Interrupted Time Series Analysis in Environmental Epidemiology: A Review of Traditional and Novel Modeling Approaches

Interrupted time series (ITS) designs are increasingly used in environmental health to evaluate impacts of extreme weather events or policies. This paper aims to introduce traditional and contemporary ITS approaches, including machine learning algorithms and Bayesian frameworks, which enhance flexibility in modeling complex temporal patterns (e.g., seasonality and nonlinear trends) and spatially heterogeneous treatment effects. We present a comparative analysis of methods such as ARIMA, machine learning models, and Bayesian ITS, using a real-world case study: estimating excess respiratory hospitalizations during the 2018 wildfire smoke event in San Francisco.

Our study demonstrates the practical application of these methods and provides a guide for selecting and implementing ITS designs in environmental epidemiology. To ensure reproducibility, we share annotated datasets and R scripts, allowing researchers to replicate analyses and adapt workflows. While focused on environmental applications, particularly acute exposures like wildfire smoke, the framework is broadly applicable to public health interventions. This work advances ITS methodology by integrating contemporary statistical innovations and emphasizing actionable guidance for causal inference in complex, real-world settings.