An Electronic Tool for the Evaluation and Treatment of Sepsis in the ICU: A Randomized Controlled Trial Supplemental Digital Content emethods: Overview of the Sepsis Tool s Development, Architecture, and Interface Electronic Medical Records (EMRs) offer the opportunity to provide computerized assistance to clinicians in patient management. The class of applications which have shown the largest impact on outcomes has been Clinical Decision Support Systems (CDSS) (1). The guidance provided by CDSSs relies on the formalization of clinical knowledge. Computerized Interpretable Guidelines (CIGs) are representations of formalized medical knowledge directly employable by Clinical Information Systems (CIS) to provide decision support. Following the Model Integrated Computing (MIC) paradigm, we developed the Sepsis Treatment Enhanced through Electronic Protocolization (STEEP) (2, 3) as a CDSS application utilizing a novel CIG as a precise formal representation of the Surviving Sepsis Campaign (SSC) Resuscitation and Management Bundles (4). Because STEEP was integrated into existing health information systems including health records, dashboards, and order entry, the application itself focused only on delivering decision support for sepsis assessment and management and deferred other aspects of clinical workflow to established systems. STEEP was built using the visual domain-specific modeling language Clinical Process Modeling Language (CPML) designed for capturing treatment protocols. CPML development occurred iteratively, first using language explicitly representing treatment trajectories as a connected, directed, bipartite graph structure. Nodes were either decision points with multiple, predefined, possible outcomes or actions representing treatment steps. This approach failed to express complex treatments efficiently and scaled poorly and was replaced by grouping treatment steps under process concepts. Processes were viewed as concurrent, asynchronous, and interactive with each other via events. Processes were ordered hierarchically in the Communicating Sequential Processes (CSP) model and could listen to events happening around them and start running only if triggering conditions were satisfied. Testing of the CSP model showed the semantics were more intuitive to clinical providers, perhaps because they more closely resembled the mental process of medical decisionmaking (2). The operational semantics specified the CPMLs runtime behavior. CPML processes have five states: Deactivated, Active, Running, Paused, and Terminated. Processes in the Deactivated state do not perform any actions. Active processes monitor runtime events and, if criteria are met, start Running with sub-processes
activated. The semantics and execution of the CPML are supported by the STEEP execution engine. For the execution of CPML models, the engine had to instantiate the CPML models and execute the state machine associated to the elements while also interpreting the guard conditions of the states. This required storing the CPML models for every patient enrolled in the system. The execution engine starts running the treatment-management process by executing the protocol models and interacting with the Treatment Management Console (TMC), a graphical user interface (GUI) accessed by providers implemented through Google Web Toolkit (GWT). GWT was chosen for its maturity and pre-existence within the EMR environment. The GUI was constructed in an iterative design-evaluation process supervised by a human factors researcher (5) and founded in the action-reaction concept in which display of adjacent patient data and management facilitate linking of cause and effect. Three user interface prototypes (a literal representation of the EGDT algorithm, an interface based on principles of ecological interface design (6), and a hybrid approach) were examined in a summative usability test protocol with fully counterbalanced repeated measures experimental design (7). Based on superior physician performance with incorporation of elements of ecological interface design (8), this prototype was developed into a final user interface which was optimized from a human factors perspective within the technical limitations of the tool s EMR environment.
Figure S1. Schematic of the Listening, Alerting, and Provider Assessment systems. At ICU admission, providers were prompted to assess whether sepsis was present. Patients assessed as septic were randomized. In those assessed as not septic at admission, the listening application monitored for development of modified SIRS criteria and if met the alerting system notified providers by text page and electronic alert. Assessment as septic resulted in randomization and assessment as not septic suppressed further alerts for 48 hours.
Video 1. Overview of the electronic sepsis tool. A video tour of the interface and capabilities of the electronic sepsis assessment and management tool narrated by Liza Weavind, MBBCh MMHC on April 7th, 2014. Author: Liza Weavind, Videographer: Andras Nadas, Participants: Liza Weavind, Length: 3min 12 sec, Size: 26.2MBs
Table S1: Multivariable regression for use of tool to enter orders Characteristic Odds Ratio 95% Confidence Interval P APACHE II score 1.15 0.74-1.80 0.539 In-hospital transfer 1.08 0.49-2.40 0.552 SICU 4.65 2.06-11.0 <0.001 Sepsis Developed in ICU 1.14 0.59-2.2 0.695 Sepsis Not Confirmed by Review 1.29 0.56-3.0 0.544 Time to assessment > 6 hours 0.00 0.00-6.7e+13 0.688 APACHE II, Acute Physiology and Chronic Health Evaluation II; SICU, Surgical Intensive Care Unit
References for Supplemental Digital Content 1. Stead WW, Hammond WE. Computer-based medical records: the centerpiece of TMR. MD Comput. Comput. Med. Pract. 1988;5(5):48-62. 2. Mathe JL, Ledeczi A, Nadas A, Sztipanovits J, Martin JB, Weavind LM, Miller A, Miller P, Maron DJ. A Model-Integrated, Guideline-Driven, Clinical Decision-Support System. IEEE Softw. 2009;26(4):54-61. doi:10.1109/ms.2009.84. 3. Nadas A, Mathe J, Weavind L, Semler M, Sztipanovits J. Lessons Learned from the Development of amodelintegrated Decision-support Application for Sepsis. In: Washington (DC); 2014. 4. Severe sepsis bundles. Resuscitation bundle. Available at: http://www.survivingsepsis.org/bundles/pages/default.aspx. Accessed December 15, 2009. 5. Stanton NA, Salmon PM, Rafferty LA, Walker GH, Baber C. Human Factors Methods: A Practical Guide for Engineering and Design. 2 edition. Burlington, VT: Ashgate Publishing Company; 2013. 6. Bennett KB, Flach JM. Display and Interface Design: Subtle Science, Exact Art. 1 edition. Boca Raton ; London: CRC Press; 2011. 7. Rubin J, Chisnell D, Spool J. Handbook of Usability Testing: How to Plan, Design, and Conduct Effective Tests. 2 edition. Indianapolis, IN: Wiley; 2008. 8. Miller A. Alerts and reminders: Is this all there is to clinical decision support? 2011. Available at: www.cs.umd.edu/hcil/sharp/workshop2011/files/miller,anne-ehr- HCILseminar.pdf#sthash.fY1CVR4W.dpuf.