Investigating the relationship between ionizing radiation and neurodegenerative mortality: addressing issues of bias, data pooling, and effect modification

Abstract

Background Much of the current understanding of the adverse health effects of ionizing radiation (IR) exposure comes from studies of Japanese atomic bomb survivors who experienced acute, high doses of IR. However, these findings may not be generalizable to populations exposed to chronic low doses of IR such as workers and the general population. The U.S. Million Person Study (MPS) was founded to investigate the effects of chronic low dose IR for improved guidance of radiation safety. To address this gap in knowledge, worker cohort studies have been and continue to be conducted to assess the potential risks associated with chronic low dose IR, including neurodegenerative diseases. These studies aim to improve the understanding of IR dose ranges and risks that underly current radiation safety policy. Recent findings underscore a potential increase in risk of neurodegenerative disease from chronic low dose IR which may be of great concern considering the significant burden of such diseases. However, there are inconsistencies in prior research findings regarding the relationship between chronic low dose IR exposure and neurodegenerative disease. Several factors we focus on that can contribute to these inconsistencies are co-exposures to non-radiological substances, lack of power to investigate specific neurodegenerative disease outcomes, and selection mechanisms like the healthy worker survivor effect (HWSE) and competing events. This dissertation research introduces novel techniques to evaluate and improve the internal and external validity of prior and future occupational studies of IR and neurodegenerative-related mortality. The specific objectives of the present study include assessing the role of co-exposures as effect modifiers and/or confounders, enhancing the understanding of the trade-offs of pooling data diverse cohorts together, and addressing selection mechanism concerns. Methods Our first objective focused on assessing the role of co-exposures in the relationship between IR and neurodegenerative-related mortality among workers at the Fernald Feed Materials Plant, referred to as Fernald, and to investigate neurodegenerative-related mortality risk across different job categories. Individual-level, brain IR dose estimates were used alongside co-exposure data from job exposure matrix application to evaluate potential confounding and effect modification. Co-exposures for this evaluation were selected a-priori based on literature review. Cox proportional hazards regression models were used to observe risk of neurodegenerative-related mortality at 10 milligray (mGy) IR brain dose across different groupings of co-exposure status, while logistic regression models were used to calculate odds of neurodegenerative-related mortality across 9 job categories. In our second objective our goal was to evaluate the implications of data pooling in the context of the relationship between IR exposure and neurodegenerative-related mortality by combining data from the Fernald, Linde Ceramics Plant, hereafter called Linde, and Mallinckrodt Chemical Works (MCW) cohorts. The Linde cohort's data was updated to prepare for merging with Fernald and MCW. Extensive data harmonization, such as collapsing continuous IR exposure measurement into a categorical form, was undertaken. Cox proportional hazards regression was again used in data analysis across three distinctive steps that analyzed different aggregation parts between the cohorts, co-exposures of uranium and silica dusts were evaluated. A meta-analysis was also conducted of these same three cohort's individual studies to qualitatively compare overall analogous estimates to the pooled analysis. Lastly, our third objective employed the parametric g-formula to address selection mechanism concerns of HWSE and competing events. The parametric g-formula estimated the effect of hypothetical interventions on the risk of neurodegenerative-related mortality to evaluate the potential for HWSE by reducing IR exposure in hypothetical intervention scenarios. Expert interviews were conducted and supported with literature review to develop feasible hypothetical intervention endpoints of reducing IR exposure with possible real-world explanations including engineering and administrative controls. To investigate competing events, in one approach we modeled a likely competing event, cardiovascular disease (CVD) mortality, and compared cumulative incidence results to another approach without modeling for CVD mortality. Results Aim 1 findings highlighted that occupational exposure to machining fluids may act as an effect modifier of the relationship between IR and neurodegenerative-related mortality. Workers highly co-exposed to machining fluids showed significantly increased risk of neurodegenerative-related mortality at 10 mGy IR brain dose (HR 1.25, 95% CI 1.11, 1.40) compared to the overall baseline estimate (HR 1.01, 95% CI 0.96, 1.06). Additionally, certain job categories such as store and supply services had elevated odds of neurodegenerative-related mortality compared to administrative workers (OR 2.30, 95% CI 0.97, 4.96). Findings in Aim 2 revealed overall results consistent with Aim 1 of no increased risk of neurodegenerative-related mortality at low/moderate IR category (HR 0.95, 95% CI 0.73, 1.24) and high IR category (HR 1.05, 95% CI 0.80, 1.37). There was no evidence of confounding or effect modification by silica or uranium dusts. The meta-analytic estimate of these same cohort's individual studies had a similar summary estimate null finding compared to the pooled analysis finding for high categorical IR (HR 0.95, 95% CI 0.75, 1.20). Results of Aim 3's application of the parametric g-formula indicated that the HWSE may not be a key factor in the relationship between IR and neurodegenerative-related mortality as cumulative incidence (CI) did not change considerably between the natural course (cumulative incidence, 16.99%), engineering control (17.43%), and administrative control (17.14%) interventions in the application modeling for competing events. When comparing approaches that modeled and did not model for competing events, we did find evidence that CVD mortality acts as a competing event for neurodegenerative-related mortality. Conclusions This dissertation found that occupational co-exposures, such as machining fluids, may modify the relationship between IR and neurodegenerative-related mortality. In a pooled analysis of worker cohorts we encountered significant data harmonization challenges that emphasized the importance of careful selection of data pooling candidates and meta-analyses as an alternative to aggregating data for some research questions. Furthermore, the selection mechanism of HWSE may not be a key factor for IR and neurodegenerative-related mortality in Fernald, but CVD mortality appears to act as a competing event to neurodegenerative-related mortality. These findings offer potential explanations and insights into the inconsistencies in prior research of IR and neurodegenerative-related mortality and highlight the need to consider co-exposures, data aggregation approaches, and selection mechanisms. This research contributes by showing the importance of evaluating the role of co-exposures as effect modifiers, providing insights into the goals of multi-cohort pooling efforts by discussing trade-offs of data pooling versus meta-analytic approaches, and exemplified a relatively innovative and novel method of addressing selection mechanism concerns in the field of occupational radiation epidemiology. Future research should focus on replicating these approaches in different circumstances that may alleviate the limitations of the present investigations to continue the overarching goal of best informing radiation safety policy for the general population and modern workforce.