Informative approaches for analysing common environmental factors in relation to health outcomes in large prospective epidemiological studies

Abstract

The study of the impact of environmental stressors on human health is a subject of great interest within the scientific community, with significant implications for public health. The aim of the thesis is to propose informative approaches for the analysis of common environmental factors in relation to health outcomes using data from large prospective epidemiological studies, and provide insights and understanding of exposure-outcome associations, including mechanisms of action and potential causal pathways, in order to prevent health effects. In Chapter 2, pooled data from the Dutch and UK cohorts of COSMOS were analysed to explore two potential exposure-outcome pathways in relation to mobile phone use and the occurrence of headaches. Results suggested that, due to the negligible exposure to RF-EMFs from texting, mechanisms other than RF-EMFs (e.g., lifestyle or behavioural factors associated with mobile phone use) were responsible for the increased risk of headache that we observed among mobile phone users. In Chapter 3, we explored the temporal dynamics of attribution of health complaints to RF-EMFs (IEI-RF) in the subcohort of AMIGO. Results showed that prevalence of IEI-RF was generally low and remained stable over time. Over 10 years of follow-up, we observed a high probability of not attributing symptoms to RF-EMF exposure anymore, which suggested that IEI-RF may be a more transient condition than previously assumed. In Chapter 4, we relied on a pluralistic approach to prospectively explore the relationship between a mixture of air pollutants (PM2.5, PM10, NO2, PM2.5 absorbance, and oxidative potential) and mortality in LIFEWORK. We applied a pluralistic approach to explore the associations and estimate the causal effect by estimating weighted quantile sum (WQS) regression and boosted regression trees (BRT) to identify the most relevant components within the mixture, and a multivariate generalized propensity score (mvGPS) model to jointly estimate the causal effects of the pollutants on overall mortality. Results highlighted the primary role of PM2.5 within the mixture. The strategy adopted in this study can serve as an example for future research aiming at understanding the intricate interplay between environmental exposures and human health by considering the health effects of environmental mixtures rather than relying on a “one-at-a-time” chemical approach. In Chapter 5, we prospectively explored the urban component of the exposome of AMIGO in relation to headaches by using a combination of machine learning techniques. We followed a two-stage approach where we first applied Boruta to identify relevant exposures in the exposome-outcome association, and then estimated causal forest to quantify the causal effect of these exposures on the occurrence of headache. These results highlight the relevance of air pollution exposure and heat island effects in contributing to the reporting of weekly headache, and emphasise the necessity of a comprehensive assessment of common urban stressors to prevent health issues. In Chapter 6, I discussed the approaches proposed to identify underlying mechanisms of action and addressed the most relevant challenges encountered in the different studies, highlighting advantages and limitations of the methods used, and providing policy implications and possible directions for future research.