The association of systemic inflammatory indices with all-cause mortality risks in patients with COPD: A cohort study based on machine learning.

This study aimed to comprehensively assess the prognostic value of routinely obtained blood-based systemic inflammatory indices in predicting all-cause mortality among individuals with chronic obstructive pulmonary disease (COPD). This retrospective cohort study analyzed data from the National Health and Nutrition Examination Survey (NHANES) cycles 2007-2010. A total of 1109 eligible adults with COPD were included, with 333 deaths recorded during the follow-up period. Eleven systemic inflammatory indices were derived from baseline hematological parameters. The associations between these indices and all-cause mortality were initially evaluated using multivariate Cox proportional hazards models. To manage high-dimensional data and identify complex patterns not captured by conventional statistical methods, machine learning (ML) algorithms were applied for feature selection, model development, and performance evaluation. Model discrimination and clinical utility were assessed using the area under the receiver operating characteristic curve (AUC), calibration plots, and decision curve analysis. Among the 1109 participants (mean age 57.9 ± 15.5 years, 52.2% male), non-survivors (n = 333) were significantly older and had a higher baseline burden of comorbidities. After full adjustment for covariates, several inflammatory indices showed statistically significant associations with all-cause mortality. The neutrophil percentage-to-albumin ratio (NPAR) and neutrophil-to-lymphocyte ratio exhibited the strongest associations, with HRs of 2.46 (95% CI: 1.64-3.69) and 2.14 (95% CI: 1.42-3.22), respectively, in the highest quartile (Q4) compared to the lowest (Q1). The neutrophil-to- high-density lipoprotein ratio also demonstrated a significant positive association (HR for Q4 vs Q1: 1.79, 95% CI: 1.18-2.70). In contrast, higher levels of the C-reactive protein-albumin-lymphocyte index index were associated with reduced risk, indicating a protective effect (HR for Q4 vs Q1: 0.49, 95% CI: 0.33-0.72). The ML-derived NPARTEST model, based on the NPAR index, achieved an AUC of 0.828 for predicting all-cause mortality, demonstrating good discriminative performance and clinical utility. Systemic inflammatory indices, particularly the NPAR and neutrophil-to- high-density lipoprotein ratio, are independently associated with all-cause mortality in patients with COPD, often exhibiting nonlinear relationships. The ML-based NPARTEST model demonstrates promising predictive performance. These findings underscore the potential of cost-effective, routinely measured blood-based biomarkers to enhance risk stratification in COPD management. External validation in diverse populations is warranted to confirm the generalizability of these results.