- Main
Applications of Causal Inference to Problems of Occupational Epidemiology
- Brown, Daniel Martin
- Advisor(s): Petersen, Maya L;
- Van der Laan, Mark J
Abstract
This dissertation concerns the application of the techniques of causal inference to problems of occupational health. The abstracts of the three works which comprise the primary substance of this disseration are reproduced below.
The healthy worker survivor effect (HWSE) is a feature of occupational cohort studies which can lead to biased estimates of the etiologic effects of exposures if the estimation procedure does not account for its sources. The HWSE arises from underlying temporal processes characteristic of working populations in which time-varying health status is a criteria for entry into follow-up as well as both a predictor and a consequence of exposure. We distinguish two sources of HWSE: left-truncation in the presence of heterogeneous susceptibility as well as time-varying confounding on the causal pathway. We apply longitudinal minimum-loss-based estimation to simulated data in order to illustrate the effect of each process on estimates of exposure response, and clarify the extent to which methodological solutions can properly adjust for the bias.
We consider the problem of the estimation of parameters of the full-data distribution from data structures in which some confounding variables are unmeasured in a portion of the population. Our focus is on evaluating approaches to implementation of an augmented inverse probability of censoring weighted targeted minimum-loss based estmator (A-IPCW TMLE) first proposed by Rose and Van der Laan. This is an inverse probability weighted estimator in which estimation proceeds using a reweighted set of fully observed data points. The weights used are the inverses the estimated probability of being fully observed which is then augmented by an estimate of the expectation of the full data influence function, given the always observed variables. The estimator's performance is compared to standard weighting approaches and multiple imputation in both a simulation study and an applied data example.
We investigate the effect of cumulative exposure to particulate matter with an aerodynamic diameter < 2.5 um (PM2.5) on the incidence of ischemic heart disease (IHD) in a cohort of aluminum workers followed for 15 years, adjusting for time-varying confounding affected by prior exposure. We use longitudinal targeted minimum-loss based estimation (TMLE) to estimate the cumulative risk difference for IHD if always exposed above an exposure cut-off compared to always exposed below, while never censored. We stratify our analyses by sub-cohort employed in the smelters versus fabrication facilities. We selected two exposure cut-offs a priori, at the median and 10th percentile, within each sub-cohort. In smelters, the estimated IHD risk difference after 15 years is 2.1% (-1.3%, 5.5%) if always exposed compared to never exposed above the median cut-off of 1.77 mg/m3 and 2.9% (0.6%, 5.1%) using the 10th percentile cutoff of 0.10 mg/m3. For fabrication workers, the estimated risk difference is 0.9% (-1.6%, 3.5%) using the median cut-off of 0.20 mg/m3 and 2.5% (0.8%, 4.1%) using the 10th percentile cut-off of 0.06 mg/m3. Results are presented as marginal incidence curves, describing the cumulative risk of IHD for each sub-cohort under each intervention regimen. By control of the time-varying confounding on the causal pathway that characterizes healthy worker survivor effect, TMLE estimated associations between cumulative PM2.5 exposure and IHD that were not detectable using standard analytical techniques in a previous report. This represents the first longitudinal application of TMLE, a method for generating doubly robust semi-parametric efficient substitution estimators, in the field of occupational and environmental epidemiology.
Main Content
Enter the password to open this PDF file:
-
-
-
-
-
-
-
-
-
-
-
-
-
-