An Emergency-deployable Albumin-enhanced NLR Derived by Machine Learning Improves Risk Stratification in Lung Cancer: A Multicenter Cohort Study

Background/Aim: Systemic inflammation is tightly linked to lung cancer prognosis, yet widely used blood-based indices show only modest discrimination. We aimed to develop a simple, albumin-enhanced inflammatory index to improve risk stratification. Patients and Methods: Using the Investigation on Nutrition Status and Clinical Outcome of Common Cancer database, 1,576 patients with lung cancer with complete baseline data were randomly split into a training cohort (n=1,104) and an internal validation cohort (n=472). LASSO regression screened prognostically informative laboratory markers. Conventional inflammatory indices were compared by Harrell’s C-index. A supervised machine-learning approach integrated serum albumin level with the neutrophil-to-lymphocyte ratio (NLR) to derive an albumin-enhanced NLR score (aNLR). Prognostic value was tested with Cox models (three prespecified adjustment levels), restricted cubic splines, Kaplan-Meier analysis, time-dependent area under the receiver operating characteristics curve, calibration, and decision curve analysis. Results: LASSO highlighted lymphocyte count, albumin level, and neutrophil count as dominant factors in predicting prognosis. Among conventional indices, NLR showed the highest discrimination (C-index 0.600). The derived aNLR markedly improved performance (overall C-index 0.727; training 0.725; validation 0.731). Using an outcome-driven cutoff (0.56), high aNLR was consistently associated with worse survival (unadjusted hazard ratio=2.39, 95% confidence interval=2.21-2.58; fully adjusted hazard ratio=2.14, 95% confidence interval=1.97-2.32; both p<0.001). aNLR also improved time-dependent area under the curve and demonstrated favorable clinical utility. Conclusion: An albumin-enhanced NLR, created by machine-learning fusion of albumin and NLR, provides substantially better prognostic discrimination than conventional inflammatory indices and supports individualized survival assessment in lung cancer.