Article Text


Anthropometric measures are simple and accurate paediatric weight-prediction proxies in resource-poor settings with a high HIV prevalence
  1. Kyly C Whitfield1,
  2. Roberta Wozniak1,
  3. Mia Pradinuk2,
  4. Crystal D Karakochuk1,
  5. Gabriel Anabwani3,
  6. Zachary Daly1,
  7. Stuart M MacLeod2,
  8. Charles P Larson2,
  9. Timothy J Green1,4
  1. 1Food, Nutrition and Health, University of British Columbia, Vancouver, British Columbia, Canada
  2. 2Department of Pediatrics, University of British Columbia, Vancouver, British Columbia, Canada
  3. 3Botswana-Baylor Children's Clinical Centre of Excellence, Gaborone, Botswana
  4. 4South Australian Health and Medical Research Institute, and the Women's and Children's Health Research Institute, Adelaide, South Australia, Australia
  1. Correspondence to Dr Tim J Green, South Australian Health and Medical Research Institute, and the Women's and Children's Health Research Institute, PO Box 11060, North Terrace, Adelaide, SA 5001, Australia;{at}


Rationale Accurate weight measurements are essential for both growth monitoring and drug dose calculations in children. Weight can be accurately measured using calibrated scales in resource-rich settings; however, reliable scales are often not available in resource-poor regions or emergency situations. Current age and/or length/height-based weight-prediction equations tend to overestimate weight because they were developed from Western children's measures.

Objective To determine the accuracy of several proxy measures for children's weight among a predominately HIV-positive group of children aged 18 months to 12 years in Botswana.

Design Weight, length/height, ulna and tibia lengths, mid-upper arm circumference (MUAC) and triceps skinfold were measured on 775 children recruited from Gaborone, Botswana, between 6 July and 24 August 2011.

Results Mean (95% CI) age and weight were 7.8 years (7.5 to 8.4) and 21.7 kg (21.2 to 22.2), respectively. The majority of children were HIV-positive (n=625, 81%) and on antiretroviral treatment (n=594, 95%). The sample was randomly divided; a general linear model was used to develop weight-prediction equations for one half of the sample (n=387), which were then used to predict the weight of the other half (n=388). MUAC and length/height, MUAC and tibia length and MUAC and ulna length most accurately predicted weight, with an adjusted R2 of 0.96, 0.95 and 0.93, respectively. Using MUAC and length/height, MUAC and tibia length and MUAC and ulna length equations, ≥92% of predicted weight fell within 15% of actual weight, compared with <55% using current equations.

Conclusion The development of nomograms using these equations is warranted to allow for rapid and accurate weight prediction from these simple anthropometric measures in HIV-endemic, resource-constrained settings.

  • weight prediction
  • anthropometry
  • paediatric
  • long bone length
  • mid-upper arm circumference

Statistics from

What is already known on this topic

  • Obtaining precise paediatric weight measurements is difficult in resource-poor settings where calibrated scales are not available, leading to inaccurate growth tracking and drug dose calculations.

  • Current age-based and height-based weight-prediction tools and equations are based on Western databases and therefore often overestimate weight in resource-poor settings.

What this study adds

  • Mid-upper arm circumference and length/height, tibia length or ulna length, all accurately predicted weight in a predominately HIV-positive group of Batswana children.

  • Simple, non-invasive anthropometric measurements have the potential to accurately predict weight; in this population, they were more accurate than available height-based and age-based equations/tools.

  • Nomograms are simple tools that maintain equation specificity and have the potential to quickly and accurately aid in weight prediction in resource-limited settings.


Although weight measurements in children are primarily used for monitoring growth and tracking nutritional status,1 they are also needed for calculating drug dosages.2–4 Drug metabolism is correlated with lean body mass, but the current standard for paediatric drug dose is calculation based on the child's weight.2 Using the child's weight, a milligram per kilogram (mg/kg) body weight dose is usually given in the form of a syrup or portion of an adult dose.5 It is estimated that up to 75% of antibiotic dosages in teaching hospitals globally are inappropriately prescribed.6 Drug dosing errors are the most common paediatric medical errors globally, with administration of doses up to 10 times greater than the correct amount.7 Researchers in Nigeria reported that over half of paediatric medications were being incorrectly dosed,8 which can lead to toxicity or ineffective treatment and/or development of drug resistance.9 In resource-rich, non-emergency settings, children are weighed on calibrated electronic weight scales10; however, these are not always available in resource-poor or emergency settings due to the cost of these scales, as well as continuing costs of maintenance4 (‘wear and tear’ from transport and heavy use11) and proper calibration, infrastructure requirements for proper use,4 lack of reliable power source11 and overheating when used outdoors in warm climates.12 It is therefore important to identify valid proxies of children's weight for use when calibrated weight scales are impractical or not available.

