Dynamic Risk Assessment Enabling Automated Interventions for Medical Cyber-Physical Systems

As in many embedded systems domains, in modern healthcare we experience increasing adoption of (medical) cyber-physical systems of systems. In hospitals, for instance, different types of medical systems are integrated dynamically to render higher-level services in cooperation. One important task is the realization of smart alarms as well as, in a second step, the realization of automated interventions, such as the administration of specific drugs. A fundamental correlated problem is insufficient risk awareness, which are caused by fluctuating context conditions, insufficient context awareness, and a lack of reasoning capabilities to deduce the current risk. A potential solution to this problem is to make systems context- and risk-aware by introducing a runtime risk assessment approach. In this paper, we introduce such an approach for a wider identification of relevant risk parameters and risk assessment model building based on Bayesian Networks (BN). This model considers not only changes in the actual health status of the patient but also the changing capabilities to detect and react according to this status. This includes changing capabilities due to adding or removing different types of sensors (e.g. heart rate sensors) and replacing sensors of the same type but with other integrity level. In addition, we present an evaluation of the approach based on a simulated clinical environment for patient-controlled analgesia.