Development and validation of a novel nutrition-inflammation prognostic score for predicting 30-day mortality in critically ill stroke patients

Objectives Malnutrition and systemic inflammation are common in critically ill stroke patients and contribute to poor outcomes. This study aimed to develop and validate a novel nutrition-inflammation prognostic score to predict critically ill stroke patients 30-day mortality and compare its performance with existing scores. Methods In this retrospective study, a total of 926 critically ill stroke patients were included. The training cohort ( n = 725) was used to develop the prognostic score. Feature selection was performed using three machine learning algorithms: LASSO, SVM-RFE, and Boruta. Four key biomarkers—high-sensitivity C-reactive protein, albumin, neutrophils, and D-Dimer—were identified. Based on these variables, a novel prognostic score, CAND, was constructed, visualized as a nomogram, and deployed as an online calculator. Cox regression analyses assessed the association between CAND defined high-risk groups and 30-day mortality, in comparison with existing nutrition-inflammation scores. The prognostic performance of CAND and these established scores was further evaluated using time-dependent receiver operating characteristic (ROC) curves, concordance index (C-index) and decision curve analysis (DCA). External validation was performed on 201 patients. Results Higher CAND scores were independently associated with increased 30-day mortality risk in both the training cohort [hazard ratio (HR) = 3.273; 95% CI: 2.413–4.437; P < 0.001] and the validation cohort (HR = 3.608; 95% CI: 1.888–6.894, P < 0.001). CAND demonstrated strong discriminative ability and prognostic performance, with a C-index of 0.863 and time-dependent area under the curve(AUC) of 0.727 in the training cohort, and a C-index of 0.831 and AUC of 0.691 in the validation cohort. Compared to existing nutrition-inflammation scores, CAND consistently outperformed them in both cohorts, as further supported by time-dependent ROC and DCA. Conclusions The CAND score, based on four objective biomarkers selected via machine learning, is a reliable and practical tool for early mortality risk stratification in critically ill stroke patients. Its application may inform timely clinical decision-making and targeted nutritional strategies.