Stunting Reversal versus Regression to the Mean in Epidemiological Studies of Child Linear Growth in Low- and Middle-Income Countries

Early childhood undernutrition in low- and middle-income countries (LMICs) is marked by population-wide linear growth faltering, whereby most children grow in height more slowly than expected for their age and a relatively high proportion of children are classified as stunted based on a height-for-age z-score (HAZ) below -2. In epidemiological studies in LMICs, patterns of linear growth are often examined by longitudinal analyses of individual children's HAZ trajectories by age; in some studies, catch-up growth is defined as an upward HAZ threshold-crossing event such as recovery from stunting or stunting reversal (i.e., crossing from below to above HAZ = -2). In this Perspective, we question the suitability of this method of studying catch-up growth and highlight its susceptibility to regression to the mean (RTM), a statistical phenomenon whereby individuals with extreme values tend to move closer to the population mean at subsequent measurement timepoints. We use simulations of anthropometric datasets to demonstrate that upward HAZ threshold-crossing events (e.g., stunting reversal) are caused by RTM, and we discuss the advantages and pitfalls of other statistical methods to address RTM. In a scoping review of epidemiological studies of linear growth catch-up in LMICs (2004 to mid-2024; N = 69), we found that RTM was uncommonly mentioned (20%), and few studies (5/69; 7%) explicitly addressed RTM. In most studies (36/69; 52%), inferences about catch-up were based on methods that are at higher risk of RTM artifacts, namely by quantifying threshold-crossing events (n = 24) or by stratification of the sample by baseline size (n = 12). We conclude that more attention to RTM is needed in analyses of linear catch-up growth in LMICs. Published studies that used RTM artifact-prone measures should be remediated, and the use of HAZ threshold-crossing events such as stunting reversal should be avoided in future research.