Transformer-based neural network for jointly predicting length of stay and critical interventions for patients

Abstract

Currently systems for Length of Stay (LoS) and clinical interventions for patients work independent of each other. However, they are highly interdependent decisions for overall medical predictions for patients. Embodiments disclosed provide a method and system for transformer-based Neural Network (NN) for jointly predicting LoS and critical interventions for patients admitted to medical facilities. A joint NN model, comprising Bidirectional Encoder Representations from Transformers (BERT) model as one of the layers, processes first day clinical notes, available in an unstructured data format, and a plurality of medical attributes of the patient available in a structured data. Further, the joint NN model jointly predicts s (a) the LoS of the patient into one of the classes comprising LONG and SHORT and (b) a Type of Intervention (ToI) for the patient into one among a plurality of classes with each class of the ToI comprising a list of critical clinical interventions.