Non-parametric estimation of spatial variation in relative risk

Area of Science:

• Epidemiology
• Biostatistics
• Geographical Information Systems (GIS)

Background:

• Estimating spatial variations in disease risk is crucial for public health.
• Existing methods may lack precision in identifying localized risk fluctuations.
• Understanding relative disease risks across regions informs targeted interventions.

Purpose of the Study:

• To develop a robust statistical methodology for estimating and visualizing spatial variations in relative disease risks.
• To introduce methods for assessing the statistical significance of estimated risk patterns.
• To provide a framework for analyzing disease risk surfaces in epidemiological studies.

Main Methods:

• Utilizing a Poisson point process model for disease occurrence.
• Applying non-parametric kernel smoothing for density ratio estimation.
• Implementing pointwise tolerance contours for risk surface visualization.
• Proposing a Monte Carlo test for the null hypothesis of uniform risk.

Main Results:

• The methodology effectively estimates spatial variations in relative disease risks.
• Pointwise tolerance contours enhance the interpretability of risk surfaces.
• The Monte Carlo test provides a statistically sound approach to assess risk uniformity.
• Demonstrated capabilities through two real-world epidemiological examples.

Conclusions:

• The proposed non-parametric kernel smoothing method offers a powerful tool for analyzing spatial disease risk.
• The inclusion of tolerance contours and Monte Carlo testing improves the reliability of risk surface interpretation.
• This approach enhances epidemiological analysis by providing statistically validated insights into geographical disease patterns.