Analysing data streams of vital signs has been a popular topic in research communities with techniques mainly focusing on detection, classification and prediction. One drawback for data classification/prediction is that the data mining model is built based on a full set of stationary data. Updating the model for sustaining the classification accuracy often needs the whole dataset including the evolving data to be accessed. This nature of model rebuilding dampers the possibility of mining vital signs in real-time and at high speed. Unfortunately, much of the past papers in the literature were based on traditional data mining models. In this paper, a data stream mining model which is flexible in configuring with different incremental data stream learning methods is tested as a real-time classification engine for mining vital data streams. A computer simulation experiment is conducted that is based on a case study of adult respiratory distress syndrome under twelve-hours of ICU treatment. The results indicate promising possibilities of performing real-time prediction by the proposed model.