Length-for-age z-scores (LAZ) and stunting prevalence (%LAZ

We simulated a synthetic cohort with a harmonically downward-shifting LAZ trajectory from birth to 24 months of age, with mean LAZs similar to the HBGDki pooled South Asian cohorts, and without any input parameters intended to differentially affect individuals’ growth across the height distribution or at different ages. We compared HBGDki empirical estimates of age interval-specific frequencies of incident stunting onset and stunting reversal with those from the synthetic cohort. Using synthetic cohorts, we examined how estimates of incident onset and reversal were affected by missing data, magnitude of the whole-population shift in the LAZ distribution and strength of the between-time-point correlation. We also compared the 3–24 month pattern of linear growth faltering expressed as age-related trajectories of average growth delay (chronological age minus height–age), mean LAZ or stunting prevalence.

Empirical estimates of age interval-specific incident stunting onset and stunting reversal in the HBGDki cohorts were similar to those observed in a synthetic cohort. Variability in LAZ threshold-crossing event rates is explained by starting LAZ, between-time-point correlation and the magnitude of the whole-population shift in the LAZ distribution. Incident stunting onset is also affected by missing data in preceding intervals. Stunting reversal occurs due to within-child variability (ie, imperfect between-time-point correlation) in the absence of any other phenomena that cause stunted children to become non-stunted at a later age. The linear growth faltering pattern based on growth delay differed from corresponding age-related trajectories of mean LAZ or stunting prevalence.

In longitudinal studies of linear growth faltering in LMICs, LAZ threshold-crossing indicators are byproducts of whole-population shifts in LAZ and within-child variability and should be interpreted accordingly. Reporting incident stunting onset and reversal rates, or analyses in which children are grouped by the timing of LAZ threshold-crossing events, may detract from efforts to understand when and why nearly all children in LMICs grow more slowly than expected for their age. Since mean LAZ and stunting prevalence are unsuitable for quantifying the rate and timing of population-average postnatal linear growth faltering, growth delay is recommended for consideration as a preferred metric.