Blood biomarker-aided cancer discrimination of screen-detected small lung nodule: The Cancer Signature Ensemble

<bold>Background:</bold> Low-dose chest computed tomography (LDCT) is the corner stone of lung cancer screening. But, its high false-positive rate is the major limitation, which may lead unnecessary invasive diagnostic procedures. The aim of this study is to develop blood biomarker index (Cancer Signature Ensemble, CSE) supporting cancer discrimination for screen-detected small lung nodules. <bold>Methods:</bold> Inclusion criteria were 1) aged 50-80, 2) smoking history ≥20 pack years (current or former smoker quitting smoking less than 15 years), 3) LDCT screen detected nodules categorized to Lung-RADS 4B or 4X and expected clinical stage IA if lung cancer would be diagnosed. Combined assessment of untargeted whole-genome methylation patterns, DNA fragmentation, and copy number variation (CSE) in cfDNA from plasma was performed with machine learning. Cancer discriminating performances of clinical models (Mayo and Brock model), CSE+clinical models (CSE-Plus) were compared. <bold>Results:</bold>. A total of 184 subjects were analyzed (male: female 174:10). 116 patients with lung cancer confirmed by biopsy and 68 patients with benign nodules (24 confirmed by biopsy, 44 of clinical and radiological benignancy criteria) were enrolled. Mean age was 66.9 ± 7.7, solid vs. part-solid nodule 163: 21, Lung-RADS 4B vs. 4X nodule 133: 51, mean diameter was 18.6 ± 5.2mm. AUC of Brock model and CSE-Plus was 0.72 and 0.87(95% CI: 0.81-0.92). AUC of Mayo model and CSE-Plus was 0.71 and 0.86 (95% CI: 0.80-0.91). This is an interim analysis, and the enrollment is on-going. <bold>Conclusion:</bold> Combination of clinical model and biomarker (CSE-Plus) improved cancer discriminating performance for screen-detected small lung nodules.