Growth Charts: From Pen and Paper to Computerised Algorithms to Detect Aberrant Growth

Petur B. Juliusson Mathieu Roelants Normal growth of children reflects good health, but a deviation of length/height, weight or head circumference from the expected track is suggestive of underlying health problems. Monitoring the growth of children is therefore a well-established practice in many countries, including the Nordics. In Norway, for example, 14 routine measurements are scheduled between birth and the age of 8 years [1]. Measurements are usually plotted on growth charts that show the distribution of a trait in healthy peers to assess whether they match expectations. This is equivalent to comparing a single measurement with the population. Growth monitoring guidelines include a set of rules that 'flag' possible aberrant growth in children below or above the statistical distribution at a given age (the trait being length/height, weight, head circumference alike), when they deviate from the target height defined by the height of their parents, or when abnormal crossing of centiles is detected. These children are then scheduled for follow-up or referral to a paediatric specialist. Such guidelines are largely based on empirical, clinical evidence but often lack a sound scientific basis. An evidence-based approach is, however, difficult to achieve as growth impairment differs between the many conditions that may impact the growth of children. We recently explored the growth trajectories in healthy children and compared these with existing Norwegian guidelines, which illustrated that while many rules were functional in terms of the number of children flagged, some rules identified clearly too many children, putting an unnecessary burden on the children, parents and healthcare workers [2]. However, despite the relatively high frequency of routine measurements and growth monitoring guidelines that often refer to too many rather than too few children, growth disorders can still be missed or diagnosed late. In other words, guidelines for 'flagging' children should be carefully chosen to optimise both sensitivity and specificity, considering the many different conditions. Automated screening procedures appear to be a promising approach that could optimise the yield of growth monitoring programs. Plotting of anthropometric measurements of children on growth charts has been practiced for many decades. With the introduction of electric patient journals, the charts were also digitised, but the underlying practice remained unchanged because we use the charts on the screen in the same way as on paper. The possibility of using algorithms to 'flag' aberrant growth in digitised measurements seems to be used only reluctantly, if at all. However, recently, some interesting work on this very topic was done in Finland, where growth-monitoring algorithms have been tested as part of an automatic screening program [3]. The Finnish algorithms consider several growth parameters simultaneously: the height-for-age standard deviation score (height SD), distance between this score and target height SDS, the change in height SDS (height velocity), BMI SDS and change in BMI SDS. In this screening program, growth considered abnormal by the automatic procedure was evaluated by a paediatric endocrinologist followed up by an electronic consultation with the GP [3]. This procedure was shown to significantly improve the identification of abnormal growth development when compared with conventional monitoring [3], including evidence for earlier detection of coeliac disease and juvenile acquired hypothyroidism [4, 5]. In the study by Lars Gelander and co-authors, published in Acta Paediatrica [6], the Finnish growth monitoring algorithms were applied to Swedish data and compared with the Swedish growth monitoring guidelines and IOTF cut-offs for obesity. As expected, far fewer children were identified as short or tall using the Finnish growth screening algorithms compared to the much broader Swedish criteria. More children were identified with obesity using the IOTF30 criteria, compared with the Finnish 99.5% specificity criteria and, interestingly, 26% of the children were identified as in need of medical intervention before they reached the IOTF30 cut-off. Gelander et al. conclude that such algorithms could be a promising approach in Sweden and other countries, but the accuracy of the referrals in their data is still unknown because that requires a prospective collection of clinical data. The analysis in the paper by Gelander et al. was limited to prepubertal years, but puberty adds a lot of variability to growth because of individual differences in the timing of puberty onset and therefore demands special consideration. Currently, large amounts of routinely collected data on the growth of children are recorded into software in child health centres and school healthcare settings, and 'clinical growth data' as a part of the follow-up in patient journals from specialist health services. In the future, we hope to see electronic growth monitoring applications that include algorithms that flag children with aberrant growth in need of follow-up or further investigation. Ideally, all growth measurements performed in primary and specialist health care services alike would be accessible in a central database. Such algorithms have the potential to process a much larger amount of information, including past growth, dynamics of growth and auxiliary information such as parental height. Validated algorithms based on multiple growth parameters may contribute to the detection of growth patterns associated with specific underlying conditions. Documentation of this process opens up the possibility of automatic feedback on diagnosis, care and outcome, and as such allows us to validate and even improve the criteria. Although we should not forget that growth as such is rarely diagnostic, but more like a triage of children for further scrutiny by a clinician, the use of electronic growth monitoring algorithms facilitates a prompt diagnosis at an early stage and improvement in the quality and equity of the associated healthcare. The reports from Finland have paved the way, and the current study by Gelander et al. challenges us to realise these ambitions. The authors declare no conflicts of interest. The authors have nothing to report.