University of Arizona
Environmental Health Science
Client: Muath Almoslem, College of Public Health, University of Arizona advisor Dr. Jeff Burgess.
Consultants: Dean Billheimer, Chen Chen, Shahin Mohammadi, Mirjana Glisovic-Bensa (author)
November 7, 2019, 1-2pm
There are 1.05 million firefighters in the US who are at risk of being exposed to harmful substances in the course of their work (NFPA, 2019). The rate of cancer diagnoses among firefighters is 9% higher compared to the general US population (CDC, 2017). This study aims to determine whether aberrant DNA methylation occurs in firefighters over time. Also, we will investigate the dose-response relationship between occupational fire exposure and any observed changes in DNA methylation rates.
The cohort consists of participants selected from the Tucson firefighter department (n = 48 participants), sampled at the time of recruitment, and after two years. The participants did not have any live-fire exposure prior to the study. The firefighters were asked to complete questionnaires affording their demographic information, body weight, height, smoking status, and live-fire occupational exposure history. We defined the study population as a non-smoker, male firefighters with no live- fire exposure, who had provided complete height and weight information at baseline and follow-up.
Blood samples were sent to the University of Utah DNA Sequencing and Genomic Core Facility to analyze the DNA methylation status using MethylationEPIC arrays (Illumina). Samples were scanned using an iScan instrument, and raw data consisted of 96 samples in image format. We will process the raw image dataset using the Bioconductor minfi package in R. We will normalize the data using Illumina’s reference factor-based normalization methods (preprocess Illumina) and Subset-quantile Within-Array Normalization (SWAN) for type I and type II probe bias correction (Zhou et al., 2019).
Statistical methods: The primary outcome is the DNA methylation status in 5 CpG sites after two years of occupational live-fire exposure.
Exposure: The primary exposure will be assessed using months as a firefighter, as the interval between the baseline and follow-up measurement varied by firefighter between 18 and 36 months, as well as the most recent fire exposure, and the number of incidents attended by each firefighter or the estimated time spent at each active fire.
Covariates: The principal investigator provided us with the following covariates to adjust for in the model:
- Body mass index (BMI) as continuous variable in kg/m2,
- Age in years,
- Ethnicity/Race (Non-Hispanic Caucasian and Hispanic Caucasian)
- Cell type.
Dr. Billheimer recommended to analyze each cell type separately.
Discussed whether to use option 1 or option 2:
Option 1: Linear mixed effect model approach:
Option 2: Linear regression on difference:
Conclusion: Option 1 more flexible but more complicated. Option 2 is simpler and better to use in this situation.
IV. Next Steps
See if these two different models will give the same results. Simulate random normally distributed data (then no need to use log) and see if two models will give us the same results.