BMI-for-Age and Weight-for-Length in Children 0 to 2 Years
OBJECTIVES: To determine the agreement between weight-for-length and BMI-for-age in children 0 to <2 years by using research-collected data, examine factors that may affect agreement, and determine if agreement differs between research- and routinely collected data.
METHODS: Cross-sectional data on healthy, term-born children (n = 1632) aged 0 to <2 years attending the TARGet Kids! practice-based research network in Toronto, Canada (December 2008–October 2014) were collected. Multiple visits for each child were included. Length (cm) and weight (kg) measurements were obtained by trained research assistants during research visits, and by nonresearch staff during all other visits. BMI-for-age z-scores were compared with weight-for-length z-scores (the criterion measure).
RESULTS: The correlation between weight-for-length and BMI-for-age was strong (r = 0.986, P < .0001) and Bland-Altman plots revealed good agreement (difference = −0.08, SD = 0.20, P = .91). A small proportion (6.3%) of observations were misclassified and most misclassifications occurred near the percentile cutoffs. There were no differences by age and sex. Agreement was similar between research- and routinely collected data (r = 0.99, P < .001; mean difference −0.84, SD = 0.20, P = .67).
CONCLUSIONS: Weight-for-length and BMI-for-age demonstrated high agreement with low misclassification. BMI-for-age may be an appropriate indicator of growth in the first 2 years of life and has the potential to be used from birth to adulthood. Additional investigation is needed to determine if BMI-for-age in children <2 years is associated with future health outcomes.
- NPV —
- negative predictive value
- PPV —
- positive predictive value
- TARGet Kids! —
- The Applied Research Group for Kids!
- WHO —
- World Health Organization
What’s Known on This Subject:
BMI-for-age growth charts are now available for growth monitoring in children younger than 2 years, although weight-for-length remains the recommended approach. If BMI-for-age performs similarly to weight-for-length, practitioners could use the same metric from birth to adulthood.
What This Study Adds:
Agreement between weight-for-length and BMI-for-age is very high, with most misclassifications close to the percentile cutoffs. BMI-for-age appears to be an appropriate anthropometric alternative in children <2 years.
Growth monitoring continues to be the most valuable clinical and public health tool to monitor growth and assess the health and nutritional status of children.1,2 Growth monitoring of children 0 to 18 years old in primary care is recommended by numerous expert bodies worldwide.3–6 In 2006, the World Health Organization (WHO) endorsed new growth reference charts that were constructed from the monitoring of growth, in a longitudinal manner, of healthy, singleton, term-born children in 6 ethnically diverse countries.7,8 These charts represent ideal growth in children under optimal environmental conditions for growth and have percentile cutoffs that can be used to classify weight status7,9 (eg, wasting, overweight, obesity) that may be practical for growth monitoring and screening.10–12
Currently, it is recommended that clinicians assess weight status by calculating and plotting weight-for-length in children 0 to <2 years of age, and then transition to BMI-for-age in children 2 years of age and older.3,13 However, the WHO Child Growth Standards (2006) also includes BMI-for-age growth reference charts for children <2 years that were not previously available.14 Using 1 tool, such as BMI-for-age, would give clinicians the ability to use BMI from birth to adulthood, track growth trajectories using 1 metric, and avoid the transition between differing measures after 2 years of age.
The association between weight-for-length and BMI-for-age in children <2 years has been explored by others, but with some limitations.15–17 Similarities in prevalence have been reported for some (eg, underweight, overweight, obesity), but not all, weight status categories when comparing weight-for-length and BMI-for-age.16,17 One study also reported a good correlation (r = 0.83, P < .0001) between the 2 measures, but included children with chronic diseases and preterm infants in their comparison.15
Recent studies also suggest the importance of comparing agreement between research- and routinely collected anthropometric measurements.18,19 Several pediatric studies report small differences for height (∼0.3 – 0.9 cm) and weight (∼0.01 – 0.04 kg) alone,20,21 exemplifying the ability to use primary care data for population growth monitoring. It is unknown whether the agreement between weight-for-length and BMI-for-age differs among these data sources.
