The secondary analysis examined spatial variation in survival benefit from mass azithromycin distribution across Dosso, Niger. Investigators pooled data from 1,390 villages collected during the AVENIR trial (children aged 1–59 months; 2020–2023). The analytic sample comprised 276,452 person-years and 3,635 deaths. Data indicated an overall mortality rate of 13.1 per 1,000 person-years and an average treatment effect (ATE) of −2.6 deaths per 1,000 person-years. Mortality was not uniform across the study area. The highest rates clustered in the southern sector, while the largest treatment benefit concentrated in the southeast, producing a northwest–southeast gradient in observed effect sizes.
Local maps generated by the study team revealed pockets where the absolute reduction in mortality exceeded the trial-wide ATE. These localized signals suggest that placespecific factors modulated the survival gain associated with biannual antibiotic distribution. The authors reported those spatial patterns without extending beyond the data presented in the source document.
Investigators applied Poisson regression frameworks to estimate spatially continuous surfaces for both child mortality and the ATE. The source states that Poisson models were used to produce smoothed, location-level estimates across the full geographic domain of the trial. This approach allowed prediction of a treatment effect at a fine spatial scale rather than restricting inference to randomized clusters alone. The analysis therefore moved from cluster-level averages to a continuous spatial field of estimated outcomes and effects, enabling identification of subcluster heterogeneity.
The paper frames this as a broadly generalizable analytic problem for large-scale, cluster-randomized trials: how to estimate where an intervention yields larger or smaller absolute benefits when geographic variation in baseline risk exists. The study team presented model-derived maps and used them to compare treatment effect magnitude across locations.
To test a pragmatic surveillance strategy, the authors aggregated village-level data to 160 primary health catchment areas. Aggregation preserved broad spatial patterns of mortality—regions of higher and lower mortality remained identifiable at the catchment level. However, the process attenuated local heterogeneity and smoothed out smaller-scale variation in treatment effect.
This result suggests that routine monitoring at the health catchment level may suffice for detecting large-scale mortality trends. At the same time, aggregation reduces sensitivity to microgeographic hotspots where the intervention’s benefit differed markedly from surrounding areas. The authors reported these trade-offs based on comparisons between village-level and catchment-level spatial estimates.
The investigators investigated whether underlying mortality rates could explain the spatial heterogeneity in treatment effects. They examined unit-level, spatially predicted treatment effects as a function of local mortality and found that baseline mortality captured much of the observed heterogeneity. The relationship between higher underlying mortality and larger absolute benefits from azithromycin was apparent in their models.
Still, the analysis also found that mortality did not account for all heterogeneity. The authors noted residual variation in predicted effect sizes after adjusting for baseline mortality patterns, indicating that other local drivers—unmeasured in this analysis—likely contributed to spatial differences in treatment response. The source did not report, and therefore this rewrite does not invent, what those additional drivers might be.
The underlying randomized trial, AVENIR, was cluster-randomized and placebo-controlled, assessing biannual azithromycin distribution to children aged 1–59 months. The trial period reported in the source spanned 2020–2023 and took place in Dosso, Niger. Trial registration is provided (NCT04224987). The authors confirmed ethical approval from the Comité National Éthique pour la Recherche en Santé in Niger and from the University of California, San Francisco Institutional Review Board. Informed verbal consent procedures for community leaders, heads of household, and caregivers were described in the source.
De-identified AVENIR trial data are available through the Open Science Framework at the link provided by the authors. The village geographic coordinates used for the fine-scale spatial analysis are not publicly shareable to protect participant confidentiality; that restriction is stated explicitly in the source material.
This work is a secondary analysis of the AVENIR trial and therefore inherits the trial’s design constraints; the source frames the spatial analysis as an added analytic layer rather than a new randomized comparison. The village coordinates required for fine-scale mapping are not shareable, limiting external reproducibility of location-specific maps. The authors also showed that baseline mortality explained much—but not all—of the variation in treatment effects, leaving open specific, data-driven explanations for the residual heterogeneity. The source did not report further causal attribution for those residual signals, and it did not provide additional analyses that identify the unmeasured local drivers. These items remain open questions for subsequent work using the trial dataset or complementary data sources.
The analysis demonstrates a replicable modeling approach to estimate localized treatment effects in large cluster-randomized trials and documents geographic variation in both child mortality and the absolute survival benefit of mass azithromycin distribution in Dosso, Niger. The source presents methods, geographic patterns, and data-sharing statements but stops short of offering causal explanations for all spatial heterogeneity; those are explicit gaps left by the authors for future investigation.