Post-surgical fall risk prediction: a machine learning approach for spine and lower extremity procedures

Abstract

In Taiwan, two key indicators of clinical care quality are pressure injuries and falls. Falls can have significant physical impacts, ranging from minor injuries like bruises to major injuries such as fractures, sprains, and severe head trauma. To assess fall risk early and implement preventive measures, this study analyzed 2,948 medical records of patients who underwent spinal and lower limb surgeries at the Veterans General Hospital in Taichung, Taiwan. Data collected included patient demographics, vital signs, health conditions, diagnoses, and medications, as well as information on their admission type and any recorded falls, to identify factors contributing to inpatient falls and to establish early warning measures. This study accounted for patients' history of falls during model training, followed by variable selection and outcome modeling using logistic regression and random forest methods. Results showed that logistic regression with fall history as part of the training data is an effective approach. Patients admitted by wheelchair or stretcher for spine or lower limb surgeries had an increased fall risk. Each additional year of age also increased fall risk. In patients with arthritis, the odds of falling decreased. Conversely, the use of psychotropic and antihypertensive drugs raised fall risk. While sleeping pills reduced it. Each degree increase in body temperature and poor vision were also associated with higher fall odds. These findings support improvements in patient care quality and help reduce caregiver workload by refining fall risk assessment processes.

Keywords: classification problems; falls; health care; logistic regression; random forests.