Deepsigns: a predictive model based on deep learning for the early detection of patient health deterioration

Abstract

CONTEXT: The accurate and early diagnosis of critically ill patients depends on medical staff’s attention and the observation of different variables, vital signs, and laboratory test results, among others. Seriously ill patients usually have changes in their vital signs before worsening. Monitoring these changes is essential to anticipate the diagnosis in order to initiate patients’ care. Prognostic indexes play a fundamental role in this context since they allow us to estimate the patients’ health status. Besides, Electronic Health Records’ adoption improved data availability, which can be processed by machine learning techniques for information extraction to support clinical decisions. The volume and variety of data stored in the EHR make it possible to carry out more accurate analyzes that allow different types of health care assessments. Nevertheless, as the amount of data available is vast and complex, there is a need for new methods to analyze that data to explore significant patterns. The use of Machine Learning (ML) techniques to generate knowledge, search for information patterns, and support clinical decisions is one of the possibilities to address this problem. OBJECTIVE: this work aims to create a computational model able to predict the deterioration of patients’ health status in such a way that it is possible to start the appropriate treatment as soon as possible. The model was developed based on the Deep Learning technique, a Recurrent Neural Networks, the Long Short-Term Memory, to predict patient’s vital signs and subsequent evaluation of the patient’s health status severity through Prognostic Indexes commonly used in the Health area. METHOD: The methodology of this work consists of the following steps carried out in sequence. The definition of the data source to be used in the creation of the model and the selection of the data, the pre-processing to create a database for the development of the model, the definition of the implementation of the model and its evaluation through comparison with other models. RESULTS: Experiments showed that it is possible to predict vital signs with good precision (accuracy > 80%) and, consequently, predict the Prognostic Indexes in advance to treat the patients before deterioration. Predicting the patient’s vital signs for the future and use them for the Prognostic Index’ calculation allows clinical times to predict future severe diagnoses that would not be possible applying the current patient’s vital signs (50% - 60% of cases would not be identified). CONCLUSION: This work’s main scientific contribution is the creation of a method for predicting vital signs based on historical data with low Mean Squared Error and its following application in the calculation of prognostic indexes with effectiveness (50% - 60% of cases that would not be identified as severe). The differential presented by this proposal stems from the fact that few works predict vital signs. Most of the works focus on predicting specific health outcomes, such as specific diagnoses, considering the current vital signs. In this work, the proposal is to predict the evolution of vital signs in the future and use these predicted signs to calculate prognostic indexes.