OrthoMortPred: Predicting one-year mortality following orthopedic hospitalization

Objective: Predicting mortality risk following orthopedic surgery is crucial for informed decision-making and patient care. This study aims to develop and validate a machine learning model for predicting one-year mortality risk after orthopedic hospitalization and to create a personalized risk prediction tool for clinical use. Methods: We analyzed data from 3,132 patients who underwent orthopedic procedures at the Central Lisbon University Hospital Center from 2021 to 2023. Using the LightGBM algorithm, we developed a predictive model incorporating various clinical and administrative variables. We employed SHAP (SHapley Additive exPlanations) values for model interpretation and created a personalized risk prediction tool for individual patient assessment. Results: Our model achieved an accuracy of 93% and an area under the ROC curve of 0.93 for predicting one-year mortality. Notably, 'EMERGENCY ADMISSION DATE TIME' emerged as the most influential predictor, followed by age and pre-operative days. The model demonstrated robust performance across different patient subgroups and outperformed traditional statistical methods. The personalized risk prediction tool provides clinicians with real-time, patient-specific risk assessments and insights into contributing factors. Conclusion: Our study presents a highly accurate model for predicting one-year mortality following orthopedic hospitalization. The significance of 'EMERGENCY ADMISSION DATE TIME' as the primary predictor highlights the importance of admission timing in patient outcomes. The accompanying personalized risk prediction tool offers a practical means of implementing this model in clinical settings, potentially improving risk stratification and patient care in orthopedic practice.