VarSee™ (Lung), a Blood-Based Multi-Protein Biomarker Panel for the Early Detection of Lung Cancer: A Prospective Pilot Diagnostic Accuracy Study

Abstract Background Most non-small-cell lung cancers are diagnosed at an advanced stage because early disease is largely asymptomatic. Low-dose computed tomography (LDCT), the recommended screening modality for high-risk individuals, is constrained by radiation exposure, limited accessibility and a high false-positive rate. We evaluated VarSee™(Lung), a serum multi-protein biomarker panel coupled to a machine-learning classifier, as a blood-based adjunct for lung-cancer detection. Methods In a consecutive, prospective, single-centre, case-control pilot study with biomarker measurement performed retrospectively on banked serum samples, seven serum proteins (CA125, CA19-9, carcinoembryonic antigen [CEA], cytokeratin-19 fragment [CYFRA 21 − 1], human epididymis protein 4 [HE4], neuron-specific enolase [NSE] and progastrin-releasing peptide [ProGRP]), quantified by chemiluminescence immunoassay (CLIA), together with the neutrophil-to-lymphocyte ratio (NLR) and four clinical covariates, were measured in 100 participants (50 with histologically confirmed lung cancer and 50 non-cancer controls) and entered into a calibrated random-forest classifier (the VarSee™ (Lung) algorithm) to generate a diagnostic score. Diagnostic performance was assessed against a histopathological/CT reference standard using stratified five-fold cross-validation, reported in accordance with STARD 2015 and, for the multivariable model, TRIPOD. Results VarSee™ (Lung) distinguished lung-cancer cases from controls with a sensitivity of 100% (95% CI 92.9–100), a specificity of 96.0% (95% CI 86.5–98.9), a positive predictive value of 96.2%, a negative predictive value of 100%, an accuracy of 98.0% (95% CI 93.0-99.4) and an area under the ROC curve of 0.997 ((95% CI 0.990–1.000). In this small cohort, sensitivity was 100% at every tumour stage (I-IV). The NLR was the dominant predictor (impurity importance 0.501), followed by CA19-9 (0.116) and CYFRA 21 − 1 (0.066); accuracy and AUC were consistent across five candidate classification algorithms (accuracy 0.950–0.980; AUC 0.995–0.997). Conclusions In this pilot, a blood-based seven-protein panel integrated with the NLR discriminated lung cancer from cancer-free controls with high sensitivity and specificity, supporting further evaluation as a radiation-free adjunct for early detection. Given the small, single-centre design and cancer-free (rather than benign-disease-matched) controls, these findings are hypothesis-generating and require confirmation in a larger, multi-centre study with external validation before any clinical or screening application is considered. Trial registration: Clinical Trials Registry - India, CTRI/2026/03/105171. Retrospectively registered (enrolment began prior to registration).