A flexible Bayesian framework to estimate age- and cause-specific child mortality over time from sample registration data

Austin E Schumacher, Tyler H. McCormick, Jon Wakefield, Yue Chu, Jamie Perin, Francisco Villavicencio, Noah Simon, Li Liu

Abstract

In order to implement disease-specific interventions in young age groups, policy makers in low- and middle-income countries require timely and accurate estimates of age- and cause-specific child mortality. High quality data is not available in settings where these interventions are most needed, but there is a push to create sample registration systems that collect detailed mortality information. Current methods that estimate mortality from this data employ multistage frameworks without rigorous statistical justification that separately estimate all-cause and cause-specific mortality and are not sufficiently adaptable to capture important features of the data. We propose a flexible Bayesian modeling framework to estimate age- and cause-specific child mortality from sample registration data. We provide a theoretical justification for the framework, explore its properties via simulation, and use it to estimate mortality trends using data from the Maternal and Child Health Surveillance System in China.

Introduction

In many low- and middle-income countries (LMICs), the absence of comprehensive vital registration systems makes it difficult to track child mortality trends accurately. Instead, health officials often rely on Sample Registration Systems (SRS), which collect data from representative sites.

Traditionally, SRS data is analyzed using "multistage" methods: total deaths are estimated first, followed by the fraction of deaths attributed to specific causes (e.g., pneumonia, diarrhea). These two estimates are then multiplied together. However, this "disconnected" approach lacks a rigorous statistical foundation. It prevents uncertainty from flowing between stages, often leading to overly narrow uncertainty intervals and potentially misleading policy recommendations.

Methodology

This paper introduces a unified Bayesian framework based on a competing risks failure process. By moving away from traditional multistage "plug-in" methods, this approach provides more accurate uncertainty intervals and uncovers hidden trends in granular age groups. Key technical innovations include:

  • The "Poisson Trick": The authors demonstrate that the complex likelihood of individuals dying from different causes can be simplified into a product of Poisson distributions for tabulated death counts and person-years, making the model computationally efficient.
  • Log-linear Hazards: Mortality rates are modeled on a log scale, allowing for flexible functional forms that include fixed effects for regions and causes, alongside random effects.
  • Temporal Smoothing: To account for trends over time, the model utilizes second-order random walks (RW2), ensuring that mortality estimates follow a smooth, realistic trajectory rather than jumping erratically year-to-year.
  • Efficient Computation: Rather than using slow MCMC simulations, the framework utilizes Integrated Nested Laplace Approximation (INLA) for faster estimation of complex Bayesian models.

Application: The China Case Study

The framework was applied to data from the Maternal and Child Health Surveillance System (MCHSS) in China (1996–2015), covering 20 years of data across six region-residency strata (e.g., "east rural," "west urban") and six granular age groups.

Trends in cause-specific mortality fractions across age groups and regions in China (2000–2015).

As shown in the figure above, the model successfully smoothed noisy empirical data to reveal distinct regional and age-specific patterns. For instance, in "west rural" regions, the model captured a high burden of prematurity-related deaths while clearly illustrating the decline in communicable diseases over time.

Key Findings

  • Likelihood Misspecification: The paper proves that multistage models suffer from "likelihood misspecification." By treating the log of the sum of mortality rates as normally distributed, they consistently produce uncertainty intervals that are too narrow. In simulations, the unified model achieved near-nominal (95%) coverage, whereas the multistage approach dropped as low as 80%.
  • Granularity and Hidden Risks: Breaking down the neonatal period (0–1 month) into "0–6 days" and "7–27 days" revealed that certain causes, like acute respiratory infections, remain significant threats in the later neonatal stage even when they appear to be declining in aggregate data.
  • Borrowing Strength: The Bayesian approach allows the model to "borrow strength" across strata. If data is sparse for a specific cause in one region, the model uses global trends and neighboring data to produce stable estimates.

Conclusion and Future Work

While the framework provides a significant advancement in mortality estimation, some limitations remain. The authors found that the China dataset was too small to reliably estimate correlations between different causes of death (e.g., measles outbreaks increasing pneumonia deaths). Additionally, the model relies on "person-years at risk," which must often be estimated using secondary demographic methods.

Future work may involve extending the framework to include environmental covariates and adapting it to handle "Verbal Autopsy" data, where causes of death are assigned probabilistically by algorithms rather than physicians.

Summary updated

Abstract

In order to implement disease-specific interventions in young age groups, policy makers in low- and middle-income countries require timely and accurate estimates of age- and cause-specific child mortality. High quality data is not available in settings where these interventions are most needed, but there is a push to create sample registration systems that collect detailed mortality information. Current methods that estimate mortality from this data employ multistage frameworks without rigorous statistical justification that separately estimate all-cause and cause-specific mortality and are not sufficiently adaptable to capture important features of the data. We propose a flexible Bayesian modeling framework to estimate age- and cause-specific child mortality from sample registration data. We provide a theoretical justification for the framework, explore its properties via simulation, and use it to estimate mortality trends using data from the Maternal and Child Health Surveillance System in China.
BibTeX
@article{schumacher2022flexible,
  author = {Austin E Schumacher and Tyler H. McCormick and Jon Wakefield and Yue Chu and Jamie Perin and Francisco Villavicencio and Noah Simon and Li Liu},
  title = {{A flexible Bayesian framework to estimate age- and cause-specific child mortality over time from sample registration data}},
  journal = {The Annals of Applied Statistics},
  year = {2022},
  doi = {10.1214/21-aoas1489},
}
Figure