Estimation of exposure-attributable fractions from time series: A simulation study

Area of Science:

• Epidemiology
• Biostatistics
• Public Health

Background:

• Public health burden analysis frequently estimates exposure-attributable fractions from time series data.
• Accurate estimation is challenging when the entire population is exposed, necessitating robust exposure-outcome association modeling.

Purpose of the Study:

• To derive asymptotic convergences for attributable fraction estimation in common time series models.
• To evaluate the performance of different models under various exposure-outcome association scenarios.

Main Methods:

• Utilized the delta method for asymptotic convergence derivations for ARMAX, Poisson, negative binomial, and Serfling models.
• Employed a Monte Carlo algorithm for Poisson regression, accounting for estimation and prediction errors.
• Conducted a simulation study with epidemic exposure and compared additive and multiplicative models.

Main Results:

• Additive models performed well for additive data, while multiplicative models were poor, and vice versa.
• The Serfling model demonstrated poor performance across all tested scenarios.
• Misspecification of the exposure-outcome association's form or delay resulted in mediocre attributable fraction estimates.

Conclusions:

• The Serfling model is not recommended for estimating attributable fractions.
• Selecting the appropriate time series model requires careful investigation of the exposure-outcome association.
• Accurate burden estimation depends on appropriate model selection and correct specification of exposure-outcome relationships.