Development and validation of anthropometric-based fat-mass prediction equations using air displacement plethysmography in Mexican infants

Early infancy body composition may program later metabolic disease risk, making accurate fat mass (FM) measurement from birth important. While air-displacement plethysmography (ADP) provides valid FM estimates, it is costly and not widely accessible, limiting its clinical and population use. Anthropometry offers a simple, low-cost alternative but requires training to ensure accuracy. Prior studies indicate multiple factors predict infant FM, including sex, age, weight, length, circumferences, and skinfolds. BMI correlates strongly with adiposity in adults and moderately in infants but does not reflect FM amount or distribution. Several FM prediction equations exist for neonates and older children, but there is a gap for the first months of life, and existing equations often perform poorly across ages and ethnicities. Few equations exist for 0–4 months and none specifically for Hispanic/Mexican infants. This study aimed to develop and validate anthropometry-based FM prediction equations for Mexican infants at 1, 3, and 6 months, using ADP as the reference method.

The literature highlights the importance of infant FM for later health and the need for age- and ethnicity-specific body composition data due to rapid growth and variability in early infancy. Evidence on sex differences in infant FM is mixed. ADP is validated for infant body composition but is impractical for routine use; anthropometry is more accessible but requires standardization. Prior FM prediction equations for neonates and children show variable accuracy, with some developed for Caucasian, Asian, and German infants, and multiethnic samples. Equations developed for older children perform poorly when applied to infants, often with significant bias and wide limits of agreement. BMI alone is limited in infancy as it does not capture FM amount/distribution; adding skinfolds and circumferences improves prediction. Studies suggest combinations including calf circumference, weight, and sums of skinfolds yield better FM estimates. Skinfolds are informative but measurement can be error-prone in infants; circumferences are simpler and more reproducible. The literature underscores the need for population- and age-specific equations incorporating multiple anthropometric measures.

Design and cohort: Secondary analysis within the OBESO (Origen Bioquímico y Epigenético del Sobrepeso y la Obesidad) perinatal cohort at the Instituto Nacional de Perinatología, Mexico City (2017–ongoing). Ethics approval Project No. 3300-11402-01-575-17; informed consent obtained. Subjects: Healthy-term newborns (≥37 weeks gestation) born to healthy adult women (≥18 years, without preexisting disease), with complete anthropometry and FM data at 1, 3, and/or 6 months. Exclusions: Postnatal/congenital diseases, adverse maternal outcomes (gestational diabetes, preeclampsia), preterm infants, and FM values <5% (below reported minima). Final sample per visit: 1 month n=133 (67 girls), 3 months n=105 (51 girls), 6 months n=101 (53 girls). Clinical data included gestational age based on first-trimester ultrasound. FM reference method: ADP measured with the PEAPOD device (COSMED, USA), calibrated per manufacturer. Infant weight measured unclothed with cap, then body volume measured; body density computed; FM (kg) calculated via PEAPOD software using Fomon’s equation. Anthropometry: Measurements by two trained nutrition professionals in duplicate (averaged) following Lohman’s methodology. Time points: birth (24–72 h), 1, 3, and 6 months. Weight (Tanita pediatric scale), length (SECA infantometer, Frankfort plane), BMI computed (weight/length²). Circumferences: head (maximal occiput–glabella), mid-upper arm (midpoint acromion–olecranon), thigh (midpoint greater trochanter–patella), calf (widest point), waist (umbilicus after exhalation); measured to nearest mm with Lufkin tape. Skinfolds: biceps, triceps, subscapular (45° fold), waist, thigh, calf using Lange caliper; duplicate (third if >2 mm difference). Statistical analysis: Three-step model development for 1, 3, 6 months separately. (1) Variable selection using LASSO on training subsets; data stratified into quintiles; 80% per quintile randomly sampled for training. Due to LASSO variability, conducted 100 blocks of 12 repetitions (1,200 beta estimates per variable); selected variables with ≥500 non-zero betas. Candidate variables: sex, gestational age at birth, weight, length, BMI, circumferences (head, waist, arm, thigh, calf), skinfolds (biceps, triceps, subscapular, waist, thigh, calf). (2) Model behavior evaluation via 12-fold cross-validation using quintile-based splitting (70% train, 30% test). Models trained with Theil–Sen regressions; partial models per fold saved; final model coefficients defined as median of partial betas. (3) Final model evaluation: Bland–Altman plots for agreement (bias, limits of agreement), Deming regression for method comparison (testing whether intercept≈0 and slope≈1), and comparisons of mean predicted vs measured FM (t-test or Mann–Whitney as appropriate). Software: Python 3.9 and scikit-learn 1.1.1 for data processing and model training; R for Bland–Altman (v0.5.1), Deming regression (v1.4), and statistical comparisons (R 4.2.1).