Proxy measures for children's weight such as age and height have been developed.13 A common proxy measure of paediatric weight is the Broselow tape, which uses length as a proxy for weight. The Broselow tape has been shown to be a more accurate estimate of children's weight than age-based formulas such as the Advanced Paediatric Life Support (APLS) and Luscombe and Owens formulas,14–16 but there is concern about its validity in areas of widespread stunting and/or malnutrition.16 Also, while the Broselow tape is user-friendly,13 it is expensive, potentially creating a barrier for use in resource-poor settings in non-emergency situations.13 ,16 ,17

Anthropometric measures may be ideal alternative proxies because they can be easily measured using simple, readily available tools. Although one proxy measure is ideal, it is more likely several anthropometric measures are required to accurately predict weight.14–16 Nomograms, simple charts depicting linear prediction equations with two measures,18 could be a better tool than traditional equations in resource-limited settings as a ruler or piece of string can be used to quickly and accurately predict weight without a calculator, time-consuming hand calculations, and with minimal training. The objective of this cross-sectional study was to determine the validity of various anthropometric measurements as potential proxies for paediatric weight in a resource-limited setting.



A convenience sample of 800 children aged 18 months to 12 years was recruited from the Botswana-Baylor Children's Clinical Centre of Excellence in Gaborone, Botswana, between 6 July and 24 August 2011. This health facility provides HIV testing to HIV-exposed children from 6 weeks of age and dispenses antiretroviral (ARV) therapy through regular 3 monthly check-up appointments. Caregivers provided informed consent and approval. Ethics approval was obtained from the University of British Columbia/Children's and Women's Health Centre of British Columbia Research Ethics Board and the Health Research Development Committee of the Botswana Ministry of Health.

Data collection

Demographic questions were obtained from the Demographic and Health Survey and included the child's birthdate, sex, HIV status and, if applicable, ARV use. Six anthropometric measurements were taken on each child: weight, length/height, ulna length, tibia length, mid-upper arm circumference (MUAC) and triceps skinfolds. All measurements were taken by two of the study investigators (RW, MP) according to the validated Food and Nutrition Technical Assistance Anthropometric Indicators Measurement Guide.12 Each measurement was taken in duplicate by the same investigator; intra-rater reliability can be found in table 1. If duplicate measures differed by more than 0.2 kg (weight), 0.5 cm (length/height and ulna and tibia lengths) or 3 mm (triceps skinfolds), a third measure was taken. Inter-rater reliability was determined by having the investigators take separate measurements on a subset of 30 participants (table 1).

Table 1

Intra-rater and inter-rater variability of anthropometric measurements taken in duplicate by two study investigators from children recruited from Botswana-Baylor Children's Clinic in Gaborone, Botswana

Anthropometric measures

Without shoes or excess clothing, children were weighed using a calibrated digital floor scale board (Seca, Birmingham, UK) to the nearest 0.1 kg. If a child could not stand on the scale unassisted due to physical conditions or a disability, or if they were frightened, the child was weighed indirectly by weighing the caregiver and the child.

Length of children under 2 years and height of children over 2 years were measured to the nearest 0.1 cm using a recumbent length board and stadiometer (Seca, Birmingham, UK), respectively.12

Ulna and tibia measurements were taken with both a long bone calliper (Rosscraft Campbell Caliper 20; Patent 4265021) and standard measuring tape. For ulna measures, the child's left arm was placed in front of the body on a flat surface with approximately 90° bend in the elbow. With the fingers extended and palm flat against the surface, the length of the ulna was measured from the proximal end of the ulna to the tip of the styloid process at the wrist.19 For tibia, the child's right leg was crossed over the left with the right ankle flexed over the left knee. The tibia length was measured from the proximal aspect of the right tibial plate and the distal end of the tibia at the ankle bone.20

MUAC was measured using a measuring tape to the nearest 0.1 cm. The midpoint of the child's left upper arm was located (and marked with a pen) by measuring the length from the tip of the child's shoulder to the base of the elbow (acromion and olecranon processes), found by bending the child's arm at a 90° angle.12 The MUAC measure was taken around the midpoint, while the child's arm was relaxed.12 Triceps skinfold was measured using a validated large skinfold calliper (Beta Technology; Patent 3008239) measured to the nearest 1 mm. With the left arm relaxed at the child's side, a vertical fold of skin was gently pulled away from the muscle at the arm's midpoint (parallel to MUAC mark) and calliper jaws were applied at a right angle to the midpoint.

