Improving Inference in Air Pollution Epidemiology: The Case for Rethinking Multipollutant Adjustment

Area of Science:

• Environmental Epidemiology
• Biostatistics
• Public Health Policy

Background:

• Air pollution poses significant public health risks, necessitating accurate identification of harmful pollutants.
• Epidemiological studies commonly use statistical adjustments for multiple pollutants to assess individual impacts and interactions.
• Multi-pollutant analyses are increasingly prevalent in air quality research and policy development.

Purpose of the Study:

• To identify potential biases introduced by indiscriminate co-pollutant adjustment in epidemiological studies.
• To investigate the mechanisms and real-world impact of collider bias in multi-pollutant air pollution research.
• To propose strategies for mitigating bias in multi-pollutant air quality and health outcome analyses.

Main Methods:

• Utilized a well-characterized Canadian national cohort to provide empirical evidence.
• Conducted a simulation study to explore the theoretical underpinnings of co-pollutant adjustment bias.
• Applied regression models with statistical adjustments for co-pollutants.

Main Results:

• Indiscriminate co-pollutant adjustment can induce noncausal associations through collider adjustment.
• This bias can distort effect estimates for individual air pollutants and their relationship with health outcomes.
• Empirical and simulation data demonstrate the realistic influence of this bias on air pollution-health research.

Conclusions:

• Multi-pollutant analyses require careful consideration to avoid introducing noncausal associations.
• Collider bias is a significant concern that can distort findings in air quality epidemiology.
• Researchers and policymakers should exercise greater caution when conducting and interpreting multi-pollutant air pollution studies.