A new time-varying coefficient regression approach for analyzing infectious disease data.

Area of Science:

• Epidemiology and Biostatistics
• Time Series Analysis
• Infectious Disease Modeling

Background:

• Accurate prediction of Coronavirus (SARS-COV-2) related deaths and hospitalizations is crucial during the global pandemic.
• Existing studies often use static models, failing to capture the evolving nature of the virus's impact.

Purpose of the Study:

• To investigate the lagged dependence between time series of Coronavirus (SARS-COV-2) case counts and death counts.
• To develop and evaluate time-varying coefficient models for predicting infectious disease outcomes.
• To assess the applicability of these models to other infectious diseases and dynamic lagged dependencies.

Main Methods:

• Employed time-varying coefficient models, specifically local polynomial regression and piecewise linear regression.
• Analyzed Canadian province-level and country-level cumulative case and death count data.
• Utilized out-of-sample prediction for rigorous model performance evaluation.

Main Results:

• Both local polynomial and piecewise linear time-varying models demonstrated effective performance in predicting COVID-19 trends.
• Local polynomial regression models generally outperformed piecewise linear models, particularly for complex lagged relationships.
• The proposed methods are readily implementable using existing R packages.

Conclusions:

• Time-varying coefficient models provide a robust framework for analyzing and predicting infectious disease dynamics.
• The choice of model (e.g., local polynomial regression) is critical for accurately capturing evolving relationships between disease metrics.
• This approach offers a flexible and efficient tool for epidemiological forecasting and understanding disease progression.