Improving Risk Prediction Models for Lung Cancer; The HUNT Lung Cancer Model and Integration of Genetic Factors and Circulating MicroRNAs

Abstract

Lung cancer is the leading cause of cancer mortality worldwide, claiming more lives than any other cancer types. Early detection of lung cancer is crucial for lung cancer survival, and lowdose computer tomography (LDCT) screening of high-risk current and former smokers has been shown to reduce lung cancer mortality by 20-24%. However, it is not only associated with benefits, but also risks and potential harms (e.g. false positive, overdiagnosis, anxiety and economic burden of the society). A major problem with LDCT screening is the screening eligibility. The current clinical criteria used (e.g. NLST, NELSON and 2021 USPSTF criteria) regarding screening eligibility has low specificity as well as sensitivity to predict who is at true high risk. To increase the benefit of CT lung cancer screening and reduce the potential harms, one needs methods to target subpopulations at highest risk of lung cancer and avoid people of low risk. The age of precision medicine urges us to develop a more risk-based approach to lung cancer risk stratification so that true high-risk individuals can be reliably predicted. Such a risk prediction approach can rely e.g. on pure clinical variables or biomarkers, or a combination. In this thesis, several aspects of this challenge are addressed, starting with the testing of a pure clinical risk prediction model (paper I), progressing to involve genetic information in risk stratification (paper II), and finally looking for microRNAs as early diagnostic/predictive biomarkers (paper III). The material used in this thesis is from several large prospective population-based studies including the Norwegian population Cohort of Norway (CONOR) (paper I); the Nord-Trøndelag Health 2 Study (HUNT2 study) and Tromsø study 4-6 (paper II); and the HUNT2 and HUNT3 studies, Norwegian Women and Cancer Study (NOWAC) and Northern Sweden Health and Disease Study (NSHDS) (paper III). The risk prediction model HUNT Lung Cancer Model (HUNT LCM) was previously shown to be more sensitive and effective compared to the NLST criteria in lung cancer risk prediction. Since then, the NELSON study came along with wider screening criteria as well as an updated recommendation from the USPSTF criteria in 2021. In paper I we further tested the lung cancer predictive performance of the HUNT LCM against the NELSON and 2021 USPSTF criteria in ever-smokers in CONOR (n=44,831), with results significantly in favor of HUNT LCM, both in lung cancer risk assessment as well as efficiency in detecting a single lung cancer case within a time horizon of six years. In paper II we showed that integrating genetic information from 22 selected single nucleotide polymorphism (SNPs) with the original eight clinical variables of the HUNT LCM can improve the predictive performance of lung cancer compared to the original HUNT LCM. This novel polygenic model, the HUNT Lung-SNP, significantly improved the lung cancer risk ranking and performance over the original HUNT LCM, and over the current clinical criteria of the NLST, NELSON and 2021 USPSTF criteria in two independent prospective cohorts, HUNT2 (n=65,240 ever-smokers) and Tromsø (n=2,663 ever-smokers). In paper III we searched for circulating microRNAs in serum as well as in plasma samples that could be useful as biomarkers for early diagnostic of lung cancer, either on their own or in combination with clinical variables. The study revealed 15 novel validated serum microRNAs as potential biomarkers for all the main histological subtypes (adenocarcinoma, squamous cell carcinoma and small-cell lung carcinoma) as well as the basket subtype nonsmall cell lung carcinoma (NSCLC) up to eight years before lung cancer diagnosis. They all achieved an independent AUC of >0.60 in both the discovery and validation serum datasets, except one. Their combined AUC reached >0.70 when the 15 microRNAs were analyzed as a signature. The study also showed that validating microRNA findings between serum and plasma should be avoided, given that the microRNA expression/level in serum and plasma seems to be different. Summing up, the results of this thesis have provided further evidence that the clinical risk prediction model, e.g. HUNT LCM, is better to predict future lung cancer risk than categorical selection criteria, and biological information such as genetical information and microRNA in blood can improve lung cancer risk prediction further. For future research, the results of this thesis should be tested prospectively in a clinical setting for lung cancer screening. In addition, the results should be explored to see whether the individual information of lung cancer risk provided by a risk prediction model, e.g. HUNT LCM or HUNT Lung-SNP, can be used along with smoking cessation programs to help motivate people to quit smoking.