The primary objective of this study was to determine the agreement between weight-for-length and BMI-for-age in healthy, term-born children aged 0 to <2 years using research-collected data. The secondary objectives were to examine if age, sex, and weight status category affect agreement, and if agreement differs between research- and routinely collected data.
Cross-sectional data were collected through the TARGet Kids! (The Applied Research Group for Kids) primary care practice-based research network in Toronto, Ontario, Canada. TARGet Kids! is a collaboration between child health researchers in the Faculty of Medicine at the University of Toronto and primary care physicians in the university’s Department of Pediatrics and Department of Family and Community Medicine. Details of study recruitment, including study protocol, have been described previously.22 The study was approved by the Hospital for Sick Children and St Michael’s Hospital Research Ethics Boards. All parents of participating children provided written, informed consent to participate in the study.
Children 0 to <2 years were recruited from 9 pediatric or family medicine primary care practices during scheduled well-child visits between December 2008 and October 2014 (n = 1632). In this study, children were included if a weight and length measurement had been obtained by a trained research assistant on the same day at any well-child visit between 0 and 2 years of age. We excluded children with gestational age <37 weeks, birth weight <1500 g, a health condition affecting growth (eg, failure to thrive, cystic fibrosis), a chronic illness (except asthma), severe developmental delay, or the absence of a parent and/or guardian fluent in English.
Trained research assistants at each primary care practice collected demographic information from parents by using a standardized data collection form adapted from the Canadian Community Health Survey.23 Demographic information included age, sex, and maternal ethnicity. Ethnicity was classified using a close-ended maternal ethnicity question designed and validated by the TARGet Kids! Collaboration that states: “What were the ethnic or cultural origins of your child’s ancestors? (An ancestor is usually more distinct than a grandparent.)”24 Response categories (described elsewhere)22,24 were then collapsed into the following subcategories: European; East, South, and Southeast Asian; African and Caribbean; Latin American; West Asian, Arab, and North African; and Mixed.
Research-collected data included weight and length measurements obtained by trained research assistants during a scheduled well-child visit (ie, a research visit). Standardized measurement techniques were used for all research-collected data; weight (kg) was measured using a precision digital scale (± 0.025%; Seca, Hamburg, Germany), and length (cm) was measured to the nearest 0.1 cm with a calibrated length board. Routinely collected data included weight and length measurements performed without the presence of a trained research assistant, and abstracted from the primary care health records of these enrolled children from any other health care visit. The method of weight and length ascertainment, including adherence to recommended protocols, including the use of standardized equipment, calibrated length boards, measurement by various team members, such as clinic nurse, physician, or other health care professional during these visits was unknown. Research and routine visits did not occur on the same date. For both research and routine visits, data from multiple visits were available for each child and were included (ie, repeated measures).
BMI (kg/m2) was calculated as weight divided by the square of length.25,26 Age- and sex-specific percentiles and the corresponding z-scores were determined by using the WHO Child Growth Standards (2006) for both weight-for-length and BMI-for-age.3,27 Percentiles and z-scores were electronically computed by using WHO Anthro software (www.who.int/childgrowth/software/en/). Z-scores were classified into weight status categories by using the following percentile cutoffs: severely underweight (z < 0.1st), underweight (z < 3rd), normal (3rd ≤ z ≤85th percentile), at-risk overweight (z >85th), overweight (z >97th), and obese (z >99.9th). These cutoffs are used to describe growth in children 0 to <2 years of age by using the WHO Child Growth Standards.3
Data Cleaning and Identifying Outliers
Data were assessed for quality by first identifying weight-for-length and BMI-for-age z scores < −4.0 and > −5.0.1 For observations with these values, we reviewed health records to compare with available data on previous or subsequent well-child visits within 2 years. Data points were removed if there was a ±1 SD difference from a previous or subsequent visit. If there was no previous or subsequent visit, the data point was also removed.
Descriptive statistics, including frequency distributions for categorical variables (eg, age, sex, ethnicity) and mean (±SD) and median (interquartile range) for continuous variables (eg, age, weight-for-length, and BMI-for-age) are presented. A Pearson χ2 test was used to detect the difference in proportions between weight status categories.
