Multidimensional Modeling for Never Smoking Lung Cancer Risk Prediction

Validation and Optimization of Multidimensional Modelling for Never Smoking Lung Cancer Risk Prediction by Multicenter Prospective Study

Lung Cancer is the leading cause of cancer-related deaths in Taiwan and worldwide and the incidence is also increasing. The payment for lung cancer which occupies the largest part of National Health Insurance expense is over 15 billion in 2018. Because about 80% lung cancer patients are smokers in western countries the low-dose computed tomography screening focuses on the smoking population It is quite different in South-East Asia particularly in Taiwan that 53% of Taiwan lung cancer are never-smokers and the etiology and the underlying mechanisms are still unknown. The preliminary results of prospective TALENT study indicated that family history plays a key role in tumorigenesis of Taiwan lung cancers but several important variables such as air pollution, biomarkers, radiomics analysis are not available limits the accuracy of lung cancer identification. Hence, it is critical to integrate most of factors involved in lung cancer formation into a multidimensional lung cancer prediction model which could benefit never-smoker lung cancers in Taiwan and East Asia even in the western countries. The investigators initiate a clinical study to validate the multidimensional lung cancer prediction model for never-smoking population by multicenter prospective study.

To achieve the goal there are four programs proposed. Program 1: Validating non-smoker lung cancer prediction model among Taiwanese population: Integration with environmental and occupational factors. The investigators aim to enhance the accuracy of lung cancer prediction among Taiwanese non-smokers by incorporating environmental and occupational risk factors. The main aim of this program is to validate and optimize existing prediction models with more comprehensive epidemiologic, environmental and occupational factors with machine learning algorithms. The other aim is to validate current PM2.5-based lung cancer risk prediction models among nonsmokers, and optimize existing model with environmental and occupational factors in higher resolution. The investigators hypothesize adding more GIS-based environmental exposure measurements, and occupational exposure using job-exposure matrix as proxy can increase the predictive power of lung cancer risk model. Program 2: Validation of autoantibody- and genetic prediction model for non-smoker lung cancer. The investigators detect the autoantibodies against p53, NY-ESO-1, CAGE, GBU4-5, HuD, MAGE A4 and SOX2 in the blood of recruited patients and detect 133 SNPs and 11 mitochondrial mutations which are highly correlated with never-smoking lung cancer in our preliminary data. The investigators will validate the prediction power of these autoantibodies and genetic biomarkers in the early diagnosis of patients with high risk of acquiring lung cancer in Taiwan. Program 3: Detection, classification, prediction of lung cancer risk in CT using deep learning and radiomics. The investigators propose an integrated platform for detecting and following up lung nodules. A similarity measurement approach between two nodules is proposed. Base on Lung RADS assessment, the investigators plan to perform CT-radiomic analysis for nodules larger than or equal to 6-8 mm diameter aimed to find nodules in higher risk of developing lung cancer. The lung nodules will be detected and followed up by using a series of AIs. The detected nodules could be used for producing report and estimating Lung-RADS. Though Lung-RADS has considered the risk of malignancy based on their categories, the expectation of this project is to efficiently select CT screen high risk lung nodule(s) by using volume measurement, morphology, texture and CT radiomics of the detected nodules in addition to Lung-RADS criteria based on nodule size and characters. Program 4: Optimization and validation of lung cancer risk and probability prediction model: prospective multicenter clinical study. The program 4 will first use retrospective cohort based the case control research design to optimize the lung cancer risk models from program 1 and the biomarker and imaging models from program 2 and 3, respectively. The prospective multi-center research design will further use to verify the optimized predictive model. The high-risk participants will be selected to measure for biomarkers and undergo LDCT. The optimized biomarker model and image feature models will be performed to predict the probability of lung cancer and compared it with conventional clinical diagnosis methods and low risk participants. Finally, the Taiwanese population suitable lung cancer screening strategy will be proposed.

Participants will receive the following things in sequence Non-smoker lung cancer prediction model among Taiwanese population by questionnaire Check autoantibodies against p53, NY-ESO-1, CAGE, GBU4-5, HuD, MAGE A4 and SOX2 in the blood of recruited patients and detect 133 SNPs and 11 mitochondrial mutations which are highly correlated with never-smoking lung cancer in our preliminary data In high risk group,arrange chest CT and detection, classification, prediction of lung cancer risk in CT using deep learning and radiomics Optimization and validation of lung cancer risk and probability prediction model: prospective multicenter clinical study

Lung cancer detection rate differences between the high lung cancer risk group and the low lung cancer risk group.

Participants will receive the following things in sequence 10,000 non-smoker participants will receive a prespecified questionnaire Autoantibodies will be checked including p53, NY-ESO-1, CAGE, GBU4-5, HuD, MAGE A4, and SOX2 in the blood of recruited participants. All 133 SNPs and 11 mitochondrial mutations will be detected which are highly correlated with never-smoking lung cancer in our preliminary data In the high-risk group, the investigators will arrange LDCT scans for four rounds to determine the lung cancer detection rate. Also, the pulmonary nodule lesions detected will be classified by Lung-RADS and prediction of lung cancer risk in CT scans using deep learning and radiomics. In the low-risk group, the matched participants will receive LDCT scans for two rounds to determine the lung cancer detection rate.

Through steps 1,2, and 3 of the above column in primary outcome 1, the lung cancer risk model will be developed with optimization and validation of lung cancer risk and probability prediction model by this prospective multicenter clinical trial. ( predicted Area under curve (AUC) > 0.8)

Inclusion Criteria: Age 50-80 years old First-degree relatives of lung cancer patients aged more than 50 - 80 years old or older than the age at diagnosis of the youngest lung cancer the proband in the family if they are less than 50 years old Exclusion Criteria: Previous history of lung cancer Another malignancy except for cervical carcinoma in situ or non-melanomatous carcinoma of the skin within 5 years An inability to tolerate transthoracic procedures or thoracotomy Chest CT examination was performed within 18 months Hemoptysis of unknown etiology within one month Body weight loss of more than 6 kg within one year without an evident cause A known pregnancy