Nutrition data science for lung cancer prevention and macrotrends

Abstract

According to the GLOBOCAN estimates, approximately 1.8 million deaths are attributable to lung cancer annually. While smoking is a well-known risk factor for lung cancer, some observational studies have identified associations between nutrition factors and lung cancer risk. However, the results are not always consistent. This thesis uncovers the association between nutrition factors and lung cancer using multiple data sources from the single nutrient, the gene-diet interaction and the macrotrend perspectives. From the single nutrient perspective, the role of dietary carbohydrates on lung cancer risk was investigated using data from the PLCO Cancer Screening Trial. Dietary carbohydrates and their food sources were categorized into quartiles by quantity and the lowest quartile was set as the reference. Multivariable binary logistic model, multinomial logistic model, and Cox regression model with time-varying coefficients were used to study the associations between dietary carbohydrates and lung cancer risk. The results showed that low-GI carbohydrates from fruits, vegetables and whole grains protected against lung cancer, while refined carbohydrates from soft drinks increased the risk of lung cancer. The association between dietary sodium and lung cancer risk was explored by multivariable Cox regression model using data from the PLCO Trial and the NIH-AARP cohort. The results indicated an inverse association between moderate sodium intake and lung cancer risk. The association needs further validation before inviting reflection on the sodium restriction policy. From the gene-diet interaction perspective, current statistical methods include the constrained maximum likelihood estimation, the unconstrained maximum likelihood estimation, and the empirical Bayesian (EB) estimation. The realistic simulation was conducted based on the single nucleotide polymorphisms (SNPs) from the PLCO GWAS data. The operating characteristics of the three statistical methods were tested in the simulated datasets. The results showed that the EB method could trade-off the type I error and statistical power for gene-diet interaction analysis. In the PLCO GWAS dataset, SNP rs7175421 was found to significantly interact with whole grain intake for lung cancer risk by the EB method. Future research can consider our workflow to select the appropriate statistical method for gene-diet interaction studies. From the macrotrend perspective, the global data of dietary factors, country income, and lung cancer incidence were extracted from the Global Dietary Database, the World Bank DataBank, and the 2019 Global Burden of Disease Study. Linear mixed model was used to study the trends of dietary sodium, fiber and whole grains by country income levels from 1990 to 2018. In addition, the linear trends of dietary sodium and lung cancer incidence were modelled by linear regression for individual countries. The study found increasing trends of dietary sodium, fiber and whole grains in middle-income countries. Moreover, none of the country-level income groups demonstrated strong agreement on trends of dietary sodium and lung cancer incidence. The dissertation investigates the relationship between dietary factors and lung cancer incidence by harnessing a holistic view of individual diet, gene-diet interaction, and dietary macrotrends at the country level. This work leverages multiple data sources to conduct nutrition data science investigations for lung cancer prevention.