Objective: To evaluate the value of the Neutrophil Percentage-to-Albumin Ratio (NPAR) in differentiating disease severity and predicting major adverse clinical outcomes in patients with sepsis and septic shock, and to compare the performance of three machine learning models (logistic regression, Random Forest, and XGBoost) incorporating NPAR with routine laboratory parameters.
Methods: A retrospective cross-sectional study was conducted on 260 adult patients admitted to the Intensive Care Unit of Thong Nhat Hospital between June 2025 and May 2026, including 127 patients with septic shock and 133 with sepsis diagnosed according to the Sepsis-3 criteria. NPAR was calculated as NEU (%) × 100 / Albumin (g/L). Three machine learning models were developed using nine input variables (NPAR, albumin, neutrophil percentage, C-reactive protein, procalcitonin, platelet count, age, creatinine, and lactate). Model performance was evaluated using stratified 5-fold cross-validation, and 95% confidence intervals were estimated by 500 bootstrap resamples.
Results: The mean NPAR was 3.12 ± 0.88, and was significantly higher in patients with septic shock than in those with sepsis without shock (3.21 vs. 3.03, p = 0.032). Major adverse clinical outcomes occurred more frequently in the septic shock group (34.7% vs. 21.4%, p = 0.026). NPAR showed a strong negative correlation with serum albumin (r = −0.780, p < 0.001) and a strong positive correlation with neutrophil percentage (r = 0.608, p < 0.001). As a single predictor, NPAR yielded an AUC of 0.421. Logistic regression demonstrated the highest predictive performance (AUC = 0.534, 95% CI: 0.44–0.62), followed by Random Forest (AUC = 0.454, 95% CI: 0.36–0.54) and XGBoost (AUC = 0.435, 95% CI: 0.35–0.52). In the XGBoost model, NPAR accounted for 11.4% of feature importance, ranking third after procalcitonin and platelet count.
Conclusions: NPAR reflects the combined effects of systemic inflammation and nutritional status in patients with sepsis. Although significantly associated with septic shock, its predictive value for adverse clinical outcomes was limited when used alone. Incorporating NPAR into machine learning models did not substantially improve predictive performance compared with conventional logistic regression in this study.
VALUE OF THE NPAR INDEX IN DIFFERENTIATING SEPTIC SHOCK FROM SEPSIS IN INTENSIVE CARE UNIT PATINET AT THONG NHAT HOSPITAL: A PRELIMINARY APPLICATION OF MACHINE LEARNING MODELS (RANDOM FOREST, GRADIENT BOOSTING)
0
Views
0
Downloads
Abstract
Keywords
Neutrophil percentage-to-albumin ratio; NPAR; sepsis; septic shock; machine learning; Random Forest; XGBoost.
References
[1]
[1] Rudd KE, Johnson SC, Agesa KM, et al. Global, regional, and national sepsis incidence and mortality, 1990-2017. Lancet, 2020, 395(10219): 200-211.
Google Scholar
[2]
[2] Pierrakos C, Vincent JL. Sepsis biomarkers: a review. Crit Care, 2010, 14(1): R15.
Google Scholar
[3]
[3] Lam SW, Bauer SR, Fowler R, Duggal A. Systematic review and meta-analysis of procalcitonin-guidance versus usual care in critically ill patients. Crit Care Med, 2018, 46(5): 684-690.
Google Scholar
[4]
[4] Sun X, Luo T, Wang L, et al. NPAR as a prognostic biomarker in solid tumors: a systematic review and meta-analysis. Cancer Manag Res, 2022, 14: 2709-2720.
Google Scholar
[5]
[5] Breiman L. Random forests. Machine Learning, 2001, 45(1): 5-32.
Google Scholar
[6]
[6] Chen T, Guestrin C. XGBoost: A scalable tree boosting system. Proceedings of KDD, 2016: 785-794.
Google Scholar
[7]
[7] Rajpurkar P, Chen E, Banerjee O, Topol EJ. AI in health and medicine. Nat Med, 2022, 28(1): 31-38.
Google Scholar
[8]
[8] Singer M, Deutschman CS, Seymour CW, et al. The Third International Consensus Definitions for Sepsis and Septic Shock (Sepsis-3). JAMA, 2016, 315(8): 801-810.
Google Scholar
[9]
[9] Vincent JL, Dubois MJ, Navickis RJ, Wilkes MM. Hypoalbuminemia in acute illness: meta-analysis of cohort studies and controlled trials. Ann Surg, 2003, 237(3): 319-334.
Google Scholar
[10]
[10] Xue M, Zhu G, Pan S, Chen C, Hu B. Neutrophil-to-albumin ratio as a predictor of 28-day mortality in sepsis. Front Med (Lausanne), 2022, 9: 861101.
Google Scholar
[11]
[11] Hastie T, Tibshirani R, Friedman J. The Elements of Statistical Learning. 2nd ed. Springer, 2009.
Google Scholar
[12]
[12] Nemati S, Holder A, Razmi F, Stanley MD, Clifford GD, Buchman TG. An interpretable machine learning model for accurate prediction of sepsis in the ICU. Crit Care Med, 2018, 46(4): 547-553.
Google Scholar
[13]
[13] Singer E, Elger BS, Madai VI. Machine learning-based mortality prediction in sepsis using routine clinical laboratory data. J Crit Care, 2023, 75: 154276.
Google Scholar
[14]
[14] Wiedermann CJ, Wiedermann W, Joannidis M. Hypoalbuminemia and acute kidney injury: a meta-analysis of observational clinical studies. Intensive Care Med, 2010, 36(9): 1657-1665.
Google Scholar
[15]
[15] Lundberg SM, Lee SI. A unified approach to interpreting model predictions. Adv Neural Inf Process Syst, 2017, 30: 4765-4774.
Google Scholar