The XGBoost model predicted independent ambulation at discharge with an AUC of 0.829, identifying cognitive function and nutritional status as key predictors.
Can an XGBoost model using preoperative clinical information predict independent ambulation at hospital discharge in patients with proximal femoral fractures?
An XGBoost model using preoperative variables like cognitive function, nutritional status, and inflammatory markers can accurately predict independent ambulation at hospital discharge in patients with proximal femoral fractures.
Absolute Event Rate: 0% vs 0%
BACKGROUND: Independent ambulation at hospital discharge is a critical determinant of discharge destination and caregiving burden in older adults with proximal femoral fractures. Preoperative prediction of walking independence may support discharge planning and early intervention. METHODS: This retrospective observational study included 350 patients who underwent surgery for proximal femoral fractures between April 2018-April 2023. Independent ambulation was defined as Functional Ambulation Category (FAC) = 5. Preoperative variables included age, Body Mass Index (BMI), cognitive function (HDS-R), psoas muscle index (PMI), nutritional and inflammatory markers (GNRI, albumin, total protein, CRP), and pre-fracture walking ability. Patients were divided chronologically into a training set (n = 250) and a temporal external validation set (n = 100). An XGBoost model was developed and evaluated using area under the receiver operating characteristic curve (AUC). SHAP analysis was applied to identify major contributing factors. RESULTS: Of the 350 patients, 186 (53.1 %) achieved independent ambulation at discharge. The optimized XGBoost model yielded a mean cross-validation AUC of 0.856 and an external validation AUC of 0.829, with accuracy = 0.771, precision = 0.872, recall = 0.774, and F1-score = 0.820. SHAP analysis identified HDS-R as the strongest contributor, followed by height, pre-fracture walking ability, GNRI, and CRP. CONCLUSIONS: Preoperative prediction of discharge-time walking independence using XGBoost demonstrated high discriminatory performance. Cognitive function, nutritional status, and inflammatory markers were key predictors. SHAP analysis enhanced interpretability, supporting clinical applicability. This model may facilitate individualized discharge planning and early intervention strategies in patients with proximal femoral fractures.
Takahashi et al. (Thu,) reported a other. The XGBoost model predicted independent ambulation at discharge with an AUC of 0.829, identifying cognitive function and nutritional status as key predictors.