Geostatistical and machine learning approaches for high-resolution mapping of vaccination coverage

Area of Science:

• Geographic Information Systems (GIS) and Spatial Analysis
• Public Health and Epidemiology
• Statistical Modeling

Background:

• High-resolution vaccination coverage maps are crucial for identifying geographic inequities.
• Existing methods for map production vary, leading to uncertainty about their reliability.
• Understanding differences in map outputs is essential for effective intervention targeting.

Purpose of the Study:

• To compare the predictive performance of different mapping methodologies for vaccination coverage.
• To evaluate the implications of these methods on spatial prioritization for public health interventions.
• To guide the selection of optimal methods for health and development metric mapping.

Main Methods:

• Utilized Nigeria Demographic and Health Survey data.
• Generated 1x1 km and district-level maps using geostatistical, machine learning (ML), and hybrid methods.
• Evaluated predictive performance through cross-validation.

Main Results:

• Five of seven investigated methods showed similar predictive performance.
• Two geostatistical approaches demonstrated the best performance.
• Two ML approaches were the worst-performing methods.
• Significant differences in spatial prioritization were observed across methods.

Conclusions:

• Geostatistical methods are recommended for producing reliable vaccination coverage maps.
• Methodological choices significantly impact spatial prioritization, potentially leading to missed underserved populations.
• The findings offer guidance for mapping various health and development indicators.