Unmatched spatially stratified controls: A simulation study examining efficiency and precision using...

Area of Science:

• Epidemiology
• Geographic Information Systems (GIS)
• Biostatistics

Background:

• Spatial analysis in epidemiology often relies on control selection methods.
• Traditional methods like simple random sampling (SRS) may not adequately represent geographic distributions.
• Efficient control selection is crucial for accurate spatial epidemiological studies.

Purpose of the Study:

• To evaluate the performance of spatially stratified random sampling (SSRS) for selecting controls in spatial analyses.
• To compare SSRS with simple random sampling (SRS) using a case study of preterm birth in Massachusetts.
• To assess the impact of SSRS on statistical model accuracy and efficiency.

Main Methods:

• Developed and applied SSRS by dividing the study area into spatial strata and selecting controls from non-cases within each stratum.
• Conducted a simulation study fitting generalized additive models using controls selected by SSRS and SRS.
• Compared mean squared error (MSE), relative efficiency (RE), bias, and map results against models using all non-cases.

Main Results:

• SSRS designs demonstrated lower average mean squared error (0.0042-0.0044) compared to SRS designs (0.0072-0.0073).
• SSRS achieved higher relative efficiency (77-80%) than SRS (71%).
• SSRS yielded more consistent statistically significant map results across simulations, particularly in low population density areas.

Conclusions:

• SSRS is a more efficient and reliable method for selecting controls in spatial analyses compared to SRS.
• The geographically balanced nature of SSRS improves model performance and spatial pattern detection.
• SSRS is particularly advantageous for spatial epidemiological studies in diverse geographic settings.