Anthropometric prediction models of body composition in 3 to 24month old infants: a multicenter international study

The study addresses the challenge that standard anthropometric indices (e.g., weight-for-length, BMI-for-age) in infants do not distinguish between fat mass (FM) and fat-free mass (FFM) and often rely on population-based cut-offs that may not be biologically relevant or generalizable across ethnic and socioeconomic groups. Early-life body composition is closely linked to later non-communicable disease risk, and accurate, feasible measures are needed for clinics and large-scale studies. Prior work shows that combining anthropometric measures with factors such as sex, gestational age, and ethnicity can improve predictions, but multi-ethnic infant equations are scarce. The research goal is to develop and validate sex- and age-specific anthropometry-based equations to predict FM and FFM from 3 to 24 months in infants from diverse countries and ethnicities.

The paper highlights limitations of existing anthropometry for assessing infant adiposity, noting that indices like WFL and BMI do not differentiate FM from FFM and that cut-offs often lack biological relevance and are based on high-income or selected populations, leading to misclassification in certain ethnic groups (e.g., underestimation of overnutrition, overestimation of undernutrition). Prior studies have developed anthropometry-based prediction equations with varying generalizability; combining anthropometric measures, gestational age, sex, and ethnicity tends to improve prediction. The literature indicates ethnic differences in body composition, with South Asian infants often having lower FFM and higher FM than European ancestry infants early in life. The review underscores the need for age- and sex-specific equations, and for tools validated across diverse populations to better reflect body composition in the crucial first 1000 days.

Design and setting: Longitudinal, prospective, multicountry study (MIBCRS) following infants from birth to 24 months in lower-middle-income (India, Pakistan, Sri Lanka), upper-middle-income (Brazil, South Africa), and high-income (Australia) countries. Ethical approvals were obtained locally; informed consent was provided. Data collection occurred 2013–2019 with IYCF guideline-based feeding during follow-up. Samples: Main cohort (Brazil, Pakistan, South Africa, Sri Lanka) provided 3–24-month data for model development and internal validation. Participants were randomly split: training (two-thirds) and validation (one-third). Training: 1896 observations total (girls: 942 from 310; boys: 954 from 340). Validation: 941 observations (girls: 441 from 154; boys: 500 from 170). External test cohort: 349 observations from 250 infants aged 3–6 months in Australia (21 girls, 30 boys), India (44 girls, 46 boys), and South Africa (59 girls with 120 observations; 50 boys with 88 observations). Body composition reference: Deuterium dilution (DD) used to measure FM and FFM at 3, 6, 12, 18, and 24 months (development/validation) and at 6 months (test group). Anthropometry: At 3, 6, 9, 12, 18, and 24 months (development/validation) and 6 months (test). Weight (Seca 376), length (Harpenden infantometer; Seca 417 in India/Sri Lanka), triceps (TSFT) and subscapular (SSFT) skinfolds with Holtain caliper, head and mid-upper arm circumferences with Seca 212 tape. WHO-MGRS-based standardized protocols; intra- and inter-observer technical error monitoring; periodic standardization; DD training; inter-laboratory comparison for deuterium analysis. Data management: REDCap (University of the Witwatersrand). Statistical analysis: Separate sex-specific linear mixed models with random intercepts were fitted to predict FM (kg) and FFM (kg). Age modeled with linear splines with knots at 9 and 18 months based on visual trajectories. Predictors: age (with splines), length (m), weight-for-length (WFL, kg/m), TSFT (mm), SSFT (mm), and ethnicity (indicator for Non-South Asian vs South Asian). Head and arm circumference did not improve models and were excluded. 95% prediction intervals for training incorporated uncertainties in random effects, fixed effects, and residual variance; for validation and test, uncertainty in fixed effects and residual variance were used (via simulation). Performance metrics: RMSE, RMSPE, MAE, MAPE, Pearson correlation; count of observations outside prediction intervals. External validation on test data; Bland–Altman analyses assessed bias and proportional error. Sensitivity analyses: Alternative age specifications (quadratic; natural cubic splines), natural splines for all predictors (4 df), log-transformed outcomes, and substitution of WFL with BMI or ponderal index; compared via error metrics and conditional AIC. Software: R 3.6.1 (lme4, merTools, cAIC4).

