Age-weighted red cell distribution width index for predicting mortality in ICU patients with acute respiratory failure: a retrospective cohort study

Acute respiratory failure (ARF) represents one of the most common critical illnesses in intensive care units (ICUs), with persistently high mortality rates. Red cell distribution width (RDW) has demonstrated value in predicting prognosis of respiratory diseases; however, conventional RDW fails to account for age-related effects on red blood cell morphology. This study aimed to evaluate the prognostic value of the Age-Weighted RDW Index (AWRI) in ICU patients with acute respiratory failure.

This retrospective cohort study utilized data from the eICU Collaborative Research Database, enrolling 24,709 ICU patients with acute respiratory failure after sequential exclusion. AWRI was defined as RDW × (1 + age/100). The primary outcome was in-hospital mortality. Multivariable Cox proportional hazards regression, restricted cubic spline analysis, and five machine learning algorithms with SHAP-based interpretability were applied to evaluate the independent prognostic value and nonlinear risk patterns of AWRI.

Among 24,709 patients (median age 66.0 years; 54.0% male), in-hospital mortality was 18.8%. AWRI was independently associated with mortality in the fully adjusted model (Q4 vs. Q1: HR 1.42, 95% CI 1.28-1.59, P < 0.001). Restricted cubic spline analysis identified a significant nonlinear relationship with two key thresholds: AWRI ≤ 16.3% was associated with markedly reduced mortality risk (HR 0.453, 95% CI 0.281-0.730) and AWRI ≥ 27% with persistently elevated risk (HR 1.31, 95% CI 1.22-1.40). In machine learning analysis, AWRI ranked as the highest-importance predictor among 17 routine admission variables (mean |SHAP| = 0.30), and the XGBoost model achieved an AUC of 0.748, outperforming APACHE IV (AUC = 0.670, P < 0.001). AWRI demonstrated superior discrimination compared with conventional RDW (AUC 0.608 vs. 0.572, P < 0.001).

AWRI independently predicts in-hospital mortality in ICU patients with acute respiratory failure, outperforming both conventional RDW and APACHE IV in discriminative ability and achieving comparable performance to SOFA-2 from a single routine blood count. Its clinically interpretable nonlinear risk thresholds and highest-ranked machine learning importance support integration into routine risk stratification at ICU admission.