Hierarchical Bayesian Decision Tool for Empirical Antibiotic Selection in Rural Australian Hospitals Using ATLAS Regional Susceptibility Priors

Rural and regional hospitals in Australia frequently treat serious infections — pneumonia, urinary sepsis, abdominal infections — that require immediate empirical antibiotic therapy before culture results are available. Clinicians choose antibiotics based on national guidelines (Australian Therapeutic Guidelines) and their hospital's local antibiogram, which shows what proportion of common bacteria are susceptible to each antibiotic.

The problem is that small rural hospitals test very few bacterial isolates — often fewer than 30 per pathogen per quarter. At these volumes, the local antibiogram is statistically unreliable. A hospital might see 8 out of 12 E. coli isolates susceptible to ceftriaxone (67%), but the true susceptibility rate could plausibly be anywhere from 40% to 88%. Clinicians currently have no principled way to reconcile their sparse local data with the broader regional picture.

This project builds a Bayesian statistical tool that combines regional susceptibility data (from the Pfizer ATLAS surveillance programme) with whatever local antibiogram data a hospital has. When local data is sparse, the tool leans on regional patterns; as local data accumulates, the tool gradually shifts toward local estimates. It then ranks candidate antibiotic regimens by balancing predicted effectiveness against collateral harms (C. difficile risk, ecological resistance pressure, cost, and patient allergies).

Improving patient outcomes: The tool gives rural clinicians a more reliable estimate of which empirical antibiotic is most likely to be effective for their specific hospital, rather than relying on imprecise local point estimates or generic national defaults. Better empirical antibiotic selection means faster effective treatment, which directly reduces morbidity and mortality from serious infections.

Strengthening stewardship: The decision layer explicitly penalises broad-spectrum antibiotics and those with high C. difficile risk. By quantifying the trade-off between coverage probability and collateral damage, the tool helps clinicians make stewardship-aligned choices — particularly in settings without on-site infectious diseases or antimicrobial stewardship pharmacist support.

Informing public health practice: The project will quantify, for the first time, where current Australian Therapeutic Guidelines recommendations are robust to local susceptibility uncertainty and where they may be suboptimal for specific regional contexts. This evidence can inform future guideline updates and targeted surveillance resource allocation.

Strengthening health systems: Rural hospitals are systematically disadvantaged in antimicrobial decision-making because their data volumes are too low for reliable local inference. This tool directly addresses that structural inequity by enabling small-volume sites to borrow statistical strength from regional surveillance, reducing the information gap between metropolitan and rural facilities.