Related Experiment Videos
Bayesian estimates of disease maps: how important are priors?
L Bernardinelli1, D Clayton, C Montomoli
1Istituto Scienze Sanitarie Applicate, Universitá degli Studi di Pavia, Italy.
Statistics in Medicine
|November 15, 1995
Summary
The fully Bayesian approach to disease mapping is sensitive to hyperprior choices. This study compared Bayesian methods to maximum likelihood for mapping insulin-dependent diabetes mellitus (IDDM) risk.
Area of Science:
- Biostatistics
- Epidemiology
- Spatial Analysis
Background:
- Disease mapping is crucial for public health surveillance.
- The fully Bayesian (FB) approach offers a robust framework for disease mapping.
- Choosing the hyperprior distribution for the dispersion parameter is a critical aspect of the FB approach.
Purpose of the Study:
- To investigate the sensitivity of rate ratio estimates in the FB approach to hyperprior choices.
- To compare the performance of the FB approach with the conventional maximum likelihood (ML) approach for disease risk mapping.
Main Methods:
- A simulation study was conducted to assess sensitivity.
- The study utilized incidence data of insulin-dependent diabetes mellitus (IDDM) from Sardinia.
- Bayesian and maximum likelihood methods were compared for mapping disease risk.
Main Results:
- Rate ratio estimates in the FB approach demonstrated sensitivity to the selection of the hyperprior distribution.
- The FB approach showed comparable or superior performance to the ML approach in certain scenarios.
- Differences in performance were observed between the two methods.
Conclusions:
- The choice of hyperprior distribution significantly impacts rate ratio estimates in Bayesian disease mapping.
- The FB approach provides a valuable alternative to conventional ML methods for disease risk assessment.
- Further research is warranted to optimize hyperprior selection in Bayesian disease mapping studies.