NIMBLE is a prospective study for blood biomarker study of lung nodules alongside analysing data which has been collected routinely as part of patient care. The primary aim of NIMBLE is to assess whether artificial intelligence and machine learning based radiomics approaches can be used to distinguish between benign disease and malignancy in a new lung nodule after a previously treated cancer, and where malignant to differentiate between metastatic recurrence or a new primary lung cancer.

1.1 Lung cancer & Indeterminate Lung Nodule Surveillance Over 46,000 cases of lung cancer are diagnosed every year in the UK, making it the 3rd most common cancer type. Lung cancer is the biggest cause of cancer mortality in the UK and worldwide due to late presentation in the majority of cases. One-year survival for lung cancer ranges from 83% at stage I to 17% in stage IV disease (CRUK data).

1.2 Incidental Lung Nodules A significant challenge posed by lung screening is the identification of incidental lung nodules. 9.3% of all patients in the NELSON study had indeterminate nodules, and only 10% of these were diagnosed with cancer.

Such nodules are very frequently picked up on CT scans performed for other reasons, and may generate anxiety and uncertainty for patients and clinicians as well as using considerable NHS CT scan capacity. Current methods of stratification are based on a combination of The British Thoracic Society guidelines and the Brock, Herder and Fleischner risk models. Depending on the size of the lesion, guidelines recommend surveillance CT scans at 3-12 monthly intervals for solid and sub-solid lesions. Previous studies have suggested that persistent sub-solid nodules have a high risk of malignancy (~63%), and using Brock guidelines, larger nodules are often referred for biopsy (Henschke, 2002). However, a proportion of patients who score highly on these models will have negative biopsies, and there is a definite need for improved stratification.

In the screening setting, identification of early lung cancers and nodules in 'Lung Health Checks' - which use 'low dose' CT (LDCT) scan screening of high-risk populations (e.g. heavy smokers) has been shown to reduce lung cancer mortality by 20-26% as observed in the National Lung Cancer Screening Trial (NLST) and NELSON studies. A number of pilot trials within the UK have led to a commitment by NHS England to roll-out a £70m national program in a number of test sites. This program will lead to an expected 10% indeterminate finding rate putting further strain on the management of indeterminate nodules. RM Partners is undertaking one of the early lung screening pilots that led to this program across two clinical commissioning groups (CCGs) in West London in 2018, inviting over 8000 patients for a lung health check. This pilot has been extended in 2019-2020 and will also be incorporated in the NHS England National program.

1.3 Imaging and blood biomarkers in lung cancer early diagnosis Recent data suggest that the application of machine-learning approaches to the NLST trial data improves radiological risk-stratification of nodules (Ardila et al., 2019). Through the retrospective RMH LIBRA study, we are currently developing radiomics and Artificial Intelligence (AI) signatures to stratify lung nodules in patients from across the London cancer alliances. There is increasing interest in multi-model approaches, and the incorporation of 'multi-omic' data may enhance diagnostic accuracy and risk stratification (Bakr et al., 2018; Lu et al., 2018).

Lung cancer biomarker development is a rapidly evolving field that spans genetics approaches such as ctDNA sequencing and methylation studies, to more indirect measures of a systemic response to active malignancy in order to indicate the presence of cancer such as metabolomic and immunophenotyping studies. There is considerable interest in using such lung nodule populations for development of lung cancer biomarkers where a positive result would represent very early stage disease. The identification of non-invasive predictive and prognostic biomarkers is therefore an important priority. This data set thus represents an important cohort to translate discovery science to patient facing clinical assays that could facilitate earlier cancer diagnosis.

1.4 Tumour Immunophenotyping Observations that cancer relapse is related to the neutrophil-lymphocyte ratio, and that lung cancer development appears related to changes in interferon signalling (Mizuguchi 2018, Beane 2019) lead us to hypothesise that immune phenotyping may have a role to play in the early-diagnosis setting. Recent advances in flow and mass cytometry now allow high dimensional immunophenoyping, through simultaneous measurement of ~40 markers per cell. Hence the central challenge of this project is to develop a more detailed understanding of the host immune phenotypes that are associated with cancer development risk, based on longitudinal high dimensional immunophenotyping, rather than low dimensional measurement of single markers. We hypothesise high dimensional data will allow a more detailed, and context resolved, set of immune phenotype states to be defined, which can be developed into accurate biomarkers to predict the risk of tumour development and relapse. Indeed, in support of this hypothesis, high dimensional immune phenotypes have already been discovered which can predict all-cause mortality in longitudinal studies of heart disease. We have conducted pilot analysis of an existing CRUK cohort of early stage lung tumour patients already recruited through the TRACERx study, to demonstrate the feasibility of high dimensional immune phenotyping in patient samples. NIMBLE will tackle an underlying challenge of work in this area which is a shortage of clinical pre/non-malignant samples with longitudinal follow up.

2. Rationale Incidental lung nodules are common, and may represent early cancers. Their assessment can result in delayed diagnosis while interval imaging is performed to assess risk.

This study will allow us to examine the potential for imaging and blood biomarkers to augment nodule stratification, and identify high-risk patients who may benefit from more frequent surveillance or earlier diagnostic procedures, and low risk patients suitable for reduced surveillance intensity. This is particularly relevant for the COVID-19 era to stratify hospital attendances and high risk interventions to those in greatest need. This project dovetails with existing radiomics and lung biomarker research (LIBRA and Lung Health Check Biomarker Study) within our early diagnosis research group.

3. Hypothesis

Primary Hypothesis: Peripheral blood Immune phenotype differences will be present between benign and malignant lung nodules, which can be developed into accurate biomarkers to predict the risk of tumour development and relapse.

Secondary hypothesis: Combined use of blood and imaging biomarkers will enhance malignancy prediction in patients with incidental lung nodules.

Exploratory hypothesis: Blood biomarkers such as immunophenotyping or metabolomics ± radiomics vector, when measured as a continuous variable will see a decrease in risk score following tumour resection or regression.

Inclusion Criteria:

• Patients under active investigation or surveillance for incidental lung nodules
• Age > 18.

Exclusion Criteria:

• Active or previous diagnosis of malignancy (within 5 years preceding baseline scan).
• Inability to give informed consent.
• Active infection (including tuberculosis or fungal infection).
• Clinician-suspected or confirmed active or recent COVID-19 infection (less than 4 weeks before CT scan or required blood sampling date).

Patients who have CT scans for lung changes (nodules) who meet the eligibility criteria.