Integrated Clinical Decision Support for Empiric Antibiotic Selection in Sepsis

As antibiotic resistance increases globally, it becomes more difficult to select empiric antibiotic therapy, particularly in patients with sepsis who stand to benefit from early adequate treatment. In particular it is difficult for clinicians to balance antibiotic stewardship principles (the need to avoid unnecessary prescribing of antibiotics that have an excessively broad spectrum of activity that favour resistance development) and under treatment. The integration of multiple risk variables for resistance are hard for clinicians to translate into clinical action, and is seemingly at odds with the natural inclination to provide heuristic/emotion-based antibiotic selection. The inappropriate treatment of sepsis is not uniformly too broad, or too narrow, and there is a need to optimize and tailor selection of antibiotic therapy to each patient, such that those that are at risk for resistant organisms receive broad therapy, and those that are not at risk, receive narrower antibiotic agents. Clinicians need support picking the right antibiotic for each patient, and from this they can potentially drive reduction of unnecessarily broad antibiotic prescribing while preserving adequacy of treatment. Individualized clinical prediction models and decision support interventions are promising approaches that meet these needs by improving the classification of patient risk for antibiotic resistant or susceptible infections in sepsis. Unfortunately, few have been validated in the clinical setting and larger rigorous studies are needed to provide the evidence to support broader clinical adoption. The investigators will perform a cluster randomized cross-over trial of an individualized antibiotic prescribing decision support intervention for providers treating hospitalized patients with suspected sepsis. The aim of this trial is to determine whether a stewardship led clinical decision support intervention can improve antibiotic de-escalation in patients with sepsis while maintaining or improving adequacy of antibiotic coverage. This decision support intervention will be based on a combination of proven decision heuristics (for Gram-positive organisms) and modelled predicted susceptibilities (for Gram-negative organisms) that are individualized to the patient. The primary outcome will be the proportion of patients de-escalated from their initial empiric regimen at 48 hours.