E-Health technologies arose as a suitable approach to support diseases diagnostics and treatment decisions, since the Internet of Things devices can monitor humans over a long period. Most of the E-Health technologies are based on machine learning to analyze and classify patients data, returning a possible diagnosis for health professionals as fast and accurate as possible. However, machine learning techniques have high computational complexity, limiting their usage to meet the real time requirements of E-Health systems. Within this context, this paper proposes an E-Health system to analyze and to classify patients data based on data streams, allowing the diagnosis of anomalies in biological exams. The applied data stream approach enables the online training of the classifiers, as well as a suitable performance for data processing. The experiments performed were based on a database of real patients. The results (considering 19 different anomalies) suggest the feasibility of proposed E-Health system, reaching 96%, 94.21%, 92.14% and 92.53% of accuracy, precision, sensitivity, and cover index, respectively, overcoming the existing solutions.