Abstract
BACKGROUND: Chronic lower respiratory diseases (CLRDs) remain major causes of global mortality. Because conventional inflammatory markers have limited prognostic utility, we developed and validated the relative neutrophil-monocyte-lymphocyte-albumin ratio (NMLAR), defined as (Neutrophil% × Monocyte% × 100)/(Lymphocyte% × Albumin [g/dL]), as a novel biomarker to predict CLRD-specific mortality. METHODS: Immune infiltration of CLRDs was analyzed based on GEO datasets. We then analyzed 9,236 adults with CLRD from NHANES 1999-2014, excluding individuals with missing core variables. Machine learning algorithms (Boruta, SVM-RFE, XGBoost) were applied to identify key predictors. Cox proportional hazards models and restricted cubic spline (RCS) functions were used to evaluate the association between NMLAR and mortality outcomes, and stratified analyses were conducted across clinically relevant subgroups. Model performance was assessed by Harrell's C-index, calibration plots, and decision-curve analysis (DCA). Findings were externally validated in NHANES 2015-2018 (n = 2,107), the MIMIC-IV v3.1 ICU cohort (n = 2,120), and a real-world Zhejiang Provincial ICU cohort (n = 161). RESULTS: Immune profiling showed increased neutrophils/monocytes and reduced lymphocytes in CLRD and acute states. Higher baseline NMLAR was consistently associated with increased risks of both all-cause and CLRD-specific mortality and demonstrated superior predictive performance compared with conventional inflammatory markers. In NHANES, fully adjusted models indicated an approximately linear dose-response, with each 1-unit increment in NMLAR corresponding to a ∼7% higher risk of all-cause mortality and an ∼8% higher risk of CLRD-specific mortality. In the MIMIC cohort, NMLAR remained independently associated with 14-365-day mortality even after adjustment for critical care-specific covariates (SOFA score, CRRT, invasive mechanical ventilation, vasopressor use), with a threshold effect identified at 12.10. In the Zhejiang ICU cohort, NMLAR independently predicted 30-day mortality (HR per unit increase ≈1.09), with a threshold at 13.32. Notably, models derived from NHANES demonstrated moderate discriminatory ability, satisfactory calibration, and clinical net benefit when externally validated in both ICU cohorts, underscoring the robustness and generalizability of NMLAR as a prognostic biomarker across diverse clinical settings. CONCLUSION: NMLAR is a simple, robust, and clinically applicable biomarker for mortality risk in CLRD, demonstrating consistent prognostic value across population-based, critical care, and real-world cohorts.