For our primary analysis, the overall degree of agreement between weight-for-length and BMI-for-age (as continuous variables) for research-collected data were evaluated by using a Pearson correlation coefficient and visually examined agreement by using a Bland-Altman plot (with 95% limits of agreement).
For our secondary analysis, we evaluated the Pearson correlation coefficients stratified by age, sex, and weight status categories. The McNemar χ2 test was used to test the difference in the proportion of observations classified into each weight status category by using weight-for-length or BMI-for-age (eg, an observation classified as normal weight status using weight-for-length, but overweight using BMI-for-age). A scatterplot of weight-for-length and BMI-for-age was graphed to visually inspect misclassification between growth status categories. Sensitivity, specificity, positive predictive values (PPV), and negative predictive values (NPV) were also calculated to determine the influence of weight status category by using weight-for-length as the criterion measure. Weight-for-length was selected as the criterion measures because it is the currently recommended method of growth monitoring in this age group.3,13
All descriptive statistics and analyses described previously were then conducted for routinely collected data to determine the difference in agreement between research- and routinely collected data. All data were analyzed by using R v.3.0.3 (Murray Hill, NJ). All hypotheses were 2-sided and P < .05 was considered statistically significant.
A total of 1632 children aged 0 to <2 years were enrolled in TARGet Kids! between October 2008 and October 2014. Measurements from 3517 research well-child visits from these children were available for analysis. After the exclusion of missing data points (height, weight, gestational age, and birth weight) and nonvalid z scores, 2190 observations remained for inclusion in the analysis (see Fig 1). Children were of mainly European descent (62.8%), although the population was ethnically diverse and boys comprised 53.6% of the children (Table 1). Of the observations included, the median age was 13.6 months (SD = 5.06). The mean weight-for-length z score was −0.06 (SD = 1.09); the mean BMI-for-age z score was −0.14 (SD = 1.12, Table 1). The proportion of observations classified with a normal weight status by using weight-for-length was 82.9% (n = 1815). For BMI-for-age, 80.5% (n = 1764) of observations were classified with a normal weight status (Fig 2).
The Pearson correlation between weight-for-length and BMI-for-age was strong, positive, and statistically significant (r = 0.985, P < .001). The Bland-Altman plot revealed that the mean of the differences between weight-for-length and BMI-for-age was near 0 (difference = −0.079, SD = 0.19) and the difference was not statistically significant (P = .68). The magnitude of the limits of agreement was < |0.5| (−0.46 to 0.31) and most observations were within the 95% confidence limits (Fig 3).
In the secondary analysis stratified by age and sex, weight-for-length and BMI-for-age were strongly and positively correlated in each of the age and sex categories (r ≥ 0.979 for all coefficients, see Table 2). When stratified by weight status categories, the strongest correlation was observed in the normal weight category (r = 0.97), but all correlations were strong and positive (underweight [r = 0.89], at-risk overweight [r = 0.85], and overweight [r = 0.87]). Figure 4 illustrates the classification of observations into weight status categories. The overall rate of misclassification was 6.3% (n = 138/2190). Misclassifications occurred near the cutoffs. The McNemar χ2 test revealed that the proportion of observations misclassified was statistically significant for wasting (P < .001), but not for other weight status categories (P > .05 for all other categories).
Among those identified by using weight-for-length, most observations (sensitivity ≥0.77 for all weight status categories) were correctly classified in the same weight category by using BMI-for-age. Sensitivity was the highest for wasting (0.92) and lowest for at-risk overweight (0.77, Table 3), whereas specificity was high in all categories (≥ 0.97). Within each weight status category, at least 75% were correctly identified in the same category; the lowest rates were observed for wasting (0.75) and at-risk overweight (0.83, Table 3). The category with the highest PPV was overweight (0.89). NPVs were high and similar for wasting, at-risk overweight, and overweight (≥ 0.97, See Table 3).
We performed all described analyses by using only routinely collected data and found similar results when compared with research-collected data (see Supplemental Tables 4, 5, and 6). The Pearson correlation coefficient between weight-for-length and BMI-for-age was 0.99 (P < .001). The mean difference between weight-for-length and BMI-for-age was −0.084 (SD = 0.20) and was not statistically significant as determined by a Bland-Altman plot (P = .67, data not shown). Pearson correlation coefficients were similar among age and sex categories (see Supplemental Table 5). Sensitivity and PPVs were highest among wasting (0.92) and overweight (0.92) observations, respectively, whereas specificity was similar and high in all categories (see Supplemental Table 6). Analyses were conducted on 1 randomly selected observation per child and the results were similar (data available on request).
