Construction and effect evaluation of a prediction model for malnutrition risk in patients recovering from stroke

The incidence of malnutrition in stroke patients during recovery is high, which seriously affects the rehabilitation outcome. Existing general nutrition screening tools have limited predictive power for this specific population.

To construct and validate a prediction model specifically for assessing the risk of malnutrition in patients with stroke during recovery.

A total of 262 patients with stroke in recovery stage were retrospectively enrolled and divided into training set (n = 176) and validation set (n = 86) according to admission time. Nutritional risk was defined as nutritional risk screening 2002 (NRS-2002) score ≥3. Clinical and laboratory indicators were collected, and Logistic regression analysis was used to screen independent predictors and construct a nomogram model. The discrimination and calibration of the model were evaluated by receiver operating characteristic (ROC) curve and calibration curve, and Bootstrap internal verification and independent temporal validation were performed.

Multivariate analysis identified age, modified Barthel index (MBI) score and albumin level as independent predictors of malnutrition. The AUC of the nomogram model was 0.869 (95%CI: 0.817-0.922) in the training cohort and 0.878 (95%CI: 0.808-0.949) in the validation cohort. The calibration curve showed good agreement between the predicted risk and the actual risk.

This study successfully constructed and verified a risk prediction model for malnutrition in stroke recovery patients including age, functional status and albumin level. As a complementary tool to the existing screening method NRS-2002, the model has good predictive power and clinical applicability. It facilitates early identification of high-risk patients and provides an evidence-based approach to individualized nutritional intervention.