Bias Analyses in Population-Based Studies of Parkinson’s Disease
- Author(s): Cui, Xin
- Advisor(s): Ritz, Beate
- et al.
This dissertation work investigates two types of biases in population-based studies of Parkinson’s disease (PD), one is survivor bias in estimating the association between PD and cancer, the other is exposure misclassification due to residential mobility and time-varying exposure. Negative associations between PD and cancer have been found in epidemiological studies. Several mechanisms were proposed to explain such reported inverse associations. In the first and second part of this work we propose two similar survivor bias mechanisms that may account for the observed negative associations with cancer both prior to and after the diagnosis of PD as an alternative explanation. Using a large Danish population-based case-control study of Parkinson’s disease as an example, we utilize causal theory, Monte Carlo methods and inverse probability-of-censoring weights techniques to quantitatively investigate how the observed negative association can be explained by the hypothesized bias. These results suggest that for cancer both before and after the diagnosis of PD, survivor bias could be an alternative explanation for the observed association between cancer and PD with reasonable bias structure and assumptions. In the last part of this work we investigate possible exposure misclassification due to residential mobility and changes in pesticide application using a California population-based case-control study of Parkinson’s disease as an illustration. We simulate scenarios where detailed residential histories were lacking and only the enrollment address was used as a proxy for all addresses to estimate long-term pesticide exposures. Results show that the exposures could be either over- or under- estimated depending on pesticide, time period of interest, as well as threshold for identifying the binary exposure variables. The exposure misclassification is not necessarily non-differential and the direction of the bias is inconsistent. When estimating long-term environmental exposures using one address only it may result in exposure misclassification and not always guarantee non-differentiality.