Development of a machine learning-based model for predicting the functional outcome of patients with proximal femur fractures

Comprehensive Summary

Nozawa et al. developed a machine learning model for predicting functional outcomes after proximal femur hip fractures. Currently, hip fractures are common in older adults, reducing quality of life and independence, but early rehabilitation efforts can improve recovery, discharge planning, and goal setting. This study created a model to predict patient independence in daily living activity classifying admission data into nine classes of degrees of independence using 2,088 cases from the Japan Association of Rehabilitation Database. Machine learning regression methods generated predictions for the continuous variable of independence, and model performance was determined using prediction accuracy and Quadratic Weighted Kappa (QWK) to be 0.340 and 0.657 respectively. As such, the model is found to be moderately aligned with physician assessments of functional outcomes. From there SHAP (SHapley Additive exPlanations) values were also identified to interpret the model's decision making by narrowing down the input features that most influence the prediction, and top predictors included the total Functional Independence Measure (FIM) score and the level of independence in activities of daily living in patients with dementia. Overall, this model offers a tool backed by data to support clinicians in planning personalized rehabilitation efforts with the potential to assist in discharge destination determination and more. These models can also improve resource allocation in elderly care after being refined to improve its accuracy and generalizability.

Outcomes and Implications

This paper introduces machine learning prediction of independence at discharge after hip fracture, influencing clinical medicine by allowing rehabilitation teams to design earlier and more personalized therapy plans that anticipate complications and create realistic timelines and expectations for families. Additionally, hospitals can now apply variables like FIM scores, age, cognitive impairment and more to reduce prolonged immobility and secondary problems like pressure injuries and thromboembolism. Discharge destination planning and the allocation of inpatient and outpatient resources can become more streamlined and equitable, promoting safer transitions of care and more efficient resource budgeting with orthopedic rehabilitation services.