Re-evaluating the impact of reduced malaria prevalence on birthweight in sub-Saharan Africa: A pair-of-pairs study via two-stage bipartite and non-bipartite matching

Pengyun Wang, Ping Huang, Yifan Jin, Yanxin Shen, Omar El Shahawy, Dae-Woong Ham, Wendy Prudhomme O’Meara, Siyu HengarXiv 2024 · DOI · arXiv

Abstract

In sub-Saharan Africa, malaria remains a significant threat to maternal and neonatal health, with approximately 32% of pregnant women infected in 2021. While individual-level interventions are known to improve birth outcomes, the community-level impact of declining malaria prevalence is more complex due to potential shifts in population-wide immunity. This study introduces a novel "pair-of-pairs" statistical framework to evaluate this relationship using data from 203,141 children across 18 countries (2000–2018). By utilizing two-stage bipartite and non-bipartite matching to maintain malaria prevalence as a continuous variable, we find that significant reductions in malaria are associated with an average birthweight increase of 98.899 grams. This methodology offers a robust alternative to traditional difference-in-differences designs that rely on data-thinning dichotomization.

Introduction: The Malaria-Birthweight Paradox

Malaria infection during pregnancy is a well-documented cause of adverse birth outcomes, most notably low birthweight. However, the relationship between community-level malaria reduction and public health outcomes is not straightforward. As malaria prevalence drops in a region, the population’s natural antimalarial immunity also tends to decline. This creates a potential trade-off: while fewer expectant mothers are exposed to the parasite, those who are infected may suffer more severe clinical consequences due to their reduced immune response.

Previous research (e.g., Heng et al., 2021) struggled to find a clear signal of community-level improvement. This study hypothesizes that the lack of evidence was not due to a lack of effect, but rather a limitation in statistical methodology—specifically the practice of "dichotomizing" continuous malaria data into binary "high" and "low" categories. Such approaches often discard significant portions of available data and obscure the nuanced dose-response relationship between disease prevalence and health.

Methodology: The Pair-of-Pairs Design

To address these limitations, we propose a "pair-of-pairs" design using a two-stage matching process. This framework allows for a difference-in-differences (DiD) analysis even when dealing with cross-sectional survey data where the same individuals are not tracked over time.

Stage One: Bipartite Geographic Matching

Because the Demographic and Health Surveys (DHS) visit different households in different years, we first paired survey clusters from "early years" (2000–2010) with clusters from "late years" (2011–2018) based on geographic proximity. By matching clusters within a 5km radius, we effectively created "pseudo-longitudinal" units, treating nearby locations as the same community measured at two points in time.

Stage Two: Non-Bipartite Quadruple Matching

In the second stage, we used optimal non-bipartite matching to pair these geographic pairs into "quadruples." Unlike traditional DiD, which splits groups into "treated" and "control," this method matches pairs that share similar socioeconomic and demographic trajectories but exhibit different magnitudes of change in malaria prevalence. This allows the Plasmodium falciparum parasite rate ($PfPR_{2-10}$) to remain a continuous variable, preserving the full richness of the dataset.

Data Integration and Imputation

The study integrated high-resolution (5km x 5km) malaria prevalence rasters from the Malaria Atlas Project (MAP) with individual-level health data from the DHS. To address the common issue of missing birthweight records in historical African surveys, we employed Bayesian linear regression to impute values based on maternal reports of infant size and other relevant covariates.

Key Insights and Results

The Power of Continuous Data

By avoiding dichotomization, our methodology increased the effective sample size by more than tenfold compared to previous studies—from approximately 18,000 to over 200,000 records. This increased power revealed a clear, statistically significant relationship between malaria reduction and birthweight.

Quantitative Impact

Our analysis estimates that every 1% reduction in the malaria parasite rate ($PfPR_{2-10}$) increases the community-level average birthweight by approximately 1.557 grams ($p < 0.001$). In regions experiencing the most dramatic success in malaria control—a 63.5% reduction in prevalence—the estimated increase in average birthweight reached 98.899 grams (95% CI: [39.002, 158.796]).

Clinical and Robustness Implications

A nearly 100-gram increase in birthweight is clinically significant, as it is associated with reduced risks of neonatal mortality and improved long-term health outcomes, including better bone density and muscle strength. Sensitivity analysis further confirmed the robustness of these findings; an unobserved confounding factor would need to explain nearly 10% of the variance in both malaria and birthweight to invalidate the results.

Conclusion

This study provides strong evidence that the benefits of reduced malaria exposure far outweigh the risks associated with declining maternal immunity. Beyond the immediate implications for malaria policy, the "pair-of-pairs" framework via non-bipartite matching offers a powerful new tool for epidemiologists. This approach is particularly valuable in settings where researchers must reconcile heterogeneous survey locations with continuous exposure measures, providing a more precise lens through which to view the impact of large-scale public health interventions.

Summary updated

Abstract

According to the WHO, in 2021, about 32% of pregnant women in sub-Saharan Africa were infected with malaria during pregnancy. Malaria infection during pregnancy can cause various adverse birth outcomes such as low birthweight. Over the past two decades, while some sub-Saharan African areas have experienced a large reduction in malaria prevalence due to improved malaria control and treatments, others have observed little change. Individual-level interventional studies have shown that preventing malaria infection during pregnancy can improve birth outcomes such as birthweight; however, it is still unclear whether natural reductions in malaria prevalence may help improve community-level birth outcomes. We conduct an observational study using 203,141 children's records in 18 sub-Saharan African countries from 2000 to 2018. Using heterogeneity of changes in malaria prevalence, we propose and apply a novel pair-of-pairs design via two-stage bipartite and non-bipartite matching to conduct a difference-in-differences study with a continuous measure of malaria prevalence, namely the Plasmodium falciparum parasite rate among children aged 2 to 10 ($\text{PfPR}_{2-10}$). The proposed novel statistical methodology allows us to apply difference-in-differences without dichotomizing $\text{PfPR}_{2-10}$, which can substantially increase the effective sample size, improve covariate balance, and facilitate the dose-response relationship during analysis. Our outcome analysis finds that among the pairs of clusters we study, the largest reduction in $\text{PfPR}_{2-10}$ over early and late years is estimated to increase the average birthweight by 98.899 grams (95% CI: $[39.002, 158.796]$), which is associated with reduced risks of several adverse birth or life-course outcomes. The proposed novel statistical methodology can be replicated in many other disease areas.
BibTeX
@misc{wang2024re,
  author = {Pengyun Wang and Ping Huang and Yifan Jin and Yanxin Shen and Omar El Shahawy and Dae-Woong Ham and Wendy Prudhomme O’Meara and Siyu Heng},
  title = {{Re-evaluating the impact of reduced malaria prevalence on birthweight in sub-Saharan Africa: A pair-of-pairs study via two-stage bipartite and non-bipartite matching}},
  year = {2024},
  eprint = {2409.19314},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2409.19314},
}
Figure