Aim: Improve the prognostic prediction of clinical variables for non-small cell lung cancer (NSCLC), by selecting from blood-biomarkers, non-invasively describing hypoxia, inflammation and tumour load.
Methods: Model development and validation included 182 and 181 inoperable stage I-IIIB NSCLC patients treated radically with radiotherapy (55.2%) or chemo-radiotherapy (44.8%). Least absolute shrinkage and selection operator (LASSO), selected from blood-biomarkers related to hypoxia [osteopontin (OPN) and carbonic anhydrase IX (CA-IX)], inflammation [interleukin-6 (IL-6), IL-8, and C-reactive protein (CRP)], and tumour load [carcinoembryonic antigen (CEA), and cytokeratin fragment 21-1 (Cyfra 21-1)]. Sequent model extension selected from alpha-2-macroglobulin (α2M), serum interleukin-2 receptor (sIL2r), toll-like receptor 4 (TLR4), and vascular endothelial growth factor (VEGF). Discrimination was reported by concordance-index.
Results: OPN and Cyfra 21-1 (hazard ratios of 3.3 and 1.7) significantly improved a clinical model comprising gender, World Health Organization performance-status, forced expiratory volume in 1 second, number of positive lymph node stations, and gross tumour volume, from a concordance-index of 0.66 to 0.70 (validation = 0.62 and 0.66). Extension of the validated model yielded a concordance-index of 0.67, including α2M, sIL2r and VEGF (hazard ratios of 4.6, 3.1, and 1.4).
Conclusion: Improvement of a clinical model including hypoxia and tumour load blood-biomarkers was validated. New immunological markers were associated with overall survival. Data and models can be found at www.cancerdata.org (DOI:10.17195/candat.2016.04.1) and www.predictcancer.org.