Intelligent alarms integrated in a multi-agent architecture for diabetes management

Resumen

This paper describes the development of an intelligent agent that interprets blood glucose monitoring data received by a telemedicine-based diabetes management service. The agent generates automatic alarms when some deviations in the patient status are detected. The agent combines different methods to produce data summaries and automatic alarms based on statistics, rule-based techniques and model-based techniques. The rule-based analysis allows the detection of severe abnormalities using different time scales depending on the quality of the received information. The model-based analysis uses a physiological qualitative model implemented with a causal probabilistic network that detects deviations in the 'insulin effectiveness' along days. The KM ( Knowledge Management) agent was tested with data from 11 patients with diabetes that used a telemedicine service during 1 year. The KM agent detected anomalous situations in the 100% of the cases where a therapy modification was decided by the healthcare professional; the agent detected abnormal data in 37% of transmissions, being able to decrease professionals' workload due to telemedicine. The advantages of an automatic response integrated in a telemedicine service are that it focuses doctors' and patients' attention on abnormal data and gives instantaneous feedback to patients reinforcing the education and the motivation aspects of the therapy.