Some findings on zero-inflated and hurdle poisson models for disease mapping

Area of Science:

• Biostatistics
• Spatial Epidemiology
• Geographic Information Systems

Background:

• Zero-inflation and hurdle models are common for excess zeros in disease mapping.
• Existing methods may lead to oversmoothing of risks in low-count geographical units.
• Naive models often fail to adequately address zero excess problems.

Purpose of the Study:

• To evaluate the effectiveness of different zero excess treatments in geographically referenced mortality data.
• To identify limitations of naive zero-inflation and hurdle models.
• To propose valid modeling alternatives for zero excess in disease mapping.

Main Methods:

• Analysis of geographically referenced mortality data sets.
• Comparison of naive zero-inflation and hurdle models with explicit probability modeling.
• Theoretical investigation of posterior distribution properties.
• Development and validation of novel modeling approaches.

Main Results:

• Naive zero-inflation and hurdle models are insufficient for addressing zero excess.
• Explicit modeling of zero probabilities, varying across areal units, is crucial.
• Flexible modeling strategies can lead to improper posterior distributions.
• Proposed valid models effectively handle zero excesses and correct oversmoothing.

Conclusions:

• Standard zero-inflation and hurdle models require careful implementation in disease mapping.
• Explicitly modeling zero probabilities is essential for accurate risk estimation.
• The proposed valid modeling alternatives provide improved solutions for zero excess in spatial mortality studies.