Bayesian analysis of space-time variation in disease risk

Area of Science:

• Epidemiology
• Biostatistics
• Medical Statistics

Background:

• Analyzing disease risk variation across space and time is crucial in descriptive epidemiology.
• Scarcity of data can compromise the accuracy of maximum likelihood estimates for area-specific risk and time-trends.
• Existing disease mapping models may not fully account for correlated spatial and temporal risk factors.

Purpose of the Study:

• To propose a novel Bayesian model for analyzing spatio-temporal disease risk variation.
• To address limitations of traditional methods when dealing with sparse epidemiological data.
• To extend existing disease mapping frameworks by incorporating random effects for both intercept and trend.

Main Methods:

• A Bayesian hierarchical model is developed, treating area-specific intercepts and linear time-trends as random effects.
• The model allows for correlation between area-specific intercepts and trends, providing a more nuanced analysis.
• The model is applied to analyze the cumulative prevalence of insulin-dependent diabetes mellitus in Sardinian conscripts (1936-1971).

Main Results:

• The proposed Bayesian model provides more robust estimates of disease risk and temporal trends compared to maximum likelihood methods with sparse data.
• The analysis revealed significant spatial variations in insulin-dependent diabetes mellitus prevalence within Sardinia.
• Incorporating correlated random effects improved the model's ability to capture complex epidemiological patterns.

Conclusions:

• The Bayesian random-effects model offers a powerful approach for descriptive epidemiology, particularly for disease mapping with limited data.
• The model enhances the understanding of spatio-temporal disease risk dynamics.
• Application to insulin-dependent diabetes mellitus in Sardinia demonstrates the model's utility in identifying geographically and temporally varying risk factors.