Can Incorrect Analysis of Time-Dependent Exposure Explain Associations between PM2.5 Exposure and Risk of Dementia?

Background: Epidemiological studies have reported positive associations between long-term exposure to particulate matter of 2.5 microns or less in diameter (PM2.5) and risk of Alzheimer's disease and other clinical dementia. Many of these studies have analyzed data using Cox Proportional Hazards (PH) regression, which estimates a hazard ratio (HR) for the treatment (in this case, exposure) effect on the time-to-event outcome while adjusting for influential covariates. PM2.5 levels vary over time. As air quality standards for PM2.5 have become more stringent over time, average outdoor PM2.5 levels have decreased substantially.

Objective: Investigate whether a Cox PH analysis that does not properly account for exposure that varies over time could produce a biased HR of similar magnitude to the HRs reported in recent epidemiological studies of PM2.5 and dementia risk.

Methods: Simulation analysis.

Results: We found that the biased HR can affect statistical analyses that consider exposure levels at event times only, especially if PM2.5 levels decreased consistently over time. Furthermore, the direction of such bias is away from the null and of a magnitude that is consistent with the reported estimates of dementia risk in several epidemiological studies of PM2.5 exposure (HR≈1.2 to 2.0).

Conclusions: This bias can be avoided by correctly assigning exposure to study subjects throughout the entire follow-up period. We recommend that investigators provide a detailed description of how time-dependent exposure variables were accounted for in their Cox PH analyses when they report their results.