Decision by Design - Decision Support for Antibiotic Prescribing in Critical Care

Decision support systems (DSS) are traditionally designed to optimise the outcomes of a decision. This thesis explores how DSS design can also be driven by the optimisation of the decision process leading to the decision, and how it may enhance the human uptake and use of DSS. It identifies which tasks could be simplified by decision support, and how to build DSS that are likely to be readily adopted and so improve decision outcomes. It tests the hypotheses that:(a) The analysis of specific process attributes of a given clinical decision task, as well as the information needs of its users, improves the design of DSS and enhances systems?impact and acceptance.(b) The complexity of the decision task is the key process attribute that, inconjunction with the information seeking of users, shapes the outcome of the design process.The work is applied to the domain of antibiotic prescribing in critical care.To explore the first hypothesis, the key attributes of prescribing decisions associated with specific prescribing subtasks and different decision-makers and decision contexts are identified and then analysed. Based on our findings, an information-processing model of decision support for an antibiotic-prescribing task is proposed. The second hypothesis is addressed by applying and comparing metrics for decision complexity including minimum message length, cognitive effort assessment and clinical algorithm structure analysis to the prescribing task. A framework is developed to select clinical decision tasks that may benefit from automation, by characterizing decision support as a process of complexity reduction for users, and these ideas are tested in the context of antibiotic prescribing for ventilator-associated pneumonia.The hypotheses are then tested by applying the task complexity framework to the design of a DSS for antibiotic prescribing in critical care. A web-based experiment and a clinical trial of the DSS are described, both of which study the acceptability and effectiveness of the system and verify the usefulness of the design framework. Specifically, in a before-after controlled trial, with no difference in patient mortality orseverity of presentation between the two periods, the use of the DSS was associated with statistically significant improvements in patient outcomes and a reduction in antibiotic usage. The length of stay and total consumption of antibiotics decreased respectively from 7.15 to 6.22 days (P=0.02) and from 1767 to 1458 defined daily doses/1000 patient days (P=0.04). The introduction of a hand-held computer-based DSS was associated with less administration of ‘broad-spectrum?antibiotics.The relative impact of the uptake of the DSS on the prescribing quality was quantified. Clinicians chose to use guidelines for one third, and pathology data or the DSS for about two thirds of cases for which they were available to assist their prescribing decisions. When used, the DSS plus pathology data improved the agreement of decisions with those of an expert panel - from 65% to 97% (P=0.002). The impact of the DSS was more significant on prescribing decisions of higher complexity. The level of decision complexity appeared to affect the choice of decision support type. Prescribing guidelines were accessed more often for lower complexity decisions, whereas the infection risk DSS plus pathology data were preferred for decisions of higher complexity.The need for measurement of the effectiveness of a DSS in improving decisions, as well as their likely rate of adoption in the clinical environment, was demonstrated. The thesis concludes with a proposal to apply the framework described to the modelling of the DSS adoption and to include task complexity and user information seeking as determinants of the design and evaluation of clinical DSS