LAUSR

We previously found that neurocritically ill patients are prone to refeeding syndrome (RFS), a potentially life-threatening complication. However, there is no unified or validated consensus on the screening tool for RFS so far. We aimed to validate and compare the performance of four screening tools for RFS in neurocritically ill patients. We conducted a single-center, observational, retrospective cohort study among neurocritically ill adult patients who were admitted to the neurocritical care unit (NCU), and who received enteral nutrition for 72 h or longer. They were scored on the Short Nutritional Assessment Questionnaire (SNAQ), the Global Leadership Initiative on Malnutrition (GLIM), the modified criteria of the Britain’s National Institute for Health and Care Excellence (mNICE), and ASPEN Consensus Recommendations for Refeeding Syndrome (ASPEN) scales to predict RFS risk via admission data. The performance of each scale in predicting RFS was evaluated. Logistic regression analysis was used to identify the independent risk factors for RFS, and they were added to the above scales to strengthen the identification of RFS. Of the 478 patients included, 84 (17.57%) developed RFS. The sensitivity of the SNAQ and GLIM was only 20.2% (12.6–30.7%), although they had excellent specificities of 84.8% (80.8–88.1%) and 86.0% (82.1–89.2%), respectively; mNICE predicted RFS with a sensitivity of 48.8% (37.8–59.9%) and a specificity of 65.0% (60.0–69.9%); ASPEN had the highest Youden index, with a sensitivity and specificity of 53.6% (42.4–64.4%) and 64.7% (59.8–69.4%), respectively. The Area Under the receiver operating characteristic Curves (AUC) of SNAQ, GLIM, mNICE, and ASPEN to predict RFS were 0.516 (0.470–0.561), 0.533 (0.487–0.579), 0.568 (0.522–0.613), and 0.597 (0.551–0.641), respectively. We identified age, Acute Physiology and Chronic Health Evaluation II (APACHE II), Sequential Organ Failure Assessment (SOFA), and Glasgow Coma Scale (GCS) score as independent risk factors of RFS, and the combination of GCS and age can improve the AUC of ASPEN to 0.664 (0.620–0.706) for predicting RFS. SNAQ, GLIM, mNICE, and ASPEN do not perform well in identifying neurocritically ill patients at high risk of RFS, although ASPEN appears to have relatively a good validity among them. Combining GCS and age with ASPEN slightly improves RFS recognition, but it still leaves a lot of room for improvement.