Accuracy of Using Weight and Length in Children under 24 mo to Screen for Early Childhood Obesity: A Systematic Review

TL;DR

This systematic review evaluates how well weight and length measurements in children under 24 months can predict obesity later in childhood.

Contribution

The study systematically reviews and compares the accuracy of anthropometric and multi-predictor models for early obesity screening in children.

Findings

Anthropometrics-alone models had AUCs ≥ 0.56 with expert guidance and ≥ 0.77 with machine learning.

Multiple-predictor models showed AUCs ≥ 0.68 with expert guidance and ≥ 0.76 with machine learning.

Models were more accurate with older initial assessments, multiple assessments, and shorter prediction windows.

Abstract

The global increase in early childhood overweight and obesity has prompted interest in early prediction of overweight and obesity to allow timely intervention and prevent lifelong consequences. A systematic review was conducted to assess the accuracy and feasibility of predicting overweight and obesity in individual children aged 3–7 y using data available in healthcare and community settings on children aged under 24 mo. This review was registered in PROSPERO (CRD42024509603) and followed the Preferred Reporting Items for Systematic Reviews and Meta-Analyses guidelines. From 7943 unique articles identified through PubMed, CINAHL, Scopus, and Google Scholar, 14 studies met the inclusion criteria, 13 from high-income countries and 1 from a middle-income country. These studies evaluated the accuracy of predicting childhood overweight or obesity in individual children using anthropometrics-alone or multiple-predictor models. Anthropometrics-alone models yielded areas under the curve (AUCs) ≥ 0.56 with expert guidance and ≥0.77 with machine learning. Multiple-predictor models yielded AUC ≥ 0.68 with expert guidance and ≥0.76 with machine learning. The inclusion of child, parental, and community predictors improved predictive accuracy but led to greater variation in performance across models. Models were more accurate when children were older at the initial assessment, multiple assessments were made, and the time between assessment and outcome prediction was shorter. Prediction models with an AUC ≥ 0.70 used machine learning to optimize variable selection, limiting their practicality for broad-scale implementation in healthcare or community settings. There is insufficient evidence on the accuracy of overweight and obesity prediction models for children in low- and middle-income countries. Existing prediction models are not well-suited for broad-scale screening of individual children for risk of early childhood overweight or obesity.

Figures