Statistical analysis

Subjects with incomplete data (n=24) or data classified as extreme outliers in WHO Anthro software (n=1) were excluded from analysis. Descriptive statistics were computed for demographic characteristics and intra-rater and inter-rater variabilities. In order to develop and then test model fits for weight, the sample was randomly divided into two groups, groups A (n=387) and B (n=388), using a computer-generated random list. Independent sample t-tests and χ2 test were used to compare characteristics of participants in groups A and B. The strength of the relationship between calliper and measuring tape measures for ulna and tibia was assessed using bivariate Pearson's correlation. Univariate general linear models with anthropometric measures were used to identify significant predictors of weight. General linear models (with weight (kg) as the dependent variable) were computed with age, sex, HIV status and ARV use as covariates. Linear regression equations, including two of length/height, ulna length, tibia length and MUAC, were developed using group A. The best three models, selected based on the lowest Bayesian Information Criterion (BIC) and highest adjusted R2 values and their accompanying pragmatic models (simplified equations with numbers rounded) were used to predict weight of group B. For each of these models, arbitrary cut-points of the proportion within 5%, 10% and 15% of the actual weight of group B children aged 1–10 years (n=258; age validity range for other established weight-prediction equations) was determined. Weight-for-age, height-for-age, body mass index (BMI)-for-age and weight-for-height z-scores were calculated using WHO Anthro and WHO Anthro Plus software programs. Nomograms for the three best models were developed using R Foundation for Statistical Computing (Vienna, Austria). All other analyses were performed with SPSS for Macintosh (V.22.0; IBM, Armonk, New York, USA), except linear regression model BIC was calculated in Stata (V.13.1) for Macintosh (StataCorp, College Station, Texas, USA). Results were considered significant at p<0.05.


Intra-rater and inter-rater variability were low, falling within 5% (and in most cases <1%) for all anthropometric measures except triceps skinfolds, which had a mean inter-rater variability of 5.7% (table 1).

Demographic characteristics of study participants can be found in table 2. Groups A and B appeared similar. Children ranged in age from 18 months to 12 years and the majority were HIV-positive (81%) and receiving ARV (95%). Nearly one-quarter of participants were stunted (height-for-age <−2SD) and 7% were wasted (weight-for-height <−2SD), indicating a medium severity of malnutrition based on WHO classifications.21

Table 2

Demographic characteristics and weight-for-age, height-for-age, BMI-for-age and weight-for-height z-scores of study participants, together and segregated into randomly selected groups A and B, of children recruited from Botswana-Baylor Children's Clinic in Gaborone, Botswana

Correlation between long bone calliper and measuring tape measurements

Long bone callipers are the standard measuring tool for ulna and tibia19 ,20 but are expensive and difficult to access in resource-limited regions compared with inexpensive and ubiquitous measuring tapes. Measuring tape and long bone calliper measurements were strongly correlated for both ulna (n=775; r=0.99; p<0.001) and tibia lengths (n=775; r=0.99; p<0.001). With this strong correlation and knowledge that measuring tapes are more likely to be employed in resource-limited settings, measuring tape rather than calliper measurements were included in all further analyses.

Anthropometric predictors of weight

Linear regression was performed on group A (n=387) to determine the best predictors of weight and produce weight-prediction regression equations that could then be evaluated using group B (n=388). The following anthropometric measures were significant predictors of weight in univariate linear regression analysis: length/height, MUAC, ulna length and tibia length (all p<0.001; data not shown). When potential confounders (age, sex, HIV status, ARV use) were added as covariates in linear regression weight-prediction models, age (p=0.04) and HIV status (p<0.001) significantly affected the model for MUAC. Linear regression weight-prediction models including two each of MUAC, length/height, ulna length and tibia length are shown in table 3. HIV was not included in these models as it was not significant when placed into linear regression models with MUAC and tibia length (p=0.72), MUAC and ulna length (p=0.90) or MUAC and length/height (p=0.57). Age was not included because it is typically difficult to obtain accurately in resource-poor regions and/or emergency situations due to lack/loss of records.

