Large retrospective validation study of metabolomic biomarkers for resectable lung cancer detection and risk assessment.

Abstract

3071 Background: Cancer can be regarded as a metabolic disease and many studies published aimed at identifying robust metabolites for lung cancer diagnosis using plasma samples. Regardless of the lung cancer staging, most of these studies were performed on relatively small sample sizes. The purpose of this study is to validate whether a panel of metabolomic biomarkers would improve risk assessment for lung cancer detection in 800 plasma samples from patients that underwent lung cancer resection, and to understand the potential role and intersection between lung cancer and other lung diseases. Methods: A blinded case-control study was performed using plasma samples from 586 patients with biopsy-confirmed lung cancer compared to 214 controls from the same institutional biorepository to evaluate the performance of a 5+ metabolites biomarker panel for lung cancer detection. The control group consists of 90 healthy individuals and 124 with other non-neoplastic lung diseases including asthma, COPD, bronchiectasis and COVID. The lung cancer subgroups include early-stage (Stage I &II) adenocarcinoma and squamous cells carcinoma, advanced stages (Stage III & IV) NSCLC and neuroendocrine tumors. In this study, plasma metabolic profiles were performed using different liquid chromatography methods coupled with a targeted and quantitative mass spectrometry approach. Metabolite concentrations, clinical data, and smoking history were used to develop logistic regression models to identify lung cancer at different stages using significant biomarkers. The area under the receiver operator characteristic curves (AUC), sensitivities and specificities at selected cut off points were calculated for each cancer stage. Results: Univariate and multivariate statistical analyses confirmed the performance of our 5+ metabolites panel (β-HBA, PC aa C38:0, PC ae C40:6, citric acid, tryptophan, and etc.) as being significantly different between controls and lung cancer cases. Linear regression model using metabolites alone yielded an AUC of 0.89 for lung cancer at all stages. When adding smoking status, the models achieved an AUC of 0.91 with sensitivity of 91% and specificity over 78% using a risk prediction threshold of 0.657. Conclusions: The blood-based metabolites panel validated on our current retrospective study exhibited robust performance for resectable lung cancer detection and risk assessment. This metabolomic panel assay demonstrates potential diagnostic applications and clinical utility for patient selection that require further follow-up and confirmation using LDCT or other lung imaging modalities. The inclusion of other lung diseases in the control group is more representative of a real clinical setting suggesting that this blood-based assay could be offered to the medical community. Further studies may also confirm its utility in the context of a lung cancer screening program.