Longitudinal growth in early childhood and vulnerability to infection and mortality

Abstract

Despite substantial improvements in child survival in the last 30 years, an estimated 5 million children still died before their 5th birthday worldwide in 2019.1,2 At the current trajectory of decline, the world will not meet Sustainable Development Goals for infant and child mortality.2,3 About 80% of under-5 deaths were due to birth, nutrition, or infectious disease-related causes.1 However, there are gaps in our understanding of how infectious diseases and other factors in pregnancy, birth, and infancy can affect vulnerability to future infections, survival, and healthy development of infants and children.4–9 Longitudinal analyses of height and weight in children can strengthen our understanding of the magnitude of infections on vulnerability and improve resiliency in child development.10 Childhood infectious disease vulnerability is likely a positive feedback loop. A child that is born premature might face obstacles in catching up to reference growth standards simply due to starting in a smaller place.11,12 Further, underweight and short stature are both associated with a greater risk of mortality due to infectious diseases.13,14 Infectious diseases broadly, and some pathogens specifically, are also associated with linear growth faltering.15–17 In this mechanism, vulnerability might reinforce itself through higher risk of infection, leading to greater vulnerability to future- and potentially more severe- infections. Being malnourished is most impactful in the first year of life, leading to increased risk of mortality from infectious causes and suppressing linear growth and cognitive development at a crucial period.18 There are several key terms that will be used throughout this dissertation. Childhood growth failure (CGF) metrics typically include height-for-age, weight-for-age, and weight-for-height z-scores (HAZ, WAZ, WHZ, respectively). These metrics are constructed from global growth reference standards. Given an individual child’s height and weight, each can be calculated based on the distance between those values and the growth reference median based on age and sex. Representing height and weight as z-scores is very useful because it removes some of the complexity of modeling non-linear growth trajectories and because it enables instantaneous comparison of a child’s growth with a reference. Children that are mildly (-1 to -2 z-scores), moderately (-2 to -3 z-scores), or severely (< -3 z-scores) below the global reference have worse health and developmental outcomes than children who are not. Stunting describes children with low HAZ, underweight describes children with low WAZ, and wasting describes children with low WHZ. Each has potentially important interpretations. For example, stunting is often considered a proxy indicator for chronically poor growth, associated with environmental and socio-demographic barriers to healthy development, and may predispose children to poor cognitive and educational attainment.9,19–21 In contrast, WHZ is most frequently used to identify children with severe acute malnutrition, a devastating condition that requires urgent medical care and therapeutic intervention to prevent deaths directly due to malnutrition like organ failure.18 This dissertation sought to systematically review and apply statistical approaches to measuring longitudinal childhood growth and vulnerability to all-cause and infectious disease-specific incidence and mortality. It is structured in four distinct chapters, representing three specific aims. A short summary of each chapter follows. Each is structured as an academic journal article and includes tables, figures, and references specific to each chapter. This document concludes with an overall reflection and proposes next steps related to this work. The first chapter of this dissertation, Statistical methods for repeated measurements of childhood growth, a systematic review and analysis, describes how the academic literature has used specific methodologies and approaches to quantify longitudinal childhood growth. Specifically, we sought to identify statistical models for accounting for repeated measurements of height, weight, head circumference, and mid-upper arm circumference among children. We identified 12,355 articles in PubMed that matched our search criteria. Of these, 2,640 were included in a full text review and 1,301 publications had repeated measurements of childhood growth and an informative description of the statistical methods used to growth on its own or how growth was associated with other outcomes such as cognitive development or mortality. Most studies (59.6%) used statistical methodologies that explicitly accounted for correlation among measurements within individual children, largely multilevel or hierarchical models, generalized estimating equations, or latent class models. Analyses of longitudinal outcomes like growth and infections should account for variation within and between individuals to appropriately make the most reliable statistical models.10 Statistical models that account for dependence in repeated observations are critical to producing valid inference because not accounting for this violates assumptions of independent and identically distributed observations in linear and generalized linear regression. There has been an increase in the number of studies that analyzed repeated measurements of childhood growth published since 2010 including wider geographical representation of the studies that collected those data. This interest in longitudinal childhood growth is exciting and research must use appropriate methods to analyze the data because it makes the results more reliable and valid and better strengthens our understanding of why growth varies between children. This review and synthesis provided a quantitative review of these methods, how they differ between studies, and provided advice for how to consider repeated measures of childhood growth in future analyses. The second chapter of this dissertation, Effects of childhood growth failure on cause-specific infectious disease incidence and mortality: a Burden of Proof study, estimated a continuous risk curve for low