Confounding in air pollution epidemiology: when does two-stage regression identify the problem?

Area of Science:

• Environmental epidemiology
• Biostatistics
• Public health

Background:

• Time-series studies are crucial for understanding the health impacts of airborne particulate matter (PM).
• Assessing confounding by copollutants is a significant challenge in environmental epidemiology.
• Existing methods for confounding assessment in PM studies require careful evaluation.

Purpose of the Study:

• To critically evaluate a recently proposed two-stage approach for assessing confounding in time-series air pollution studies.
• To investigate the potential for misleading inferences from this method, particularly concerning particulate matter (PM) and health effects.
• To highlight the importance of considering additional copollutants in epidemiological analyses.

Main Methods:

• The study analyzes a two-stage regression approach applied to independent city-specific time-series data.
• First stage: Fitting separate models for health effects vs. PM and confounder vs. PM within each city.
• Second stage: Regressing city-specific slopes from the first stage to infer direct effects and confounding.

Main Results:

• The proposed two-stage method's interpretation of nonzero intercepts (direct PM effect) and slopes (confounding) may be flawed.
• A counterexample demonstrates that the presence of an additional copollutant can lead to incorrect conclusions.
• Inferences regarding the direct pathway from PM to health effects could be overestimated or underestimated.

Conclusions:

• The evaluated two-stage method for confounding assessment in particulate matter (PM) epidemiology may yield unreliable results.
• The method's assumptions can be violated by unmeasured or additional copollutants, leading to biased interpretations.
• Researchers should exercise caution when applying this method and consider alternative or supplementary analytical strategies.