Related Experiment Videos
Statistical methods for geographical surveillance in veterinary epidemiology.
A Biggeri1, D Catelan, L Rinaldi
1Dept. of Statistics "G. Parenti", University of Florence, Florence, Italy.
Parassitologia
|August 3, 2006
Summary
This study reviews spatial clustering methods for disease risk analysis. Bayesian models and non-parametric tools effectively identified disease clusters, with spatial ranges around 600 meters.
Area of Science:
- Spatial statistics
- Epidemiology
- Geographic Information Systems (GIS)
Background:
- Spatial clustering and cluster detection are crucial for understanding disease risk and identifying anomalous case aggregation.
- Numerous hypotheses exist regarding the mechanisms driving disease aggregation, necessitating robust analytical methods.
Purpose of the Study:
- To review and compare statistical methods for marked point data (case/control) in disease risk analysis.
- To assess the ability of various methods to describe spatial disease intensity, test for randomness, and locate significant excesses.
- To provide an informal guideline for spatial analysis using real-world data.
Main Methods:
- Review of spatial clustering techniques including Kernel density estimation, Ripley's K function, Cuzick-Edwards test, SatScan, and Bayesian Gaussian Spatial Exponential models.
- Application of these methods to analyze data on fecal contamination and dog parasitic diseases in Naples, Italy.
- Comparison of non-parametric and Bayesian approaches for spatial analysis.
Main Results:
- Kernel density estimation showed high sensitivity to bandwidth, overemphasizing localized excesses.
- Ripley's K function and the Cuzick-Edwards test demonstrated consistency in detecting clusters.
- SatScan failed to detect significant excesses in the analyzed dataset.
- The spatial range of clusters was approximately 600 meters, indicating the presence of several small clusters.
- Bayesian models proved powerful in reconstructing the spatial phenomenon and provided parameter inferences consistent with non-parametric analyses.
Conclusions:
- Bayesian Gaussian Spatial Exponential models and non-parametric methods like Ripley's K and Cuzick-Edwards test are effective for spatial disease risk analysis.
- The choice of method and parameters (e.g., bandwidth) significantly impacts cluster detection results.
- Spatial analysis revealed disease clusters with a range of approximately 600 meters, relevant for public health interventions.