Elderly inpatient fall risk factors: A study of decision tree and logistic regression

The elderly fall events have become the major public health issue in the world these days. In the United States, falls are the leading cause of accidental death for the elderly. This study applied logistic regression and decision tree to construct inpatient fall risk assessment model in elderly patients. By case-control method, we collected 602 fall and non-fall patients' data, including demographic variables, physiological variables, Barthel indexes, hospitalization risk factors for falls from a regional teaching hospital in northern Taiwan. The result shows six important variables: 1. whether the patient has fallen in the past year 2. cognitive problems, disorientation, irritability during hospitalization 3. movement responses 4. dizziness during hospitalization 5. unsteady gait and use of walking aids during hospitalization and 6. length of stay. For surgical patients, the ways of emergency admission were significantly related to patient fall. A new risk factor - movement response is also significant to predict the elderly inpatient falls. A patient would have a high risk of fall if his/her movement response score is less than six points. The accuracy of fall prediction with training data and validation data are both 70%. The findings provide suggestions for nursing department to identify the high-risk patients and to take preventive measures when they are admitted into hospital.