Machine learning based association between inflammation indicators (NLR, PLR, NPAR, SII, SIRI, and AISI) and all-cause mortality in arthritis patients with hypertension: NHANES 1999–2018

Background This study aimed to evaluate the relationship between CBC-derived inflammatory markers (NLR, PLR, NPAR, SII, SIRI, and AISI) and all-cause mortality (ACM) risk in arthritis (AR) patients with hypertensive (HTN) using data from the NHANES. Methods We employed weighted multivariable logistic regression and WQS regression to explore the relationship between inflammatory markers and ACM in AR patients, as well as to determine the weights of different markers. Kaplan–Meier curves, restricted cubic splines (RCS) and ROC curves were utilized to monitor cumulative survival differences, non-linear relationships and diagnostic utility of the markers for ACM risk, respectively. Key markers were selected using XGBoost and LASSO regression machine learning methods, and a nomogram prognostic model was constructed and evaluated through calibration curves and decision curve analysis (DCA). Results The study included 4,058 AR patients with HTN, with 1,064 deaths over a median 89-month follow-up. All six inflammatory markers were significantly higher in the deceased group ( p < 0.001). Weighted multivariable logistic regression showed these markers’ elevated levels significantly correlated with increased ACM risk in hypertensive AR patients across all models ( p < 0.001). Kaplan–Meier analysis linked higher marker scores to lower survival rates in AR patients with HTN ( p < 0.001). WQS models found a positive correlation between the markers and hypertensive AR patients ( p < 0.001), with NPAR having the greatest impact (70.02%) and SIRI next (29.01%). ROC analysis showed SIRI had the highest AUC (0.624) for ACM risk prediction, closely followed by NPAR (AUC = 0.618). XGBoost and LASSO regression identified NPAR and SIRI as the most influential markers, with higher LASSO-based risk scores correlating to increased mortality risk (HR, 2.07; 95% CI, 1.83–2.35; p < 0.01). RCS models revealed non-linear correlations between NPAR (Pnon-linear<0.01) and SIRI (Pnon-linear<0.01) with ACM risk, showing a sharp mortality risk increase when NPAR >148.56 and SIRI >1.51. A prognostic model using NPAR and SIRI optimally predicted overall survival. Conclusion These results underscore the necessity of monitoring and managing NPAR and SIRI indicators in clinical settings for AR patients with HTN, potentially improving patient survival outcomes.