SUMMARY

Radiation-induced health effects have been extensively studied for almost a century. While health risks from radiation exposure at high doses and rates are relatively well-established from various epidemiological studies, the risk at low doses and rates is difficult to distinguish from baseline cancer risk and confounding factors. This study aims to investigate associations between residential radon exposure and any associated health risks, specifically lung cancer, using a novel approach that integrates multivariable methods and machine learning techniques. Radon exposure has been linked to lung cancer; however, there are limited research methods for adopting advanced methods for practical risk assessment.
This work has five primary aims:
1. Evaluating multivariable and machine learning methods for modeling population-level lung cancer incidence due to environmental radon exposure;
2. Developing a radon concentration prediction model that accurately estimates exposure risk at a smaller geographic scale (ZCTA, Zip-Code Tabulated Area);
3. Exploring the relationship between lung cancer incidence and radon concentration, alongside other potential interacting variables, at this smaller geographic scale;
4. Analyzing and discussing the data quality for modeling lung cancer risk with radon exposure; and
5. If warranted, create a framework for alternative radon-health risk exposure and/or chronic low-dose radiation exposure models.
The proposed methodology involves adopting Poisson regression and Poisson random forest to analyze various factors affecting lung cancer incidence and radon concentrations at the county and ZIP Code Tabulation Areas (ZCTAs) levels. The ultimate goal is to generate improved predictions and estimates that inform public health strategies and interventions. This study's significance lies in its potential to enhance understanding of radon-induced health risks, thereby improving understanding of public health outcomes from chronic low-dose radiation exposures.