height-for-age, weight-for-height, and weight-for-height z-scores (HAZ, WAZ, and WHZ, respectively) for diarrhea, lower respiratory infections, malaria, and measles incidence and mortality. Previous efforts to quantify the risk of these infectious disease outcomes were based on categorical definitions of childhood growth failure (CGF) and were limited to fatal outcomes.13 In this chapter, we describe how, for the first time, a continuous risk curve for each of the three CGF indicators was produced using a flexible, hierarchical, Bayesian meta-regression tool. We applied the Global Burden of Disease study (GBD) Burden of Proof criteria to our models and show that, with high confidence, low HAZ, WAZ, and WHZ are significant risk factors for mortality due to diarrhea, lower respiratory infections, and measles. Further, our analysis indicates novel associations between childhood growth metrics and the risk of malaria mortality and the risk of infectious disease incidence for each of the distinct infectious diseases. The results from this work are part of the GBD 2021 estimates of the burden of childhood growth failure for 195 countries, 5 age groups, and by sex from 1990-2021. These updated burden estimates are described in the third chapter, Quantifying the fatal and non-fatal burden of disease attributable to childhood growth failure: An analysis from the Global Burden of Disease study 2021. We found that the burden of CGF is not equally distributed geographically, with substantial health loss attributable to infectious disease incidence and mortality concentrated in sub-Saharan Africa and South Asia. We show that, despite reductions in the prevalence of CGF in many locations, the percent attributable burden hasn’t declined as fast as the attributable burden rate, suggesting that reducing CGF is essential to accelerate declines in under-5 child mortality. Reducing its burden may be necessary but insufficient for countries to meet Sustainable Development Goals for under-5 mortality, demonstrating that interventions to reduce mortality due to other causes, especially neonatal causes of death, are also required to meet international goals for child survival. The two papers of this aim will both be submitted for peer review and publication in an academic journal and the second will be shared with the GBD Collaborator Network, a group of thousands of experts in disease burden and epidemiology, for critical feedback prior to being submitted for peer review. This chapter is based on final results from GBD 2021 as of November 2023. These papers advance our understanding of the consequences and stakes of childhood growth failure by quantifying infectious disease risk based on metrics of CGF and by quantifying the overall burden of disease and the magnitude of health loss associated with CGF on a global level. The fourth and final chapter of this dissertation, Longitudinal trajectories of early childhood vulnerability: a pooled analysis of cohort studies of growth and mortality, uses a dataset of 58 studies to quantify vulnerability to all-cause mortality. Using vulnerability as a framework to describe the complex, interdimensional factors contributing to population- or individual-level risk of poor health and developmental outcomes. Recently, vulnerability has been used to segment newborns into birth phenotypes, or categorical groups based on their gestational age and birthweight.12 These phenotypes are strongly associated with the probability of neonatal mortality. A separate analysis used a standardized, pooled dataset of prospective cohort studies from low-and middle-income countries to produce robust, longitudinal estimates of the incidence and recovery of stunting and wasting.11,22 Our analysis uses this same pooled dataset and adds 18 other studies, standardized across growth metrics, maternal characteristics, birth outcomes, and sociodemographic indicators to quantify vulnerability in the first two years of life. We defined vulnerability as the risk of all-cause mortality and used a survival analysis to jointly estimate the impact of continuous values of HAZ and WHZ on that risk. This non-linear, compounded risk can be calculated for any child given their age, sex, weight, and height. We then used a multidimensional model to estimate latent class trajectories in HAZ and WHZ to quantify how children grow in these metrics simultaneously. Membership in each trajectory cluster describes a child’s magnitude and trend in vulnerability over time as they age from birth to 2 years. The analysis showed that children born to poorly nourished mothers and mothers with lower education were more likely to have lower starting and deteriorating vulnerability than their peers. This novel analysis used a very large, pooled dataset of tens of thousands of children and hundreds of thousands of anthropometric observations to produce robust quantitative estimates of vulnerability, transforming a useful framework into a practical and actionable way to identify children that are most likely to suffer from mortality and to intervene upon them to prevent it. This paper will be finalized and submitted to peer review and publication in an academic journal. Altogether, this dissertation has advanced our understanding of how childhood growth failure affects the risk of disease burden among children younger than 5. It surveyed the landscape of statistical methods to assess repeated measurements of growth and applied lessons from that review to building appropriate and robust statistical models to quantifying outcomes of CGF. It has quantified risks of cause-specific infectious disease incidence and mortality and generated estimates of the burden of disease caused by CGF at the global level. Finally, repeated growth measurements were used to generate trajectories in childhood vulnerability and describe which children are most likely to have positive and negative slopes in their vulnerability as they age. Reducing early childhood vulnerability associated with childhood growth failure ultimately requires multi-sectoral policies, strategies, and interventions.23 This dissertation has reiterated how urgent those efforts are and created a framework for identifying the children most vulnerable. Thanks for reading!