Global, Regional and National Estimation of Firstborn Prevalence, 1990-2022

Thesis (Master's)--University of Washington, 2020Introduction: Birth order, and in particular whether or not a child is firstborn, is known to be directly correlated with risk of specific neonatal diseases and indirectly associated with health outcomes beyond the neonatal period. Accurate and detailed knowledge of firstborn prevalence is therefore important in its own right and also a powerful tool for improving our understanding of multiple child health conditions. Data on birth order is not available in every location or year, so modeling is required. Previous models have used all age total fertility rate as a covariate which, although simpler, obscures important cohort fertility trends and subsequently may bias results. This analysis develops a new modeling strategy for estimating firstborn prevalence that explicitly incorporates the interaction between maternal age and historical fertility trends of maternal cohorts, exploring the model and prediction performance comparisons between the two approaches. Methods: Data on birth order from population-representative surveys and vital registration statistics were extracted and tabulated to inform an annual model of firstborn prevalence. Total fertility rate estimates from the Institute for Health Metrics and Evaluation were recomputed to derive a measure of fertility specific to maternal age and cohort, and the new age-specific measure was used as a covariate in a three step spatiotemporal Gaussian process regression to estimate maternal-age-specific firstborn prevalence across all GBD 2020 locations annually from 1990 to 2022. To first directly compare the effect of modeling by maternal age compared to across all ages, one dataset restricted to sources with complete coverage of the reproductive age range was modeled using both approaches and assessed through in-sample error metrics and qualitative comparison. To generate more comprehensive final estimates, a third model was then run using the age-specific model settings and full dataset (no restrictions on complete age coverage). Results: When run on the same data, the age-specific model performed better than the all age model in terms of in-sample fit. Both models had a small negative bias (-3.55% in the all-age model and -2.85% in the age-specific model). When the age-specific model was run on the full dataset, firstborn prevalence globally in 2019 was estimated to be 38.9% (36.9 – 40.9), an increase of 8.72% (1.50 – 16.8) since 1990. Firstborn prevalence in 2019 was highest in the Republic of Korea and Taiwan and lowest in Guinea-Bissau and the Solomon Islands. In 1990 firstborn prevalence was highest in San Marino and Greenland and lowest in Tuvalu and Qatar. Globally, firstborn prevalence was highest among women 10-14 years old and lowest among women 45-49 years old. Discussion: Modeling firstborn prevalence by maternal age improves upon modeling across all ages by improving in-sample fit, removing assumptions about comparability of maternal age ranges, and producing age-specific estimates that are more informative and more actionable. The set of complete and cross-nationally comparable estimates of firstborn prevalence can be used as predictive covariates to model other health outcomes, and the age-specific modeling framework reveals more specific drivers of changes in firstborn prevalence and presents opportunities for modeling additional fertility scenarios