Primary care giver and children's body-mass-index: A deep neural network model for use in primary paediatric care.

Abstract

Introduction: The familial environment is one of the major determinants of children's development of body weight during infancy and adolescence, in particular the socioeconomic status and anthropometric characteristics of the primary care giver. Thus, the aim of the present study is to utilise information on the familial environment of children and adolescents to identify an optimal prediction algorithm for estimating their expected body-mass-index (BMI) in the course of their development.

Methods: Data from Cohort '08 and Cohort '98 of the National Longitudinal Study of Children in Ireland are used (N = 37,960 and 27,499, respectively). The optimal prediction algorithm of children's BMI was identified by means of deep neural network models, with socioeconomic status and anthropometric characteristics of the primary care giver as predictors. Training and validation of the optimal model was performed with 80% and 20% of the total sample, respectively.

Results: The optimal deep neural network model yielded substantial improvements in prediction accuracy of children's BMI. The Pearson correlation between observed and predicted values obtained with the deep neural network was r=0.69, representing an improvement of about 50% in comparison to a simple linear model.

Conclusion: The predicted values of deep neural network models offer acceptable accuracy to be used as a communication tool in educational prevention programmes targeting families with children at higher risk of overweight and obesity in paediatric settings.