Using Nursing Notes to Predict Length of Stay in ICU for Critically Ill Patients

Abstract

Managing resource critical facilities like Intensive Care Units (ICUs) is an important task for hospital management officials. Predicting how long a patient is going to stay in ICU is considered an important problem for managers. Several attempts have been made to solve this problem using different types of clinical data that are available from the past. While a number of studies have deployed classification models that use structured clinical variables, recent advances in Natural Language Processing models have opened up the possibilities of using unstructured text data like nursing notes, discharge summaries, etc. for prediction. In this work, we have proposed the use of CNN and LSTM based prediction networks along with transformer-based language models for representing the notes data. The proposed model can predict with a much higher accuracy rate than any other existing model. The dataset used for the experiment is MIMIC, which is an anonymized dataset that contains detailed records of around 40,000 patients most of whom were critically ill. We use the first day’s nursing notes for prediction since that can provide most relevant and valuable input to planning.