Evaluation of a Method to Indirectly Adjust for Unmeasured Covariates in Large Administrative Data Cohort Analyses: An Analysis of Associations between Fine Particulate Matter and Mortality in the 2001 Canadian Census Health and Environment Cohort (2001 C

摘要

Background. The method of indirect adjustment for unmeasured confounding is a promising technique in large cohort studies, however no formal evaluation of non-linear survival models has been conducted. Objectives: To describe and evaluate the indirect adjustment method for air pollution -mortality relationships with a large longitudinal cohort, the 2001 Canadian Census Health and Environment Cohort (CanCHEC, N=2.4 million), and a representative matching dataset (pooled cycles of the Canadian Community Health Survey, [CCHS], N=450,000) with detailed behavioural risk factor information. Methods: We compared the distribution of exposure to fine particulate matter (PM2.5) among subjects across multiple characteristics (age, sex, etc.), and examined the direction and magnitude of correlations amongst variables common to both datasets. We assessed the performance of indirect adjustment on non-linear Cox proportional hazard models by comparing estimates in the CanCHEC with and without indirect adjustment for known variables. Further, we estimated the degree of bias correction by removing variables available in both the CCHS and CanCHEC (e.g. education, income) and evaluated differences. Results: At baseline PM2.5 distribution profiles across population characteristics were similar in both cohorts. PM2.5 levels for CCHS participants tended to consistently be 1.8-2.0 μg/m3 lower than the CanCHEC cohort, likely due to sampling differences between urban and rural areas since applying a sample-weighting scheme to the CCHS largely corrected for this discrepancy. Initial non-linear validation tests indicate possible underestimation of the PM2.5 association with further testing on-going. Conclusions: Formal evaluation using a large longitudinal mortality cohort and representative health survey will help establish protocols that other jurisdictions can use to assess the viability of the indirect adjustment method when information on potential confounding variables are not available.