ABSTRACT Lung adenocarcinoma (LUAD) is the most common lung cancer subtype diagnosed in the US; characterized by a broad spectrum of biological behaviors and clinical trajectories. Yet, LUAD is managed uniformly based on clinical stage, with the potential for under- and over-treatment of aggressive and indolent lesions, respectively. This contributes both to suboptimal lung cancer outcomes and unnecessary morbidity, mortality and healthcare costs. While histologic grade of resected tumors correlates with patient outcome, it is only available after surgical treatment and cannot be used to inform pre-surgery management or surgical planning. We have developed and validated CANARY, a radiomic biomarker that predicts LUAD aggressiveness. We have further developed two gene expression biomarkers from resected FFPE Stage I LUAD for predicting indolent or aggressive tumor histology. These gene expression biomarkers are insensitive to intratumoral heterogeneity, suggesting that they might retain good performance when measured in limited tissue available from small, presurgical biopsies. This is potentially transformative as histologic assessment of these small biopsies is frequently insufficient for predicting tumor aggressiveness. Our goal is to refine and validate these radiomic and gene expression biomarkers and then integrate them into a single model for detecting indolent and aggressive Stage I LUAD, which is supported by our preliminary data. To accomplish these goals, we will prospectively enroll a cohort of patients undergoing transthoracic or transbronchial biopsy for suspected lung cancer and collect additional specimens for research. In the subset of tumors who are later resected for Stage I LUAD, we will perform a central pathologic assessment of tumor grade. Predicting tumor histologic grade at resection will be the primary endpoint for assessing the performance of the integrated presurgical prediction model. Refinement of the radiomic biomarker will involve testing whether the addition of features extracted from the peri-nodular lung using deep learning can improve the prediction of the Stage I LUAD histologic grade. Refinement of the gene expression biomarker will involve determining their performance in biopsy tumor tissue relative to resected tumor tissue and optimizing the biomarkers for assessment in biopsies. Finally, we will develop and assess an integrated model combining both radiomics and gene expression. As a secondary endpoint, we will compare the association between recurrence free survival and predicted tumor grade vs. actual tumor grade at resection. An improved ability to predict tumor aggressiveness prior to treatment has the potential to transform the management of Stage I LUAD as it could allow clinicians and patients to confidently choose precisely tailored treatment. The team from Boston University, Boston Medical Center, Vanderbilt University Medical Center, and Lahey Hospital & Medical Center has the diverse expertise in lung cancer clinical care, advanced bronchoscopy, interventional radiology, histology, pathology, radiology, radiomics, molecular biology, genomics, bioinformatics, deep learning and biostatistics required to complete this project.

PROJECT NARRATIVE There is currently no clinically accepted way to predict a lung cancer's likely aggressiveness prior to removing the tumor via surgery, resulting in imprecise treatment. We propose to characterize the aggressiveness of early-stage lung adenocarcinoma (the most common lung cancer subtype) before treatment, based on detailed and quantitative analysis of chest CT and molecular analysis of a small biopsy of the tumor obtained before surgery. The ability to determine the aggressiveness of lung cancer could facilitate treatment tailored to individual patients: improving lung cancer survival in patients with the most aggressive tumors and decreasing complications from lung cancer treatment in patients with the least aggressive tumors.