This paper proposes a Web services-based infrastructure to support clinical decision support systems (CDSS), applied to distributed medical domains within the hospital information system (HIS), to provide clinical support in high-risk medical environments. This work extends an Extensible Markup Language (XML)-based framework for medical data interoperability, and integration of CDSS into the neonatal intensive care unit (NICU), developed previously by the authors.

The CDSS includes artificial neural networks (ANN), case-based reasoning (CBR) tools, and alert detection systems. These services will at first be accessed locally, and eventually from remote locations, via wireless communication by authenticated users.

Since potential performance limitations in providing real-time CDSS output to physicians, using more and more handheld devices, contributes a limiting factor to the system’s successful implementation, a thorough investigation of those possible limitations prior to implementation is necessary. Therefore, software performance engineering (SPE) was applied from the early design stages of the proposed system, in order to ensure that the system would meet its performance requirements, and to identify possible solutions for relieving existing performance limitations of the prototype.

In this paper, only a representative subset of the complete Web services infrastructure is modeled, involving one CDSS as a core Web service, and accessing the patient’s electronic patient record (EPR). In section 1, the authors briefly present the medical, as well as the performance perspective of the proposed systems. The next section reports on the system analysis, considering both functional and nonfunctional requirements. The application of SPE to the proposed systems is presented in section 3, which explains, in detail, the model parameter estimation and the sequence diagram, encompassing the entire functionality for Web service invocation. This diagram is annotated with performance information according to the unified modeling language (UML) profile for schedulability, performance, and time. The resource metrics and architectural design presented in the previous sections are used to build a layered queuing network (LQN) model in section 4, which is to be solved using the analytical solver, LQNS. The results of the performance analysis are presented in the last section, and include the base case results, an identification of software and hardware bottlenecks, and a report on the results obtained by introducing both threads and processors, to mitigate these bottlenecks. This section goes on to include a sensitivity analysis with regard to the Web service processing time, and the number of visits made to the disk by the database task.

Altogether, this paper is a nice case study of applying SPE methods to an important application.