Predicting Length of Stay in Geriatric Patients Using an Ensemble Learning Method

Prolonged hospital stays that exceed the expected duration pose a challenge to patient care and utilization of hospital resources. Leveraging routine clinical data and knowledge about the intensity of medical care, a comprehensive predictive model was developed and validated to assess the risk of prolonged hospitalization. The model, adjusted for patients aged 50 years and older, integrates classification and regression models, applied from the fourth to the fourteenth day of hospitalization to account for updated patient data. Separate predictive models were trained for homogeneous subgroups of patients based on emergency department diagnosis. Moreover, a dashboard was developed to support physicians in clinical practice, facilitating the prompt identification of patients at high risk of delayed discharge. This approach addresses the pressing need to optimize resource allocation and mitigate the risks associated with prolonged hospitalization, particularly among elderly patients.