Abstract: BACKGROUND: Prevention of fractures after stroke is very important, and there are currently no models to predict the occurrence of hip fractures after stroke.
OBJECTIVE: To investigate the risk factors leading to hip fracture in stroke patients and to establish a risk prediction model to visualize this risk. 
METHODS: A total of 439 stroke patients were selected from the Affiliated Hospital of Xuzhou Medical University from June 2014 to June 2017, including 107 males and 332 females, with a mean age of (71.38±9.74) years. They were divided into fracture group (n=35) and non-fracture group (n=404) according to the presence or absence of hip fracture. Univariate and multivariate analyses were used to determine the risk factors for hip fracture after stroke. The data were randomly divided into training set (70%) and test set (30%). Nomogram predicting the risk of hip fracture occurrence was created based on the results of the multifactor analysis, and performance was evaluated using receiver operating characteristic curve, calibration curves, and decision curve analysis. A web calculator was created to facilitate a more convenient interactive experience for clinicians. 
RESULTS AND CONCLUSION: (1) Univariate analysis showed significant differences between the two groups in the number of falls, smoking, hypertension, glucocorticoids, number of strokes, Mini-Mental State Examination, visual acuity level, National Institute of Health Stroke Scale, Berg Balance Scale, and Stop Walking When Talking scale scores (P < 0.05). (2) Multivariate analysis showed that number of falls [OR=17.104, 95%CI (3.727-78.489), P=0.000], National Institute of Health Stroke Scale [OR=1.565, 95%CI(1.193-2.052), P=0.001], Stop Walking When Talking [OR=12.080, 95%CI(2.398-60.851), P=0.003] were independent risk factors positively associated with new hip fractures. Bone mineral density [OR=0.155, 95%CI(0.044-0.546), P=0.012] and Berg Balance Scale [OR=0.840, 95%CI(0.739-0.954), P=0.007] were negatively associated with new hip fractures after stroke. (3) The AUC values of nomogram were 0.956 and 0.907 in the training and test sets, respectively, and the calibration curves showed a high agreement between predicted and actual status with an area under the decision curve of 0.038 and 0.030, respectively. (4) These findings conclude that the number of falls, low bone mineral density, low Berg Balance Scale score, high National Institute of Health Stroke Scale score, and positive Stop Walking When Talking are risk factors for hip fracture after stroke. Based on this, a nomogram with high accuracy was developed and a web calculator (https://stroke.shinyapps.io/DynNomapp/) was created. 

Key words: stroke, osteoporosis, hip fracture, fall risk, nomogram, prediction model