Zero-inflated spatio-temporal models for disease mapping.

Area of Science:

• Biostatistics
• Epidemiology
• Spatial Statistics

Background:

• Spatio-temporal disease incidence data often exhibit excess zeros, complicating standard statistical modeling.
• Traditional frequentist analysis of complex spatio-temporal models with random effects is computationally intensive.
• Bayesian methods using Markov chain Monte Carlo (MCMC) offer computational convenience but may not align with frequentist objectives.

Purpose of the Study:

• To analyze geographical and temporal variability of disease incidence in spatio-temporal count data with excess zeros.
• To propose and evaluate a frequentist approach for zero-inflated spatio-temporal modeling using data cloning.
• To facilitate computationally convenient estimation and prediction for disease incidence patterns.

Main Methods:

• Utilized zero-inflated Poisson models with random effects to account for excess zeros.
• Employed spatio-temporal models incorporating conditionally autoregressive (CAR) spatial smoothing and B-spline temporal smoothing.
• Applied the data cloning (DC) method for computationally convenient frequentist analysis, yielding maximum likelihood estimates.

Main Results:

• The data cloning approach provides a computationally feasible frequentist method for zero-inflated spatio-temporal disease modeling.
• This method facilitates straightforward prediction and calculation of standard errors or prediction intervals for disease incidence.
• The approach was successfully illustrated using real-world monthly pediatric asthma hospital visit data from Manitoba, Canada.

Conclusions:

• The proposed data cloning method offers an effective frequentist solution for analyzing spatio-temporal disease data with excess zeros.
• This approach enhances the ability to understand and predict geographical and temporal disease incidence patterns.
• The study demonstrates the utility of data cloning in complex epidemiological modeling.