BEAR - Bayesian Estimation of Antimicrobial Resistance Prevalence

BEAR aims to improve data-driven population infection management using a novel simplified algorithmic application of Bayesian inference. The algorithm enables infection specialists in resource-restricted settings to more accurately estimate AMR prevalence in their local populations, better informing antimicrobial stewardship, infection prevention and control and drug deployment programmes.

Dr Alex Howard (solution lead)

Consultant in Medical Microbiology and Infectious DIseases

Liverpool University Hospitals NHS Foundation Trust

University of Liverpool

The prevalence of antimicrobial resistance (AMR) is a crucial metric in population infection management. Antimicrobial formularies, antimicrobial stewardship, infection prevention and control, and deployment of new drugs are examples of population infection management strategies that hinge on understanding AMR prevalence. For example, AMR prevalence may inform whether an antimicrobial agent should be used as a first-line treatment for urinary tract infection, the resource-effectiveness of AMR screening strategies, and the likely effects of deployment of drugs in new populations (e.g., neonates).

Real-world clinical microbiology data may not represent the true prevalence of AMR in the population for whom a decision needs to be made because many people who have never had microbiology specimens sent may have different patterns of AMR risk factors (e.g., healthcare exposure, antimicrobial treatment), and estimates for seldom-tested antimicrobial agents are vulnerable to extreme estimation errors due to small sample sizes.

In resource-rich settings, the approach to this problem is to test more specimens in the hope that the observed prevalence in a larger sample size begins to better represent the true population prevalence. However, in resource-restricted global settings this approach is resource intensive and often infeasible.

Our target audience is local, regional, and international decision makers in antimicrobial stewardship, infection prevention and control, and new drug deployment. We seek to support these people by providing them with accessible means to better inform their AMR-combatting activities using resource-efficient targeted antimicrobial susceptibility testing. We have reached out to stakeholders in the local area and internationally with the proof-of-concept data and received useful feedback. We intend to disseminate findings in an open-source peer-reviewed journal and to engage other stakeholders in a rage of global settings through pre-existing AMR research networks.

BEAR provides public good by giving decision makers a tool free of charge to make better population-level antimicrobial management and infection prevention and control decisions, resulting in less antimicrobial harm and harm from multidrug-resistant organisms across whole populations. The solution also provides a sustainable means of maintaining and carrying forward more reliable antimicrobial susceptibility data to inform future decision making at local and international level. We also plan to publish the work in an open-source peer-reviewed journal.

BEAR targets under-resourced healthcare settings where large-scale antimicrobial susceptibility testing and/or access to high computational power are not feasible. Its tangible impact would be via better informed decision making to inform local antimicrobial resourcing, deployment, use, and infection prevention and control - this could result in reduced antimicrobial treatment harm and spread of multi drug-resistant organisms. The solution would drive this solution using a dashboard interface for which clinicians / healthcare decision makers would be given a short training session. They would then be able to input their prior belief into the interface along with either manual input or upload of a small sample of local antimicrobial susceptibility data. BEAR would then output an estimate of local prevalence that they could use to inform population infection management decisions such as policy, as well as providing an ongoing record of changes in local antimicrobial susceptibility prevalence for reporting / future infection management purposes.

The impact would be scaled by the development of a user-friendly graphic interface that would streamline the process of clinicians uploading and/or manually inputting antimicrobial susceptibility data and their prior knowledge. An interactive e-learning package will also be developed for BEAR users with support for multiple languages, along with a website that houses links to the open-source software required, a means of users providing feedback, the open-source Github code, and proformas for entry of antimicrobial susceptibility data. The intervention will then be trialled via healthcare contacts at sites linked to the Centres for Antimicrobial Optimisation Network (CAMO-NET). Initial rollout will be restricted to sites with basic laboratory infrastructure to facilitate targeted testing if required.

Performance will be measures using antimicrobial usage metrics (i.e., defined daily doses, indications, IV to oral step-downs, WHO Access/Watch/Reserve agent usage), outcome metrics (i.e., mortality from infection, admission/readmission with infection, multi drug-resistant organism colonisation/infection, Clostridioides difficile diarrhoea), and fidelity measures such as BEAR usage, qualitative user feedback, and composition/coverage/adherence to local and guidelines).

The potential barriers to BEAR implementation in some areas would be: antimicrobial agent availability (potentially limiting the ability of BEAR to fully make an impact on population infection management decisions by limiting treatment options); electrical power and temporary internet access (required to run hardware, download open-source software and access online resources); local infection knowledge/expertise (required to provide an informed opinion on likely antimicrobial susceptibility rates that provide prior AMR prevalence belief to inform estimates, and an understanding of which drug-pathogen combinations it will be effective for); the intervention requires some degree of infrastructure for the provision and dissemination of infection management decisions at population level (e.g., guidelines) in order to be maximally effective.

The resources that will be pursued for this will be hardware and face-to-face training/education for pilot areas where required. A BEAR network will also be formed for pilot sites to share challenges, ideas, and solutions to barriers. Beyond the initial rollout, solutions to facilitate basic in-situ antimicrobial susceptibility testing for sites without basic laboratory infrastructure will be pursued.

We are applying to the Trinity challenge to help a simple mathematical solution cross the implementation gap into global clinical population health infection management practice. We aim for the effect of this to be a sustainable, resource efficient way to provide equity of population infection management strategy and policy provision to a range of global settings. Once implemented, we hope BEAR to take the form of an easily accessible user interface that local specialists use as part of their routine practice to help reduce harm from antimicrobial use and antimicrobial resistance at population level.

We would like to access expertise in implementation science for global resource-restricted settings in order to help with initial pilots and overcoming local barriers. We would also benefit from routes to assistance with language translation if required were the intervention to scale beyond the initially established network.