A novel development indicator based on population-average height trajectories of children aged 0–5 years modelled using 145 surveys in 64 countries, 2000–2018

Abstract

IntroductionChildren’s growth status is an important measure commonly used as a proxy indicator of advancements in a country’s health, human capital and economic development. We aimed to assess the feasibility of using Super-Imposition by Translation And Rotation (SITAR) models for summarising population-based cross-sectional height-by-age data of children under 5 years across 64 countries.MethodsUsing 145 publicly available Demographic and Health Surveys of children under 5 years across 64 low-income and middle-income countries from 2000 to 2018, we created a multicountry pseudo-longitudinal dataset of children’s heights.ResultsSITAR models including two parameters (size and intensity) explained 81% of the between-survey variation in mean boys’ height and 80% in mean girls’ height. Size parameters for boys and girls (relative to the WHO child growth standards) were distributed non-normally around a mean of −5.2 cm for boys (range: −7.9 cm to −1.6 cm) and −4.9 cm for girls (range: −7.7 cm to −1.2 cm). Boys exhibited 10% slower linear growth compared with the WHO (range: 19.7% slower to 1.6% faster) and girls 11% slower linear growth compared with the WHO (range: 21.4% slower to 1.0% faster). Variation in the SITAR size parameter was ≥90% explained by the combination of average length within the first 60 days of birth (as a proxy for fetal growth) and intensity, regardless of sex, with much greater contribution by postnatal intensity (r≥0.89 between size and intensity).ConclusionsSITAR models with two random effects can be used to model child linear growth using multicountry pseudo-longitudinal data, and thereby provide a feasible alternative approach to summarising early childhood height trajectories based on survey data. The SITAR intensity parameter may be a novel indicator for specifically tracking progress in the determinants of postnatal growth in low-income and middle-income countries.

Complementary and alternative metrics for tracking population-level trends in child linear growth

“…We also included the Super-Imposition by Translation and Rotation (SITAR) intensity parameter [ 14 ], which is a scaling factor that reflects the velocity of a linear growth curve relative to the mean velocity. The SITAR size parameter was not included as we previously found it to be nearly interchangeable with <5y mean HAZ [ 15 ]. Metrics reported in the literature that incorporate stunting prevalence rather than present new statistical expressions of height-age distributions (e.g., “Composite Index of Anthropometric Failure”) were not considered [ 16 ].…”

“…The SITAR growth model was then fit for each DHS survey in comparison to the mean growth curve of all surveys. The intensity parameter, which represents the velocity of a growth curve as a proportion of the velocity of the mean growth curve, was estimated from the SITAR model for each survey as previously described [ 15 ]. All descriptive and model-derived metrics were generated accounting for DHS survey sampling design using strata, cluster, and sample weights.…”

“…Where α i is a random height effect for population i, h is a cubic B-spline function, β i is a random age effect, γ i is the intensity parameter representing how compressed or expanded the growth rate of one observed population i is compared to the global average growth rate, and ε it are residuals [14,15].…”

Prenatal and postnatal exposure to NO2 and rapid infant weight gain – A population‐based cohort study

“…The SITAR method has been implemented for modelling anthropometric growth as captured by various measures, during different age ranges and populations, and has been validated against other methods of growth modelling 41–45 . The SITAR method is based on a shape‐invariant, nonlinear mixed‐effects model, which estimates the average growth curve in the population alongside up to three random‐effect parameters per individual.…”

“…or 6 weeks,39,40 and our SITAR model at 5 weeks (FigureS2). In addition, we examined exposure to NO x , which is comprised mainly of NO 2 and NO and is highly correlated with NO 2 exposure (Pearson's r = 0.96).2.4 | Statistical analysis2.4.1 | Modelling the outcomeThe SITAR method has been implemented for modelling anthropometric growth as captured by various measures, during different age ranges and populations, and has been validated against other methods of growth modelling [41][42][43][44][45]. The SITAR method is based on a shape-invariant, nonlinear mixed-effects model, which estimates the average growth curve in the population alongside up to three randomeffect parameters per individual.…”