Abstract: Objective To explore the risk factors of refeeding syndrome (RFS) in the elderly patients with severe stroke, and to construct a risk prediction model to guide clinical assessment. Methods Convenience sampling was used to enroll the elderly patients with severe stroke admitted to Nanjing First Hospital from August 2023 to September 2024 in the study. The participants were divided into RFS group and non-RFS group based on RFS occurrence, and their general information, and the scores of nutrition risk screening (NRS) 2002, enteral nutrition (EN) tolerance, and National Institutes of Health Stroke Scale (NIHSS)were collected.Multiple logistic regression analysis was used to investigate the influencing factors of RFS, and a prediction model for RFS was developed.The performance of the model was evaluated by receiver operating characteristic (ROC) curve, calibration curve and Hosmer-Lemeshow (HL) fit goodness test. Results A total of 253 patients were enrolled, and 54(21.34%) of them presented with RFS. Protein supplementation during refeeding, pre-refeeding serum level of albumin, NRS2002 score, EN tolerance score, and BMI<18.5 were independent influencing factors for RFS in the elderly patients with severe stroke (P<0.05). The area under the ROC curve of the predictive model was 0.897 (95% CI :0.835－0.958), with a maximum Youden index of 0.714, sensitivity of 0.940, and specificity of 0.774. The HL test showed a high degree of model fit, and the calibration curve showed good consistency. Conclusions The model shows good fit and high predictive performance. Clinical nursing staff should focus on elderly patients with BMI<18.5, additional protein supplementation during refeeding, low pre-refeeding serum albumin concentration and elevated NRS2002 score or EN tolerance score.

Key words: severe illness, stroke, aged, refeeding syndrome, risk prediction model