0.12 to 0.35]), data reported by fathers (0.09 [95% CI: À0.24 to 0.41]), or by parents with a higher BMI (0.03 [95% CI: À0.25 to 0.31]) (see Table 2).

| DISCUSSION
Our study found that parents of Australian preschool-aged children report on average only slightly below researcher-measured height (mean difference: À0.27 cm) and weight (mean difference: À0.06 kg) data, which had minimal implications on calculated BMI z scores (mean difference: À0.01) and BMI (mean difference: 0.40).In terms of classification of weight status, parent-reported data provided a fair (weak) estimate of child weight status categorisation (Cohen's Kappa: 0.24).
T A B L E 2 Differences in body mass index classification using parent-reported and researcher-measured height and weight data.These results are consistent with international studies in preschool-aged children which found that parents underestimate child BMI by between 0.3 and 0.5 kg/m 2 . 8,9However, the agreement in BMI categorisation was weaker than a previous Australian study of children aged 4-11 years which found moderate agreement for BMI categorisation (Cohen's kappa of 0.59) despite a larger underreporting of weight (0.5 kg) and height (0.9 cm) on average. 20Visual inspections of the plot also identified groupings of data values.This could be partly due to differences in the way weight and height were recorded.This was up to two decimal points for measured and usually in integer values for parents.
Self-reported height and weight data represents a cost effective and pragmatic method for informing population health status at scale. 21It has been suggested that providing parents with explicit instructions when reporting anthropometric data could improve the accuracy of parent-reported data.A range of reporting biases may impact the validity of self-reported height and weight data; therefore, adjustment methods are commonly used in analysis models to improve self-reported estimates.This study provides indication of differences in reporting with better estimates for BMI categorisation when the child was male, where reported by mothers, and parents were ≤25 kg/m 2 .Such findings are consistent with previous studies internationally. 8A study with over 9000 German children found that girls were more likely to be misclassified then boys, and parent's perception of weight status (girls being perceived as "too fat" and boy perceived as "too thin") influenced this report. 22It is perhaps unsurprising that mothers report more accurately on their children's weight and height, given they commonly hold the role of primary carer.
The primary limitations of our current investigation are a relatively small sample of participants, attending ECEC services recruited to participate in a web-based menu-planning program, relatively low service response rate (47%) and cross-sectional study design.
The findings are however consistent with the only other Australian study with younger children. 20

| CONCLUSIONS
In a sample of Australian pre-school aged children, we found little difference in mean BMI z-scores between parent-reported and research measured data.Despite this, reliability of BMI categorisation was weak suggesting that weight and height data could provide reasonable estimates for population or large-scale studies, however this is unlikely to be an accurate indicator of adiposity at the individual level.
Further exploration of the factors that influence parent-report is needed in more representative samples of Australian pre-school aged children.

For
child height, weight, BMI (kg/m 2 ), and BMI-z scores differences between the two methods, we assessed mean estimates using linear mixed model regression controlling for clustering within ECC services.Subgroup analyses were conducted based on child sex (male vs female), and parent characteristics including parent sex (male vs female) and parent weight status (BMI ≤ 25 kg/m 2 vs BMI > 25 kg/ m 2 ).Intra-class correlations (ICC) were calculated and Bland-Altman plots used to explore the agreement between individual's absolute
Demographic characteristics of participating parents and their children (N = 89), as reported by parents.