UC Davis

AIM: Pre-injury frailty has been investigated as a tool to predict outcomes of older trauma patients. Using artificial intelligence principles of machine learning, we aimed to identify a signature (combination of clinical variables) that could predict which older adults are at risk of fall-related hospital admission. We hypothesized that frailty, measured using the 5-item modified Frailty Index, could be utilized in combination with other factors as a predictor of admission for fall-related injuries. METHODS: The National Readmission Database was mined to identify factors associated with admission of older adults for fall-related injuries. Older adults admitted for trauma-related injuries from 2010 to 2014 were included. Age, sex, number of chronic conditions and past fall-related admission, comorbidities, 5-item modified Frailty Index, and medical insurance status were included in the analysis. Two machine learning models were selected among six tested models (logistic regression and random forest). Using a decision tree as a surrogate model for random forest, we extracted high-risk combinations of factors associated with admission for fall-related injury. RESULTS: Our approach yielded 18 models. Being a woman was one of the factors most often associated with admission for fall-related injuries. Frailty appeared in four of the 18 combinations. Being a woman, aged 65-74 years and presenting a 5-item modified Frailty Index score >3 predicted admission for fall-related injuries in 80.3% of this population. CONCLUSION: Using artificial intelligence principles of machine learning, we were able to develop 18 signatures allowing us to identify older adults at risk of admission for fall-related injuries. Future studies using other databases, such as TQIP, are warranted to validate our high-risk combination models. Geriatr Gerontol Int 2025; 25: 232-242.