Length of Stay Predictions using Text Notes
Patients with long acute Lengths of Stay (LOS) or delayed discharge, known as Alternate Level of Care (ALC) are known to have adverse effects on patient flow throughout the hospital. With inpatient units operating at or near capacity, the lack of available beds contributes to long waiting times in the emergency department. In addition, prolonged ALC days are associated with worse functional status for the patient at discharge.
Using available inpatient admission notes, a Latent Dirichlet Allocation (LDA) model was used to extract features from raw text data. These features were then used to predict which patients will require long LOS and which patients will require ALC. We then compare the performance of the topic features to typically available structured features at the time of admission.
Early identification of patients requiring long LOS or ALC stays through real time risk stratification could have positive impacts on the patient expectations, better discharge planning, scheduling of hospital resources and improved patient flow.
