A big-data approach to producing descriptive anthropometric references: a feasibility and validation study of paediatric growth charts
Résumé
Background: Both national and WHO growth charts have been found to be poorly calibrated with the physical growth of children in many countries. We aimed to generate new national growth charts for French children in the context of huge datasets of physical growth measurements routinely collected by office-based health practitioners.Methods: We recruited 32 randomly sampled primary care paediatricians and ten volunteer general practitioners from across the French metropolitan territory who used the same electronic medical records software, from which we extracted all physical growth data for the paediatric patients, with anonymisation. We included measurements from all children born from Jan 1, 1990, and aged 1 month to 18 years by Feb 8, 2018, with birthweight greater than 2500 g, to which an automated process of data cleaning developed to detect and delete measurement or transcription errors was applied. Growth charts for weight and height were derived by using generalised additive models for location, scale, and shape with the Box-Cox power exponential distribution. We compared the new charts to WHO growth charts and existing French national growth charts, and validated our charts using growth data from recent national cross-sectional surveys.Findings: After data cleaning, we included 1 458 468 height and 1 690 340 weight measurements from 238 102 children. When compared with the existing French national and WHO growth charts, all height SD and weight percentile curves for the new growth charts were distinctly above those for the existing French national growth charts, as early as age 1 month, with an average difference of -0·75 SD for height and -0·50 SD for weight for both sexes. Comparison with national cross-sectional surveys showed satisfactory calibration, with generally good fit for children aged 5-6 years and 10-11 years in height and weight and small differences at age 14-15 years.Interpretation: We successfully produced calibrated paediatric growth charts by using a novel big-data approach applied to data routinely collected in clinical practice that could be used in many fields other than anthropometry.