Our results indicate high agreement between weight-for-length and BMI-for-age with low misclassification overall. The Bland-Altman plots were symmetrical on visual inspection and no systematic bias was identified. Our results demonstrated high specificity (≥97%) and most of those identified in any weight status category were correctly classified (≥75%). Most misclassifications occurred near the cutoffs and misclassification was not statistically significant for any weight status category, except for wasting. The agreement between weight-for-length and BMI-for-age was similarly high in routinely collected data, indicating the potential for routinely collected data to be used for growth monitoring and for research and public health purposes.
Previous research has supported the use of BMI-for-age for growth monitoring in the first 2 years of life. Nash et al15 reported a Pearson correlation coefficient of 0.83 (P < .0001) between weight-for-length and BMI-for-age in a small (n = 547) population recruited from a pediatric tertiary care setting. Nash et al15 included children with chronic conditions affecting growth (eg, cystic fibrosis, failure to thrive, congenital defects) and 18% of the sample constituted preterm infants. Our exclusion of children with chronic disease may have resulted in a higher correlation coefficient (0.99, P < .001).
Nash et al15 also reported fewer children identified as at-risk overweight (ie, ≥85th percentile) by using BMI-for-age (12.5%) compared with weight-for-length (18.2%).15 We did not identify any differences in prevalence at this cutoff or at other cutoffs. Our sample size was larger and included only healthy children, which may account for this difference. De Onis et al17 reported a similar prevalence for weight-for-length and BMI-for-age in overweight in children <5 years, whereas Mei et al16 reported no difference for wasting and overweight children in the same age category.
The potential for routinely collected data to be used for research and public health surveillance has been recently demonstrated, particularly for weight and length measurements in children.20,21 Both studies report high agreement for weight and length between research- and routinely collected data.20,21 Our results align with these findings: that routinely collected data appears to be an accurate source of information.
Our overall rate of misclassification was low (∼6%), but this rate was different between weight status categories. PPVs were lowest for wasting, at-risk overweight, normal, and overweight (see Table 3). For wasting, as high as 25% of those identified as wasted by using BMI-for-age were not identified as wasted by using weight-for-length. For normal, at-risk overweight, and overweight, the proportion of misclassified children was 17%, 13%, and 11%, respectively. These percentages indicate that the 2 measures may not be entirely interchangeable, although it appears that misclassification is occurring near the percentile cutoffs (see Fig 3). In the future, it will be necessary to determine how these differences in classification affect longitudinal child health outcomes.
Our study has several strengths. Children in this study were recruited as part of TARGet Kids!, a primary care practice-based research-based research network, which is an ethnically diverse cohort of children with research-collected data from numerous well-child visits in the first 2 years of life.22 Second, children with conditions affecting growth, and those born preterm and very low birth weight were excluded from our analysis. Furthermore, we verified the data quality of both research- and routinely collected data by identifying outlier z scores and then determining both their biological plausibility and consistency with measurements from other well-child visits. In addition, we have used the WHO recommended guidelines for classifying weight status, which is applicable to all children 0 to <2 years worldwide, regardless of socioeconomic status, ethnicity, or feeding patterns.7
One potential limitation of our study is that data were combined for children <3rd percentile (ie, wasting [<0.1st] + severe wasting [>3rd]), as well as for those >97th percentile (ie, overweight [>97th] + obesity [>99.9th]). A small sample size at these extreme values impeded our ability to examine the agreement in these weight status categories separately. Additionally, we excluded those children for whom birth weight and gestational age were not known (n = 954); however, their inclusion may have provided additional insight. Second, although we have used the recommended percentile cutoffs by the WHO to define weight status categories, the validity of these categories in younger children remains poorly understood. Many of these cutoffs were validated for older children only or chosen as statistical, rather than clinical, criteria.25,28–30
Although weight-for-length was used as the criterion measure, it may be argued that a more accurate and proximate measure of body fat (eg, skinfold test, dual energy x-ray absorptiometry) be considered a “gold standard” to assess weight status.31–33 We did not collect data on these measures in our study. Importantly, though, weight-for-length is the currently recommended measure to use for growth monitoring by the American Academy of Pediatrics4 and the Canadian Pediatric Society3 in children <2 years. Indeed, weight-for-length in children <2 years as a marker of weight status has been associated with obesity34,35 and cardiometabolic outcomes36 at other points throughout childhood. There is also evidence to support BMI-for-age as a marker of weight status,37–39 and it performs similarly to dual energy x-ray absorptiometry to predict cardiovascular outcomes throughout childhood.38,40 Last, the generalizability of our results is not known. Although we excluded preterm and very low birth weight infants, we did not have enough data to compare certain demographic characteristics of our population with those of the WHO Multicentre Growth Reference Study that represents ideally growing children (eg, singleton birth rate, smoking mothers, and breastfeeding practices).
