ESPE Yearbook of Paediatric Endocrinology

Brief summary: Growth data from three independent longitudinal cohort studies (Gothenburg GrowUp 1974 (n 1596); Gothenburg GrowUp 1990 (n 1890); Edinburgh Growth Study (n 145)) were used to train machine learning (ML) to predict adult height (AH) based on growth measurements until the age of 6 years. Five ML algorithms were tested. A random forest model predicted best, with sex and height at age 3.4–6.0 years being the most influencing factors. The model was cross-validated using unrelated data and revealed a prediction accuracy of r=0.88, with an overprediction of AH in short children and an underprediction in tall children. The average prediction error for AH was −0.4±4.0 cm.

Assessing growth in children and predicting adult height is the daily business of a pediatric endocrinologist. Tools used are longitudinal height measurements that are recorded on growth charts to visualise an individual’s growth trajectory against a selected reference population and/or radiologic evaluation of skeletal maturation in comparison to a healthy reference population. In the latter approach, AH is then most often predicted according to Tanner-Whitehouse and/or Bayley-Pinneau methods, or the newer online version BoneXpert. Accuracy of these methods is generally quite disappointing with a standard error of about 5–6 cm.

Electronic health records are in standard use in most health care centers and accumulate huge amounts of patients’ data. The use of such data for personalized, real-time decision-making is currently implemented in many aspects of medicine. So it is nice to see that this is also the case for growth assessments and AH prediction. Remarkably, the ML algorithm developed by the investigators performed better for boys and girls at age 6 years than all other older classic prediction methods. It is able to explain 75–77% of the variability in AH, while the remaining variability is due to pubertal growth and remains unexplained and unpredictable by the algorithm.

This study published in 2021 shows that Ze’ev Hochberg never got tired to learn and apply novel methods to his research which he intended to translate into clinical practice in order to improve children’s healthcare. Moreover, the collaborative network between Haifa, London and Goteborg supporting this study shows that he was an internationally acknowledged team player and leader.