Table 3

Predictors of weight (kg) in group A participants (n=387) recruited from Botswana-Baylor Children's Clinic in Gaborone, Botswana, using a linear regression model*

The adjusted R2 of all models in table 3 are between 0.87 and 0.96, which indicate that 87%–96% of the variance in outcomes could be explained by the variables included in the model. The BIC is also shown in table 3; lower BIC indicates better model fit. As such, the weight-prediction equations of the three models with the lowest BIC and highest adjusted R2, models 1, 2 and 4, were computed and can be found in table 4.

Table 4

Percentage agreement between actual weight of children in group B aged 1–10 years and predicted weights using models 1, 2 and 4 and weight-prediction models from the literature*†

Weight-prediction models: percentage agreement between predicted and actual weight

The agreement between weights predicted using equations from models 1, 2 and 4 and their accompanying pragmatic equations (rounded numbers) and actual weights of children aged 1–10 years in group B (n=258) are shown in table 4. As a comparison, actual weights are also compared with selected weight-prediction equations from the literature: the APLS,22 Luscombe and Owens,23 Theron et al,24 mid-arm circumference25 (MAC) and Leffler and Hayes26 equations. While ≥92% of predicted weight from any of models 1, 2 or 4 fell within 15% of actual weight, 55% or less (only 10% and 12% for Theron and Luscombe and Owens equations, respectively) fell within 15% of actual weight of equations from the literature. Pragmatic equations of models 1, 2 and 4 had poor predictive power, with only 47%, 51% and 4% of predicted weights, respectively, falling within 15% of actual weight.


Prediction equations from simple anthropometric measurements were shown to be accurate proxies of weight among this predominately HIV-positive group of Batswana children. Although this is to be expected because the equations were based on data derived from this population, as seen in table 4, the equations developed from this study more accurately predicted weight than currently commonly employed weight-prediction tools. The Broselow tape, a specialised length measuring tape (42.2–146.59 cm)27 labelled with colour-banded weight ranges rather than length to predict weight for paediatric drug dosing,28 is one of the most widely used weight proxy drug dosing tools.4 However, compared with actual weights, the Broselow tape only accurately categorised 62% (n=160 of 258) of Group B participants aged 1–10 years. A recent accuracy verification of the Broselow tape indicated that it was more accurate for children ≤25 kg;14 however, in this study, only 61% (n=133 of 219) of children aged 1–10 years weighing ≤25 kg in group B were correctly categorised. The Broselow tape more often overestimated (34%, n=88 of 258) than underestimated (4%, n=9 of 258) the weight of children aged 1–10 years in group B.

It was anticipated that some of the other commonly employed weight-prediction equations and tools derived from Western populations would not accurately predict weight among this population due to high HIV rates and medium severity of malnutrition.21 For example, both the APLS equation and Broselow tape were created using 1977 American National Center for Health Statistics data,2 and the Luscombe and Owen's formula amended the APLS formula using data from UK children to better reflect the heavier, ‘modern day child’.23 However, age-based and height-based tools cannot be dismissed altogether. A recent study created a new age and height banding system using the WHO growth standards and found that drug dosing for six commonly prescribed paediatric medications would have been accurate among children <5 years in Bangladesh and Uganda.3

The WHO Multicenter Growth Reference Survey highlighted that children around the world share milestones of weight and length/height regardless of ethnicity,29 suggesting that the equations developed here may be generalisable to other populations. However, specialised weight-prediction proxies may be required for HIV-positive children due to weight loss and failure to thrive (slow weight gain and linear growth) caused by HIV-mediated nutrient malabsorption, increased energy expenditure and limited appetite and/or food intake30–32 and greater risk of diarrhoea, tuberculosis and other comorbidities.33 However, 95% of HIV-positive children in this study were receiving ARV treatment, which has previously been shown to normalise growth.34 Although some ARV drugs may include lipoatrophy or lipohypertrophy among their adverse effects, the large sample size in this study obviated the need to consider such effects on an individual basis. While anthropometric weight-prediction equations developed in this study are highly predictive of weight among this group of predominately ARV-treated HIV-positive Batswana children, they should undergo further testing among other groups of HIV-positive children (both ARV-treated and otherwise), children without HIV, and in other resource-limited settings where growth faltering and/or malnutrition are common.

