Date of Award
Spring 2024
Document Type
Dissertation
Degree Name
Doctor of Philosophy (PhD)
Department
Public Health
First Advisor
Rogne, Tormod
Abstract
Non-Hodgkin lymphoma (NHL) is one of the most common hematologic malignancies in the world. Despite substantial efforts to identify causes and risk factors for NHL over the past few decades, the etiology of NHL is largely unclear. The overarching goal of this three-part dissertation is to use different meta-analytical and causal inference methods to identify and evaluate potential environmental risk factors for NHL. For the first project of this dissertation, we conducted an umbrella review to summarize the range, strength, and consistency of all associations between environmental risk factors and NHL reported in published meta-analyses. We identified 85 meta-analyses of summary level data reporting 257 associations for 134 unique environmental risk factors and 10 NHL subtypes. The vast majority of (79, 93%) meta-analyses of summary level data were rated as having critically low quality based on established evidence rating criteria. Most (225, 88%) associations presented either non-significant or weak evidence. Only 5% of the associations, primarily those for autoimmune and infectious diseases, were supported by the highest level of evidence (P < 10-6, at least 1000 NHL cases, largest study in the review reporting a nominally significant result, minimal between-study heterogeneity, and no evidence of publication bias). Autoimmune diseases have long been suspected as possible risk factors for NHL, and potential mechanisms for the relationship include chronic inflammation, antigen stimulation, and overlapping genetic susceptibility. To demonstrate the use of meta-analytical and causal inference methods to identify and evaluate potential environmental risk factors for NHL, we selected all ten statistically significant associations between autoimmune diseases and NHL (Behçet's disease, coeliac disease, dermatitis herpetiformis, psoriasis, rheumatoid arthritis, sarcoidosis, systemic lupus erythematosus, Sjögren's syndrome, systemic sclerosis, and type 1 diabetes [T1D]) for further evaluation in the second and third projects of the dissertation. In particular, given that these associations were all from observational studies, which are generally susceptible to different biases, this project highlighted the need for additional research evaluating their validity. For the second project of this dissertation, we (1) systematically identified and summarized quantitative bias analysis (QBA) methods proposed in the peer reviewed literature and generated a comprehensive classification tool that can facilitate the identification of QBA methods for studies with similar characteristics, and (2) used the classification tool and QBA methods to evaluate the impact of a potential unmeasured confounder (Epstein-Barr virus [EBV] infection) on the associations between the ten autoimmune diseases and NHL. We identified 55 QBA methods for summary level data, of which half (28, 51%) were designed for unmeasured confounding. Using the classification tool, we identified and applied five QBA methods for unmeasured confounding to the associations between ten autoimmune diseases and NHL. These analyses suggested that while effects of EBV infection as an unmeasured confounder could not significantly change the observed associations between autoimmune diseases and NHL risk, other unmeasured confounders, such as those with significant differences in prevalence rates between general population and autoimmune disease patients, could nullify or even reverse the observed associations from meta-analyses of observational studies. For the third project of this dissertation, we conducted a series of Mendelian randomization (MR) analyses to evaluate the associations between the previously identified ten autoimmune diseases and risk of NHL. The MR studies, which were carried out using large-scale genetic association studies, are less susceptible to unmeasured confounding than conventional observational studies. We observed negative associations between T1D and the risk of NHL, and between sarcoidosis and the risk of NHL (odds ratio [OR] 0.95, 95% confidence interval [CI]: 0.92 to 0.98, and OR 0.92, 95% CI: 0.85 to 0.99, respectively). No significant associations were found between the other eight autoimmune diseases and NHL risk. The findings from this MR study suggested that the effect estimates from conventional observational studies evaluating the associations between autoimmune diseases and an elevated risk of NHL may have been overestimated due to confounding. To triangulate the evidence, we compared the consistency of results in terms of direction and significance from the umbrella review (which identified meta-analyses, Project 1), bias analyses (Project 2), and MR analyses (Project 3). While results from the meta-analyses suggested that all ten autoimmune diseases were significantly associated with an increased risk of NHL, the bias anayses suggested that three were not significantly associated with and four were signficantly associated with a decreased risk of NHL, respectively. The MR analyses suggested that only two autoimmune diseases (T1D and sarcoidosis) were signficantly associated with a decreased risk of NHL; the other eight autoimmune diseases were not significantly associated with NHL. Overall, none of the associations were aligned across all three approaches. By combining meta-analytical and causal inference methods, we found that the associations between ten autoimmune diseases and an increased risk of NHL from published observational studies and meta-analyses are likely susceptible to unmeasured confounders. Future studies are warranted to further examine the impact of additional unmeasured confounders and other systematic errors (i.e., information and selection bias) on the associations between the ten autoimmune diseases and NHL. The novel integration of different meta-analytical and causal inference approaches for evidence triangulation proposed in this dissertation can serve as a template to guide researchers in environmental epidemiology when conducting investigations of environmental risk factor-health outcome associations.
Recommended Citation
Shi, Xiaoting, "Utilizing Meta-Analytical and Causal Inference Methods to Evaluate the Impacts of Environmental Risk Factors on Non-Hodgkin Lymphoma" (2024). Yale Graduate School of Arts and Sciences Dissertations. 1393.
https://elischolar.library.yale.edu/gsas_dissertations/1393
