Spatial-temporal generalized additive model for modeling COVID-19 mortality risk in Toronto, Canada

Area of Science:

• Epidemiology
• Biostatistics
• Geospatial Analysis

Background:

• COVID-19 pandemic presents significant public health challenges.
• Understanding spatial-temporal patterns of mortality is crucial for effective response.
• Geo-referenced data analysis requires advanced modeling techniques.

Purpose of the Study:

• To develop and evaluate a spatial-temporal generalized additive model for COVID-19 mortality.
• To investigate the influence of neighborhood-level factors on mortality patterns.
• To compare individual-level versus aggregated data modeling approaches.

Main Methods:

• Application of a spatial-temporal generalized additive model.
• Incorporation of neighborhood factors (population density, income) using 2D spline smoothers.
• Modeling temporal changes in spatial patterns with 3D tensor product smoothers.
• Comparison of in-sample and out-of-sample predictive performance.

Main Results:

• The spatial-temporal model effectively captures patterns not explained by covariates.
• Individual-level data analysis demonstrated superior model fit.
• Individual-level analysis achieved higher predictive accuracy for epidemic peaks.
• Aggregated data analysis showed limitations in capturing granular spatial-temporal dynamics.

Conclusions:

• Individual-level data analysis is more effective for modeling COVID-19 mortality.
• The proposed spatial-temporal model enhances understanding of disease dynamics.
• Accurate prediction of epidemic peaks is improved using granular data and advanced modeling.