We have demonstrated high agreement with limited misclassification between weight-for-length and BMI-for-age in healthy children 0 to <2 years and found that agreement is similar between research- and routinely collected data. If BMI-for-age were to replace weight-for-length as the weight status standard for children 0 to <2 years, this may enable improved monitoring of longitudinal growth patterns in young children. Future studies are required to examine unmeasured weight status categories, including severe wasting and obesity, where classification and agreement may be lower, and to determine the predictive ability of BMI-for-age for long-term health outcomes as compared with weight-for-length.
*TARGet Kids! Collaboration. Co-Leads: Catherine S. Birken, Jonathon L. Maguire; Advisory Committee: Eddy Lau, Andreas Laupacis, Patricia C. Parkin, Michael Salter, Peter Szatmari, Shannon Weir; Scientific Committee: Kawsari Abdullah, Yamna Ali, Laura N. Anderson, Imaan Bayoumi, Catherine S. Birken, Cornelia M. Borkhoff, Sarah Carsley, Shiyi Chen, Yang Chen, Denise Darmawikarta, Cindy-Lee Dennis, Karen Eny, Stephanie Erdle, Kayla Furlong, Kanthi Kavikondala, Christine Koroshegyi, Christine Kowal, Grace Jieun Lee, Jonathon L. Maguire, Dalah Mason, Jessica Omand, Patricia C. Parkin, Navindra Persaud, Lesley Plumptre, Meta van den Heuvel, Shelley Vanderhout, Peter Wong, Weeda Zabih; Site Investigators: Murtala Abdurrahman, Barbara Anderson, Kelly Anderson, Gordon Arbess, Jillian Baker, Tony Barozzino, Sylvie Bergeron, Dimple Bhagat, Nicholas Blanchette, Gary Bloch, Joey Bonifacio, Ashna Bowry, Anne Brown, Jennifer Bugera, Douglas Campbell, Sohail Cheema, Elaine Cheng, Brian Chisamore, Ellen Culbert, Karoon Danayan, Paul Das, Mary Beth Derocher, Anh Do, Michael Dorey, Kathleen Doukas, Anne Egger, Allison Farber, Amy Freedman, Sloane Freeman, Keewai Fung, Sharon Gazeley, Donna Goldenberg, Charlie Guiang, Dan Ha, Curtis Handford, Laura Hanson, Hailey Hatch, Teresa Hughes, Sheila Jacobson, Lukasz Jagiello, Gwen Jansz, Paul Kadar, Tara Kiran, Lauren Kitney, Holly Knowles, Bruce Kwok, Sheila Lakhoo, Margarita Lam-Antoniades, Eddy Lau, Fok-Han Leung, Alan Li, Jennifer Loo, Joanne Louis, Sarah Mahmoud, Roy Male, Vashti Mascoll, Rosemary Moodie, Julia Morinis, Maya Nader, Sharon Naymark, Patricia Neelands, James Owen, Jane Parry, Michael Peer, Kifi Pena, Marty Perlmutar, Navindra Persaud, Andrew Pinto, Tracy Pitt, Michelle Porepa, Vikky Qi, Nasreen Ramji, Noor Ramji, Jesleen Rana, Alana Rosenthal, Katherine Rouleau, Janet Saunderson, Rahul Saxena, Vanna Schiralli, Michael Sgro, Susan Shepherd, Hafiz Shuja, Barbara Smiltnieks, Cinntha Srikanthan, Carolyn Taylor, Suzanne Turner, Fatima Uddin, Joanne Vaughan, Thea Weisdorf, Sheila Wijayasinghe, Peter Wong, Anne Wormsbecker, Ethel Ying, Elizabeth Young, Michael Zajdman, Ian Zenlea; Research Team: Charmaine Camacho, Arthana Chandraraj, Dharma Dalwadi, Ayesha Islam, Thivia Jegathesan, Tarandeep Malhi, Megan Smith, Laurie Thompson; Applied Health Research Center: Christopher Allen, Bryan Boodhoo, Judith Hall, Peter Juni, Gerald Lebovic, Karen Pope, Jodi Shim, Kevin Thorpe; Mount Sinai Services Laboratory: Azar Azad.
