A Bayesian Model Based on Local Phenotypic Resistance Data to Inform Empiric Antibiotic Escalation Decisions

Area of Science:

• Infectious Diseases
• Clinical Microbiology
• Biostatistics

Background:

• Clinicians often escalate empiric antibiotic therapy without microbiology guidance due to poor clinical progress.
• Escalation decisions should consider how resistance to initial antibiotics impacts resistance to subsequent options, a concept termed escalation antibiogram (EA).
• Understanding uncertainty in EA results for specific patient subgroups is crucial for clinical application.

Purpose of the Study:

• To develop and apply a Bayesian model for estimating antibiotic resistance rates and uncertainty.
• To calculate escalation antibiograms for various patient populations, including intensive care unit (ICU), haematology-oncology, and pediatric patients.
• To assess the applicability of hospital-wide EAs to specific patient subgroups.

Main Methods:

• A Bayesian model was developed to estimate antibiotic resistance rates in Gram-negative bloodstream infections using phenotypic resistance data.
• The model generates credible curves to fit resistance data, incorporating an integrated penalisation term for adaptive smoothing.
• Credible intervals were used to illustrate uncertainty in resistance rate estimates, particularly for smaller patient subgroups.

Main Results:

• Resistance rates to empiric and escalation antibiotics were calculated for 10,486 bloodstream infections across the general hospital population and specific groups (ICU, haematology-oncology, pediatric).
• Significant differences in resistance rates were observed between patient subgroups; for example, piperacillin/tazobactam resistance was 27.3% in ICU patients versus 13.4% in the general population.
• The model estimated the probability of inferiority between antibiotics and highlighted differences in optimal escalation options across patient groups.

Conclusions:

• Bayesian-informed escalation antibiogram (EA) analysis is valuable for guiding empiric antibiotic switches by estimating local resistance rates and comparing options with uncertainty measures.
• Hospital-wide EAs cannot be reliably applied to specific patient groups due to significant variations in resistance patterns.
• The model provides a quantitative approach to managing uncertainty in antibiotic resistance data for improved clinical decision-making.