Um modelo proativo de antecipação de ações de times de resposta rápida baseado em análise preditiva

Abstract

The mobile and ubiquitous computing has allowed the emergence of solutions that enable real-time monitoring of signals coming from sensors and processing for applications that can perform actions according to the conditions found. This feature enables the use of this technology for monitoring health conditions of patients, called ubiquitous healthcare. In several situations, in order to save his lives, it is necessary to analyze the vital signs of patients to prevent any collapses. This work is part of these conditions and is aimed at anticipating the actions of rapid response teams based on predictive analysis, proposing the Predictvs model. A Rapid Response Team intends to prevent deaths in patients who have clinical deterioration outside of intensive care units in hospitals environments. Differently of related works, which are concerned only with intensive care environments, the Predictvs model seeks to anticipate the actions of teams of rapid response through the analysis of vital signs of patients with the use of early warning scores and linear regression. The scientific contribution of the presented model is that we could better predict possible collapse situations of patients, through the monitoring and analysis of vital signs. The Predictvs evaluation was performed with the use of scenarios, implementation of a prototype and several simulations. Analyzes performed with about 228,000 measurements from a public dataset showed good results, where the accuracy of the prediction for the next measurement was very high, reaching more than 99% in the case of heart rate and 100% in arterial oxygen saturation, surpassing 95% in other vital signs. In addition, the false negative index was considerably lower, reaching less than 1% in heart rate and arterial oxygen saturation. The rate of false positives was also low, although not so much as that of false negatives. However, predictions for three or more future measurements show a drop in accuracy (even showing relatively expressive set values with several physiological signals above 98%) and an increase in the number of false negatives and, mainly, false positives.