A nomogram for predicting the risk of malnutrition in hospitalized older adults: a retrospective study.

Background: Malnutrition is highly prevalent but under-recognized in hospitalized older adults, which is closely related to increased risk of adverse clinical outcomes and mortality. It is crucial to identify high-risk individuals at an early stage and manage them promptly. This study aimed to explore the predictive factors and develop a nomogram model for predicting the risk of malnutrition in hospitalized elderly patients.

Methods: We conducted a retrospective study of data collected from 456 older individuals admitted to geriatric wards from four hospitals in China between August 2020 and December 2020 (136 in the malnutrition group and 320 in the non-malnutrition group). Least Absolute Selection and Shrinkage Operator (LASSO) regression and stepwise multivariate logistic regression were applied to screen predictors and create a nomogram. The predictive performance of the model was assessed by receiver operating characteristic (ROC) curve, concordance index (C-index) and calibration curve. The clinical utility was estimated by decision curve analysis (DCA). Youden's Index was used to identify the optimal cut-point of the nomogram.

Results: Four independent predictive factors were utilized to construct the nomogram model after being selected by LASSO regression and multivariate logistic regression, namely body mass index (BMI), heart failure, frailty and hemoglobin. C-index of the model was 0.906 (95% CI: 0.872-0.939) and the area under the curve (AUC) was 0.906. The optimal cut-point of the nomogram was 82.74 with a sensitivity of 78.7% and specificity of 92.2% (Youden's index: 0.709). The calibration curve demonstrated a high degree of consistency between predicted probability and actual observation. The DCA indicated a favorable clinical benefit for the nomogram.

Conclusions: We have established a multi-dimensional nomogram model to predict the risk of malnutrition in Chinese hospitalized older adults. The model yields favorable predictive performance and clinical utility, which provides an effective approach for rapid identification of high-risk malnourished older individuals in clinical practice.