Geographically linking population and facility surveys: methodological considerations.

Area of Science:

• Public Health
• Geographic Information Systems
• Health Services Research

Background:

• Understanding the link between health services and population health outcomes is crucial.
• Geographically linking independent datasets is an efficient method but presents methodological challenges.
• This study investigates misclassification errors from using facility samples and displaced household clusters.

Purpose of the Study:

• To quantify misclassification errors in linking health facility data with household survey data.
• To assess the impact of using facility samples versus facility censuses.
• To evaluate the effect of geographic displacement of household clusters on data linkage accuracy.

Main Methods:

• Utilized 2007 Rwanda Service Provision Assessment (RSPA) and 2007-2008 Rwanda Interim Demographic and Health Survey (RIDHS) data.
• Created five facility samples and five household cluster displacements to simulate real-world data.
• Employed four geographic techniques to link facility and cluster datasets, creating 36 paired datasets for analysis.

Main Results:

• Facility sampling introduced greater misclassification errors than cluster displacement, leading to underestimation of service access.
• Directly linking clusters to the nearest facility misclassified distance for over 50% of clusters.
• Linking to all facilities within an administrative boundary minimized misclassification, but errors persisted with sampling and displacement.

Conclusions:

• Linking facility samples with displaced household clusters is not recommended for accurate analysis at low geographic scales.
• Misclassification errors from geographic displacement can bias estimates of the relationship between service environment and health outcomes.
• Consider potential linkage errors when designing facility surveys requiring integration with population data.