Tumor-infiltrating lymphocytes, environmental exposures, genetic mutations and the risk of selected neoplasms

Approximately 20 million new cancer cases and 9.7 million cancer deaths occurred in 2022, making cancer the third cause of death in the world. Cancer types with the highest burden on population’s health are respiratory and digestive cancer, and among the latter colorectal and liver cancer have the highest mortality rates. Previous work suggested that tumor-infiltrating lymphocytes (TILs) detected in cancer tissue are predictive of better prognosis among cancer patients. TILs can be evaluated visually by pathologists in hematoxylin and eosin-stained tissue sections. However, TIL assessment made by experts is a time-consuming task and is affected by interobserver variability. Hence, machine learning-based methods, such as deep learning, have been tested to analyze whole slide images of cancer tissue to overcome such potential issues related to human-based TIL estimation and to improve reproducibility. Previous studies evaluating TILs assessed using deep learning algorithms as a prognostic marker among cancer patients were focused on a specific cancer type only and mostly used cancer-specific algorithms for TIL estimation. Also, none of them investigated the combined use of TILs and other factors, such as exposure to environmental pollutants and genetic mutations, in the assessment of prognosis among cancer patients. The results described in this thesis suggest that TILs evaluated using a deep learning algorithm validated for many different cancer types may be useful for cancer prognostic stratification, especially among head and neck cancer patients. Also, among evaluated factors, TILs and environmental exposure to air pollutants classified as carcinogens showed the most promising results as prognostic markers, while clusters of genetic mutations obtained using k-means algorithm showed no clear associations with cancer survival. Hence, the present thesis provides a proof of concept for the integration of machine learning and classical epidemiological methods for the identification of prognostic factors in patients with respiratory and digestive cancer.