ViSiBiD: A learning model for early discovery and real-time prediction of severe clinical events using vital signs as big data

The advance in wearable and wireless sensors technology have made it possible to monitor multiple vital signs (e.g. heart rate, blood pressure) of a patient anytime, anywhere. Vital signs are an essential part of daily monitoring and disease prevention. When multiple vital sign data from many patients are accumulated for a long period they evolve into big data. The objective of this study is to build a prognostic model, ViSiBiD, that can accurately identify dangerous clinical events of a home-monitoring patient in advance using knowledge learned from the patterns of multiple vital signs from a large number of similar patients. We developed an innovative technique that amalgamates existing data mining methods with smartly extracted features from vital sign correlations, and demonstrated its effectiveness on cloud platforms through comparative evaluations that showed its potential to become a new tool for predictive healthcare. Four clinical events are identified from 4893 patient records in publicly available databases where six bio-signals deviate from normality and different features are extracted prior to 1–2 h from 10 to 30 min observed data of those events. Known data mining algorithms along with some MapReduce implementations have been used for learning on a cloud platform. The best accuracy (95.85%) was obtained through a Random Forest classifier using all features. The encouraging learning performance using hybrid feature space proves the existence of discriminatory patterns in vital sign big data can identify severe clinical danger well ahead of time.