Time-stratified case-crossover studies for aggregated data in environmental epidemiology: a tutorial

Aurelio Tobias,Yoonhee Kim,Lina Madaniyazi

DOI: https://doi.org/10.1093/ije/dyae020

IF: 9.685

2024-02-14

International Journal of Epidemiology

Abstract:Abstract The case-crossover design is widely used in environmental epidemiology as an effective alternative to the conventional time-series regression design to estimate short-term associations of environmental exposures with a range of acute events. This tutorial illustrates the implementation of the time-stratified case-crossover design to study aggregated health outcomes and environmental exposures, such as particulate matter air pollution, focusing on adjusting covariates and investigating effect modification using conditional Poisson regression. Time-varying confounders can be adjusted directly in the conditional regression model accounting for the adequate lagged exposure–response function. Time-invariant covariates at the subpopulation level require reshaping the typical time-series data set into a long format and conditioning out the covariate in the expanded stratum set. When environmental exposure data are available at geographical units, the stratum set should combine time and spatial dimensions. Moreover, it is possible to examine effect modification using interaction models. The time-stratified case-crossover design offers a flexible framework to properly account for a wide range of covariates in environmental epidemiology studies.

public, environmental & occupational health

What problem does this paper attempt to address?

The paper aims to address the issue of how to use the time-stratified case-crossover design in environmental epidemiology to study the short-term association between environmental exposure and health events. Specifically, the paper addresses this issue through the following aspects:

1. **Implementation of Time-Stratified Case-Crossover Design**:
- The paper details how to apply the time-stratified case-crossover design to aggregate data to study the relationship between environmental exposure (such as particulate air pollution) and health outcomes.
- This is achieved using a conditional Poisson regression model by adjusting covariates and investigating effect modification.

2. **Controlling for Long-Term Trends and Seasonality**:
- The time-stratified case-crossover design can control for long-term trends and seasonal effects of unmeasured time-varying confounders, as well as the day of the week effect, by controlling for the stratified set of time strata in the conditional regression model.

3. **Adjusting for Time-Varying Covariates**:
- Environmental time-varying covariates (such as temperature and relative humidity) can be adjusted in the conditional regression model using appropriate lagged exposure-response functions.

4. **Handling Subgroup Time-Invariant Covariates**:
- For subgroup time-invariant covariates (such as age groups), the typical time series dataset needs to be reshaped into a long format and conditioned on the extended stratified set.

5. **Investigating Effect Modification**:
- Effect modification can be investigated by fitting interaction terms between environmental exposure and potential effect-modifying covariates in the conditional regression model.

6. **Multi-Site Studies**:
- When environmental exposure data are available at the geographic unit level (such as cities), the stratified set should combine time and spatial dimensions, matching geographic units by day of the week within the month and year. This is known as the spatiotemporal case-crossover design.

Through these methods, the paper provides a flexible and effective framework for environmental epidemiology research, allowing for the appropriate consideration of various covariates to more accurately estimate the short-term association between environmental exposure and health events.