Abstract 20P
Background
Lung cancer is one of the most common and deadliest cancers with a 5-year survival rate of ∼ 20%. Approximately 70% of cases are detected when the disease is regional or distant and have lower survival rates than localized cancers. Improving the number of cases detected in a localized state could lead to better patient outcomes.
Methods
Non-small cell lung cancer (NSCLC) plasma samples were tested for protein biomarkers carried by extracellular vesicles (EVs) following isolation and analysis using multiplex ELISA. In total, 161 lung cancer cases (Stage I = 88, Stage II = 34, Stage III = 20, Stage IV = 19) and 683 control subjects (no known cancer diagnosis) were analyzed. The case cohort consisted of 41% smokers with a median age of 63 years and 42% female population while the control cohort consisted of 12% smokers with a median age of 58 years and 52% female population.
Results
A machine learning algorithm identified an initial set of 35 informative EV biomarkers for differentiation of NSCLC cases from controls. The ROC showed an AUC of 0.987 (95% CI: 0.981 – 0.993) with a sensitivity of 97% (CI: 93% - 99%) at a specificity of 92% (CI: 90% - 94%). When performance was broken down by stage, we found similar performance across all stages tested with sensitivities of 98%, 94%, 95% and 100% for stages I – IV, respectively. We then reduced the number of biomarkers used from 35 to 11 to identify the most informative, yielding a 92% sensitivity at 92% specificity. To better understand performance in a future independent validation, we performed an in-silico experiment randomly partitioning the dataset into ten training and validation sets (67%/33%). This showed a mean AUC of 0.982 (range 0.968-0.991), mean sensitivity of 95% (range 85% - 100%) and mean specificity of 92% (range 86% - 95%) across the ten held-out validation sets.
Conclusions
Our pilot study suggests a potential approach for detecting lung cancer at early stages, when treatment can be more effective, by using a liquid biopsy test based on EV biomarkers. Our results warrant further testing with independent datasets for biomarker and algorithm validation. Future planned studies will also evaluate the impact of controls with confounding disorders, such benign lung nodules, as well as COPD to better evaluate the best use in the clinical setting.
Clinical trial identification
Editorial acknowledgement
Legal entity responsible for the study
The authors.
Funding
Biological Dynamics.
Disclosure
J.P. Hinestrosa: Financial Interests, Personal, Full or part-time Employment: Biological Dynamics. J.M. Lewis, H. Balcer, G. Schroeder: Financial Interests, Institutional, Full or part-time Employment: Biological Dynamics. P. Billings: Financial Interests, Institutional, Member of the Board of Directors: Biological Dynamics. All other authors have declared no conflicts of interest.