- Sample: From 348 cohort participants, data available for 292 newborns; excluded 120 healthy infants without complete data/follow-up and 33 preterm. Excluded FM<5% from 7 infants (1M n=6; 3M n=1). Final per visit: 1M n=133, 3M n=105, 6M n=101. Mean gestational age 39.00 (1.06) weeks. - Anthropometry/body composition: FM approximately doubled from 1M to 3M and increased ~33% from 3M to 6M; no sex differences in FM. - Variables retained in final equations: • 1 month: thigh circumference (TC), BMI, waist skinfold (WSF), thigh skinfold (ThSF), subscapular skinfold (SSF). • 3 months: waist circumference (WC), TC, calf circumference (CC), BMI, calf skinfold (CSF), SSF, triceps skinfold (TSF), gestational age at birth. • 6 months: TC, CC, WSF, ThSF, SSF, gestational age at birth. - Model performance: R² values: 1M: 0.54; 3M: 0.69; 6M: 0.63. Correlations between predicted and ADP-measured FM were high (1M r=0.734; 3M r=0.831; 6M r=0.791; all p<0.001). - Agreement and mean differences: No significant differences between mean predicted vs measured FM: • 1M: 0.62 vs 0.60 kg (p=0.77). • 3M: 1.20 vs 1.35 kg (p=0.80). • 6M: 1.65 vs 1.76 kg (p=0.55). Bland–Altman bias (95% CI): • 1M: −0.021 (−0.050 to 0.008); LoA: −0.352 (−0.401 to −0.302) to 0.310 (0.260 to 0.359); out of LoA: ~3.01% lower and upper. • 3M: 0.014 (0.090 to 0.195); LoA: −0.377 (−0.466 to −0.287) to 0.663 (0.573 to 0.752); out of LoA: 2.97% lower, 1.98% upper. • 6M: 0.108 (0.046 to 0.169); LoA: −0.478 (−0.584 to −0.373) to 0.694 (0.589 to 0.800); out of LoA: 1.08% lower and upper. - Deming regression: High correlations observed; intercept/slope did not meet equivalence (intercept≈0, slope≈1) at 1M, 3M, 6M (e.g., at 3M slope 0.743 [0.612–0.875], intercept 0.215 [−0.344 to 0.775]). - Equations (from Table 2): • 1M: FM (kg) = 0.068·TC + 0.018·BMI + 0.026·WSF + 0.010·ThSF + 0.009·SSF − 1.082. • 3M: FM (kg) = 0.006·WC + 0.074·TC + 0.078·CC + 0.062·BMI + 0.024·CSF + 0.054·SSF − 0.062·TSF − 0.053·(gestational age at birth) − 1.045. • 6M: FM (kg) = 0.030·TC + 0.163·CC + 0.023·WSF + 0.034·ThSF + 0.019·SSF − 0.050·(gestational age at birth) − 0.858.

The study developed anthropometry-based FM prediction equations for Mexican infants at 1, 3, and 6 months, addressing a gap in age- and ethnicity-specific tools for early infancy. Predicted FM correlated strongly with ADP measurements, and models explained roughly 54–69% of FM variability. Despite high correlations, Deming regression indicated nonequivalence (slope/intercept not meeting 1/0), and Bland–Altman analyses showed small but significant mean biases at 3 and 6 months, suggesting slight underestimation on average. Including multiple anthropometric dimensions (BMI, circumferences from trunk and limbs, and several skinfolds) likely enhanced prediction performance relative to reliance on BMI or few measures alone. The use of absolute FM (kg), rather than percentage, aligns with recommendations for anthropometric modeling. Month-specific equations account for rapid growth dynamics and changing FM accretion in early months, improving applicability. Given the cost and limited availability of ADP, these equations provide a practical alternative for clinical and research settings in similar populations, although care is needed with skinfold techniques and recognition of modest underestimation at older ages in this range.

Anthropometry-based equations can provide accessible, inexpensive estimates of infant FM when reference methods are unavailable. The proposed month-specific equations for Mexican infants in the first 6 months showed high correlations with ADP-measured FM and acceptable agreement, supporting their use in comparable clinical and research contexts. Future research should prioritize establishing FM cutoffs linked to metabolic risk, evaluating acceptable clinical variation, and externally validating and simplifying models for broader clinical feasibility across diverse populations.

- Sample size was relatively small, which may limit model prediction power and generalizability. - Models include multiple anthropometric measures, some of which (especially skinfolds) require trained personnel, standardized technique, and specific equipment, potentially limiting routine clinical implementation. - Significant mean bias was observed at 3 and 6 months, indicating modest underestimation; acceptable clinical variation thresholds are not clearly established. - External validation in other Hispanic/Mexican populations and across different settings was not performed.