We thank all of the participating families for their time and involvement in TARGet Kids! and are grateful to all practitioners who are currently involved in the TARGet Kids! practice-based research network.
- Accepted April 22, 2016.
- Address correspondence to Catherine Birken, MD, Child Health and Evaluative Sciences, Research Institute, Division of Paediatric Medicine, Department of Paediatrics, The Hospital for Sick Children, Rm 109801, 10th Fl, Peter Gilgan Centre for Research and Learning, 686 Bay St, Toronto, ON M5H 0A4 Canada. E-mail:
FINANCIAL DISCLOSURE: The authors have indicated they have no financial relationships relevant to this article to disclose.
FUNDING: Funding of the TARGet Kids! research network was provided by the Canadian Institutes of Health Research Institute of Human Development, Child and Youth Health, the Canadian Institutes of Health Research Institute of Nutrition, Metabolism and Diabetes, the SickKids Foundation, and the St. Michael’s Hospital Foundation. The Paediatric Outcomes Research Team is supported by a grant from The Hospital for Sick Children Foundation. The funding agencies had no role in the design and conduct of the study, the collection, management, analysis and interpretation of the data, or the preparation, review, and approval of the manuscript.
POTENTIAL CONFLICT OF INTEREST: The authors have indicated they have no potential conflicts of interest to disclose.
- de Onis M,
- Habicht JP
- ↵Canadian Task Force on Preventive Health Care. Recommendations for growth monitoring, and prevention and management of overweight and obesity in children and youth in primary care. CMAJ. 2015;187(6):411–421
- de Onis M,
- Garza C,
- Victora CG,
- Onyango AW,
- Frongillo EA,
- Martines J
- de Onis M,
- Lobstein T
- Gibson R
- WHO Multicentre Growth Reference Study Group
- Nash A,
- Secker D,
- Corey M,
- Dunn M,
- O’Connor DL
- de Onis M,
- Blössner M,
- Borghi E
- Bryant M,
- Santorelli G,
- Fairley L, et al; Born in Bradford Childhood Obesity Scientific Group
- Howe LD,
- Tilling K,
- Lawlor DA
- Carsley S,
- Borkhoff CM,
- Maguire JL, et al; TARGet Kids! Collaboration
- Statistics Canada
- Mei Z,
- Grummer-Strawn LM,
- Pietrobelli A,
- Goulding A,
- Goran MI,
- Dietz WH
- Zhang Z,
- Lai HJ
- Onyango AW,
- de Onis M,
- Caroli M, et al
- Albanese CV,
- Diessel E,
- Genant HK
- Hussain Z,
- Jafar T,
- Zaman MU,
- Parveen R,
- Saeed F
- Fields DA,
- Goran MI
- Taveras EM,
- Rifas-Shiman SL,
- Belfort MB,
- Kleinman KP,
- Oken E,
- Gillman MW
- Steinberger J,
- Jacobs DR,
- Raatz S,
- Moran A,
- Hong CP,
- Sinaiko AR
- Lindsay RS,
- Hanson RL,
- Roumain J,
- Ravussin E,
- Knowler WC,
- Tataranni PA
- Copyright © 2016 by the American Academy of Pediatrics