Introduction: To tackle antimicrobial resistance, strategies to reduce inappropriate consumption of antibiotics have been implemented in health systems in the UK. In secondary care, behavioural antimicrobial stewardship (AMS) interventions have had limited effect in optimising doctors’ antibiotic prescribing practices as reflected in the observed deviation from prescribing guidelines. A powerful analytical tool is required to capture the determinants of antibiotic prescribing decision-making within the complexity of the health system to inform intervention design and allow subsequent impact evaluation.
Methods: A System Dynamics (SD) model was constructed to enable the simulation of antibiotic prescribing decision-making processes in English hospitals. Literature review, secondary analysis of patient records, healthcare professional interviews and survey responses provided the data from SD model development. A qualitative SD model was formulated and validated through a series of tests to ensure the credibility of the model to simulate doctors’ behaviours under different scenarios. Computer-based simulations were performed to predict doctors’ prescribing outcomes measured by the level of guideline compliance.
Results: Maximal improvement in doctors’ guideline compliance at empiric stage could be achieved if senior doctors’ guideline compliance was increased, combined with the microbiology laboratory turnaround time of blood cultures within 24 hours. At review stage, doctors’ decision-making could be enhanced if more opportunities for specialist microbiology input were provided. Improving guideline compliance in junior doctors alone had limited impact on overall prescribing outcomes.
Conclusions: SD modelling enables capture of causal mechanisms and prediction of prescribing outcomes by transparent qualitative mapping and quantitative simulation of complex, dynamic prescribing decision-making processes. SD acknowledges that doctors’ sub-optimal antibiotic prescribing is a feature of bounded rationality compounded by multiple contextual interacting influences. AMS interventions can be re-designed for maximum effect with minimal additional resources, if the most influential healthcare professionals and the most critical events in the prescribing decision-making processes are targeted.