Sep 5, 2026
2:00 PM12:00 PM
ePoster display
PD074
COMPARATIVE ACCURACY OF TRADITIONAL PREDICTIVE EQUATIONS AND NOVEL PHOTO-BASED MODELS FOR ESTIMATING WEIGHT IN HEALTHY CHILDREN: A PILOT STUDY
L. Buyle1,*, C. Schoemans 2, B. Jansen 2, K. Huysentruyt1
1Department of Paediatric Gastro-enterology, Hepatology and Nutrition, UZ Brussels, 2Department of Electronics and Informatics (ETRO), Vrije Universiteit Brussel (VUB), Brussels, Belgium
Rationale: Traditional weight predictive equations (WPE) do not accurately predict weight. This pilot study compares traditional weight prediction formulas with a novel photography-based model to evaluate their accuracy in healthy children. The overall aim of this study is to utilize this algorithm for accurate weight prediction enabling rapid medical decisions.
Methods: Data from 86 healthy children (5-9 years) were analysed. Prediction errors were categorized into predefined bands. Acceptable error was defined as ≤5% difference. We estimated SMPL-based weight from 2D images and corrected residuals using Hybrid Huber regression with leave-one-out cross-validation.
Results: Data from a total of 86 children (45, 52% girls) for the WPE were analyzed. Data from 84 children were analyzed for the photo-based model. The mean (SD) weight, height, BMI, mid upper arm circumference, triceps skin fold and subscapular skin fold z scores were respectively 0.14 (0.99), 0.26 (1.17), 0.01 (0.93), 0.00 (0.96), -0.05 (0.76) and -0.06 (0.89). Table 1 shows that the photo-based model has the highest percentages of accurate estimates (69.8%) followed by formula CEEW1 (55.8%). The Argall, Luscombe and APLS-ALT WPE were more prone to major overestimation and the APLS WPE to major underestimation of the measured weight.
Image:
Conclusion: Most weight prediction equations grossly over- or underestimate the actual weight in children. Our photo-based model provided the most accurate weight predictions.
Disclosure of Interest: None declared