Oral Presentation Australian and New Zealand Obesity Society Annual Scientific Conference 2024

Development of risk prediction models to identify rapid weight gain during infancy: a machine learning approach (#43)

Miaobing Zheng 1 , Yuxin Zhang 1 , Rachel Laws 1 , Peter Vuillermin 2 , Jodie Dodd 3 , Li Ming Wen 4 , Louise A Baur 4 , Rachael Taylor 5 , Rebecca Byrne 6 , Kylie Hesketh 1
  1. Institute for Physical Activity and Nutrition, School of Exercise and Nutrition Sciences, Deakin University, Geelong, Victoria, Australia
  2. Barwon Health, Geelong, Victoria, Australia
  3. Discipline of Obstetrics and Gynaecology, The Robinson Research Institute, The University of Adelaide, Adelaide, South Australia, Australia
  4. School of Public Health and Sydney Medical School, The University of Sydney, Sydney, New South Wales, Australia
  5. Department of Medicine, University of Otago, Dunedin, New Zealand
  6. School of Exercise and Nutrition Sciences, Faculty of Health, Queensland University of Technology, Kelvin Grove, Queensland, Australia

Backgrounds: Compelling evidence shows that rapid weight gain(RWG) during infancy is highly predictive of subsequent obesity. Infant RWG is a useful proxy marker denoting future obesity risk that could facilitate obesity risk prediction from infancy. Leveraging novel machine learning approaches, this study aims to develop and validate risk prediction models to identify RWG by age one year using prenatal and early postnatal factors.

Methods: Pooled data from seven Australian and New Zealand cohorts (n=5233) were used. Predictors included maternal pre-pregnancy BMI, smoking status, gestational age, number of siblings, infant sex, birth weight, breastfeeding duration, and timing of solid introduction.  Eight machine learning algorithms were used to predict infant RWG, which was defined as a change in weight-for-age z-score ≥ 0.67 from birth to around age one year. Pooled data were randomly split into a training and a test dataset for model training and validation respectively. Model consistency was evaluated using five-fold cross-validation. Model predictive performance was evaluated by Area Under the Curve (AUC), accuracy, precision, sensitivity, and specificity.

Results: Average prevalence of infant RWG was 27%.  All machine learning methods showed acceptable to excellent discrimination with AUC ranging from 0.75 to 0.85. Accuracy, which indicates the overall correctness of the model, spanned from 0.70 to 0.78. Precision, that measures the model's ability to avoid false positives, ranged from 0.70 to 0.77. The range of sensitivity(true positive cases) and specificity(true negative cases) of all models was 0.68 - 0.80 and 0.65 - 0.78, respectively. Random Forest and Gradient Boosting Classifier algorithms showed the most favourable predictive accuracy.

Conclusion: Utilising prenatal and early postnatal factors, the resulting machine learning based prediction models showed acceptable to excellent ability to identify and distinguish between infants with or without RWG. Such models could be feasibly integrated in clinical practice for early obesity risk assessment.