Assessing and attenuating the impact of selection bias on spatial cluster detection studies

Area of Science:

• Epidemiology
• Biostatistics
• Geographic Information Systems (GIS) in Public Health

Background:

• Spatial cluster analyses are vital in epidemiology for detecting disease risk hotspots using case-control data.
• Case-control studies are prone to selection bias due to subject non-participation, potentially impacting spatial analysis findings.

Purpose of the Study:

• To systematically evaluate the impact of non-participation on spatial cluster analysis in case-control studies.
• To develop and validate a spatial algorithm for correcting non-participation bias in disease risk mapping.

Main Methods:

• A simulation study was conducted using the local spatial scan statistic.
• Scenarios varied non-participation rates, locations, and disease risk intensity.
• A novel spatial algorithm was proposed to adjust for spatially structured non-participation.

Main Results:

• Lower control participation than case participation significantly inflated false-positive rates for artificial clusters.
• Non-participation outside true risk zones reduced the power to detect actual disease hotspots.
• The proposed algorithm effectively reduced false positives and maintained sensitivity in detecting true risk zones.

Conclusions:

• Non-participation bias poses a significant threat to the validity of spatial cluster analyses in epidemiological research.
• The developed spatial algorithm offers a promising method to mitigate non-participation bias.
• Increased attention to non-participation effects is crucial for accurate spatial epidemiology studies.