Design and implementation of a deep recurrent model for prediction of readmission in urgent care using electronic health records

There has been a steady growth in machine learning research in healthcare, however, progress is difficult to measure because of the use of different cohorts, task definitions and input variables. To take the advantage of the availability and value of digital health data, we aim to predict unplanned readmissions to the intensive care unit (ICU) from a publicly available Critical Care dataset called
Medical Information Mart for Intensive Care (MIMIC-III). In this research, we formulate a heterogeneous LSTM and CNN architecture specifically to create a model of readmission risk. Our proposed predictive framework outperformed all the benchmark classifiers such as support vector machine, random forest and logistic regression models on all performance measures (AUC, accuracy and precision) except on recall where random forest performed slightly better. Predictions from these models will help in resource planning and decrease mortality or length of stay in clinical care settings.