Early prediction of functional impairment at hospital discharge in patients with osteoporotic vertebral fracture: a machine learning approach

Abstract

Although conservative treatment is commonly used for osteoporotic vertebral fracture (OVF), some patients experience functional disability following OVF. This study aimed to develop prediction models for new-onset functional impairment following admission for OVF using machine learning approaches and compare their performance. Our study consisted of patients aged 65 years or older admitted for OVF using a large hospital-based database between April 2014 and December 2021. As the primary outcome, we defined new-onset functional impairment as a Barthel Index ≤ 60 at discharge. In the training dataset, we developed three machine learning models (random forest [RF], gradient-boosting decision tree [GBDT], and deep neural network [DNN]) and one conventional model (logistic regression [LR]). In the test dataset, we compared the predictive performance of these models. A total of 31, 306 patients were identified as the study cohort. In the test dataset, all models showed good discriminatory ability, with an area under the curve (AUC) greater than 0.7. GBDT (AUC = 0.761) outperformed LR (0.756), followed by DNN (0.755), and RF (0.753). We successfully developed prediction models for new-onset functional impairment following admission for OVF. Our findings will contribute to effective treatment planning in this era of increasing prevalence of OVF.