Decision analysis: a framework for critical care decision-making

Physicians and nurses need assistance with the difficult, high-consequence choices they encounter in caring for critically ill patients. Decision analysis offers a framework for clarifying these complex, uncertain, and dynamic decision problems. This dissertation proposes to integrate decision analysis into critical care practice by using computer programs known as intelligent decision systems (IDSs) that help physicians and nurses formulate, evaluate, and appraise decision models. Critical-care decisions can be grouped into two major categories. Strategic decisions encompass the diagnostic and therapeutic interventions that mark major milestones in the patient's critical care course. This dissertation identifies a generic model structure that complements existing methods for building IDSs for these decisions. Operational decisions encompass the repeated--and frequently delegated--adjustments of supportive therapy that are made in the interval between strategic decisions. This dissertation introduces several concepts and procedures that enable IDSs to be built and used for this class of decisions. In particular, it simplifies an otherwise extremely complex multistage sequential decision problem by using a sequence of two-stage models. It also proposes six generic knowledge maps that capture the extremely complex relevant medical knowledge. The dissertation presents a procedure in which the physician delegates life-support decisions to the nurse by providing a set of instructions for performing a decision analysis. These instructions are termed a decision class analysis (DCA). Assisted by an IDS, the physician creates the DCA by selecting knowledge maps and decision model appraisal rules from a library previously created by critical care specialists and decision analysts. The physician then transmits this DCA to an IDS integrated with the nurse's bedside electronic patient charting system. Each time new data become available, the nurse uses the bedside IDS to execute the DCA. The resulting formulation and evaluation of decision models provides updated recommendations to guide adjustment of the life-support therapy. Appraisal of each updated model provides proactive alarms that indicate the need for additional therapies, for additional monitoring, or for calling the physician to consider modifying the DCA. A pilot decision engineering project has applied these concepts to the problem of adjusting inspired oxygen concentration in a demonstration IDS, Orchestra.