Testing the harvesting hypothesis by time-domain regression analysis. II: covariate effects

Area of Science:

• Environmental Epidemiology
• Biostatistics
• Time-Series Analysis

Background:

• Investigating the time-scale log-linear regression model's efficacy for detecting mortality displacement (harvesting) in air pollution time-series data.
• Extending prior research by Fung et al. (2004) on the Dominici et al. (2003) model.
• Exploring mortality displacement under pure and mixed frailty models.

Purpose of the Study:

• To evaluate the capability of time-scale log-linear regression to identify mortality displacement.
• To assess the impact of covariates, such as temperature, on the detection of mortality displacement.
• To determine the reliability of the model when faced with potential confounding factors.

Main Methods:

• Conducted a simulation study using pure and mixed frailty compartment models.
• Analyzed time-series data relating air pollution to excess mortality.
• Examined time-scale coefficients of log relative risk under different model specifications.

Main Results:

• A characteristic mortality displacement pattern was identified in the pure frailty model with a moderate pollution effect.
• This pattern disappeared upon the introduction of temperature as a covariate.
• An incorrectly specified model (omitting temperature) produced a false mortality displacement effect.

Conclusions:

• Time-scale regression demonstrates limited value for detecting mortality displacement in air pollution time-series data.
• The presence of covariates like temperature significantly impacts the model's ability to accurately detect mortality displacement.
• Model misspecification can lead to erroneous conclusions regarding mortality displacement.