• Developed sex-specific prediction equations for FM and FFM across three age bands: 3–9, 10–18, and 19–24 months, including an ethnicity indicator (Non-South Asian), length, WFL, TSFT, and SSFT with age splines at 9 and 18 months. Regression coefficients (Table 3) showed: length and WFL positively associated with FM and FFM; TSFT and SSFT positively associated with FM but negatively with FFM; age effects were non-linear with slope changes at 9 and 18 months. Marginal R²: FM 0.60 (boys), 0.56 (girls); FFM 0.90 (boys), 0.89 (girls). Conditional R²: 0.66–0.67 (FM), 0.91 (FFM). • Internal and external validation: RMSEs were similar across training, validation, and test sets. For FM (males): RMSE training 0.48 kg, validation 0.51 kg, test 0.55 kg; for FM (females): 0.46, 0.49, 0.52 kg. For FFM (males): 0.48, 0.51, 0.57 kg; for FFM (females): 0.47, 0.50, 0.54 kg (Table 5). • Error metrics: For FM, validation RMSPE and MAPE were higher than test; for FFM, test RMSPE and MAPE were higher than validation. MAPE for FFM was consistently <10% across groups (e.g., males: training 5.2%, validation 6.0%, test 9.1%). MAE for FFM ranged ~0.35–0.46 kg across groups. • Age-group trends: RMSEs increased in older age groups (e.g., 19–24 months: FM RMSE up to 0.66 kg in boys; FFM RMSE up to 0.66 kg in boys). • Country-specific performance: For test data (3–6 months), FFM RMSE showed best agreement with validation in South Africa (M/F 0.46/0.45 kg) compared to Australia (0.51/0.33 kg) and India (0.77/0.80 kg). For FM predictions, Sri Lanka (in training/validation cohorts) had higher RMSPE (boys 47.1, girls 55.2) and MAPE (boys 33.7, girls 36.3) than overall means. • Bland–Altman analyses: No proportional bias for FFM in males; mean bias −0.33 kg (underestimation). Evidence of proportional bias for FM in both sexes (r≈0.24–0.27) and for FFM in females (r = −0.26). • Sensitivity analyses: Alternative model forms (quadratic age; natural splines; log-transformed outcomes) produced similar prediction errors; models with natural splines for all predictors had best conditional AIC but comparable errors. Substituting BMI or ponderal index for WFL did not improve performance.

The study met its aim by creating and validating multi-ethnic, sex- and age-specific anthropometric equations to estimate infant FM and FFM, addressing limitations of conventional anthropometry that cannot partition weight into FM and FFM. The models demonstrated consistent performance across internal and external datasets, especially for FFM, indicating clinical and epidemiological utility where direct body composition techniques are impractical. Incorporating length, WFL, TSFT, SSFT, and ethnicity captured biologically relevant variation; skinfolds improved FM prediction but were inversely related to FFM after adjustment, aligning with physiological expectations. Non-linear age effects justified the spline approach, reflecting changing growth and tissue accretion dynamics across infancy. External validation highlighted that equations generalize better within similar populations; performance was strongest where cohorts overlapped (e.g., South Africa). Overall, the findings suggest these equations are best applied to FFM prediction, with FM derivable as weight minus FFM, thereby reducing error in FM estimation.

This multicenter, multi-ethnic study provides sex- and age-specific anthropometry-based equations to predict FM and FFM in infants aged 3–24 months. The equations showed good validity across internal and external datasets, with superior predictability for FFM versus FM. Given the diversity of contributing populations, these tools may improve assessment of early-life body composition in research and practice, particularly in resource-limited settings. The results underscore the importance of applying equations to populations similar to those used for model development. Future work could expand validation to additional regions and age windows, refine models with alternative functional forms (e.g., splines) or additional predictors where feasible, and evaluate clinical impact of using FFM-based assessments on early-life nutrition interventions.

• External validation was limited to the 3–6-month age range, whereas equations span up to 24 months. • Some country-specific performance varied (e.g., higher FM RMSPE/MAPE in Sri Lanka), suggesting population-specific differences may limit generalizability. • Bland–Altman analysis indicated systematic underestimation of male FFM (~0.33 kg) and proportional bias for FM in both sexes and for female FFM. • Although natural-spline models showed better conditional AIC, final equations used linear splines for interpretability and parsimony; alternative specifications might further improve fit. • Head and arm circumference were measured but excluded due to limited added value; unmeasured factors (e.g., gestational age at birth for all infants) may also influence prediction. • No a priori power analysis for equation development was conducted.