Pragmatic rounding of equations for easier calculation is common22 ,23 ,25 ,26 but may lead to considerable differences in predicted weights. Rounding coefficients in equations for models 1, 2 and 4 resulted in marked differences in percentage agreement between actual and predicted weights. For example, when model 1 equation 1.486MUAC+0.910T−27.980 was rounded to a more pragmatic (and more easily hand-calculated) 1.5MUAC+T−30, the accuracy within 15% of actual weight changed from 94% to 47% (47% difference). Similarly, within 10% and within 5%, accuracy changed from 80% to 40% (40% difference) and 52% to 31% (21% difference) when using model 1 versus pragmatic model 1, respectively. Therefore, creating tools that use exact equations, such as nomograms,18 may be better than rounding equations in maintaining accuracy of weight predictions. In addition, speed and ease of use, minimal training requirements and their use with commonly available supplies (ruler, string or even the measuring tape used to take anthropometric measures) are several of the advantages of nomograms over both full and pragmatic equations (requiring more training and a calculator) in the accurate prediction of paediatric weight (see figure 1). We envision use of such nomograms for both routine drug dosing and growth monitoring in resource-poor clinical settings and in emergency settings such as refugee camps.

Figure 1

Weight-prediction nomograms developed from models 1, 2 and 4 are shown.

Since ulna and tibia lengths were predictive of weight among this population, other long bones, namely the humerus, should also be investigated in future studies. Since the humerus is already measured as part of the MUAC measurement protocol,12 future research should explore relationships with humerus and MUAC to determine if one set of simple measures (MUAC, including humerus) rather than two (MUAC and ulna, tibia, or length/height) are as predictive of weight.

Height/length may be difficult to measure in some resource-poor settings as recumbent length boards and stadiometers are not always present. Since MUAC, ulna and tibia measures only require a simple measuring tape, models 1 and 2 are more ideal weight-prediction tools than model 4 for use in resource-poor settings.

We acknowledge that there are more sophisticated statistical methods than the split sample approach for validating predication models (ie, bootstrapping); however, we deliberately chose this split sample approach as it is more easily replicable in low-resource settings.

Until a calibrated weight scale is manufactured that remedies shortfalls for use in resource-poor settings, an accurate proxy of weight is required. An ideal calibrated weight scale should be inexpensive, durable and able to withstand extreme temperatures and other exposures (eg, sand, dirt, water), powered by a renewable resource (eg, solar) and easy to maintain and calibrate.4 ,11 The benefit of using other anthropometric measures as a proxy is that these additional measures that would not always be collected for drug dosing could also be used for growth monitoring and screening for malnutrition, which are both common concerns in resource-poor settings.


In the absence of a calibrated scale, anthropometric measurements including MUAC and length/height, ulna length or tibia length can be used to accurately predict weight and were more accurate than commonly employed age-based and height-based equations in this predominately HIV-positive group of Batswana children. Nomograms or other tools such as a mobile phone-based application using these prediction equations can be used to predict weight without losing equation accuracy. To determine if these equations are generalisable to other resource-poor settings, further research is needed.


The authors gratefully acknowledge the support of Keofentse Mathuba, Tebo Dipotso, Keboletse Mokete and Mogakolodi Clins Mmunyane of the Botswana-Baylor Children's Clinical Centre of Excellence for patient recruitment and demographic data collection. They also thank Dr Nathan J Lachowsky for statistical consultations.


View Abstract


  • Contributors Study protocol was developed by RW, MP, GA, SMM, CPL and TJG. Data collection was facilitated by GA, SMM and CPL and conducted by RW and MP. Data analysis was completed by KCW, RW, CDK, ZD and TJG. KCW and RW prepared the initial manuscript under the supervision of TJG and all authors reviewed and revised this work.

  • Funding This work was supported by the BC Children's Hospital Centre for International Child Health and the University of British Columbia Vitamin Research Fund.

  • Competing interests None declared.

  • Patient consent Obtained.

  • Ethics approval University of British Columbia/Children's and Women's Health Centre of British Columbia Research Ethics Board and the Health Research Development Committee of the Botswana Ministry of Health.

  • Provenance and peer review Not commissioned; externally peer reviewed.

Request Permissions

If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.

Linked Articles