Personalised antimicrobial susceptibility testing with clinical prediction modelling informs appropriate antibiotic use

Abstract

Antimicrobial susceptibility testing is a key weapon against antimicrobial resistance. Diagnostic microbiology laboratories use one-size-fits-all testing approaches that are often imprecise, inefficient, and inequitable. Here, we report a personalised approach that adapts laboratory testing for urinary tract infection to maximise the number of appropriate treatment options for each patient. We develop and assess susceptibility prediction models for 12 antibiotics on real-world healthcare data using an individual-level simulation study. When combined with decision thresholds that prioritise selection of World Health Organisation Access category antibiotics (those least likely to induce antimicrobial resistance), the personalised approach delivers more susceptible results (results that encourage prescription of that antibiotic) per specimen for Access category antibiotics than a standard testing approach, without compromising provision of susceptible results overall. Here, we show that personalised antimicrobial susceptibility testing could help tackle antimicrobial resistance by safely providing more Access category antibiotic treatment options to clinicians managing urinary tract infection.

Fig. 1. Performance of probability predictions on the validation dataset.

Plots (a–l) correspond to the 12 antimicrobial agents for which susceptibility prediction models were developed. Purple lines represent receiver operating characteristic (ROC) curves for the predictive value of binary logistic regression models for susceptible (‘S’) results when applied to a validation dataset, with chance level (level of performance if the model had no predictive value) represented by black dashed lines, and area under the curve (AUC) provided in inset boxes. Source data are provided as a Source Data file.

Fig. 2. Specimen pathway sensitivity analysis.

Susceptibility prediction performance throughout the simulated laboratory specimen pathway expressed as AUC-ROC values, as further information becomes available in the form of organism ID (identification) and other antimicrobial susceptibility testing (AST) results. Source data are provided as a Source Data file.

Fig. 3. Microsimulation study results-per-specimen analysis.

The number of susceptible results per specimen provided by application of the personalised approach (PDAST) and standard fixed approach in the individual-patient simulation (microsimulation) study for all agents (left) and for WHO Access agents (right). The six results were chosen based on the PDAST or standard panel recommendations from the 12 actual real-world results that were available in the study dataset, representing a simulation of a real-world testing decision and outcome. Blue and red circles represent median values, blue and red lines represent the interquartile range, and grey dot cloud darkness represents the number of results from all specimens. Source data are provided as a Source Data file.

Fig. 4. Microsimulation study decision threshold sensitivity analysis.

a, b display the effect of varying testing decision threshold on number of susceptible results per panel for all agents and WHO Access category agents respectively. Plots c and d display the effect of varying testing decision threshold on the percentage of panels with least one susceptible result for all agents and WHO Access category agents respectively. In plots a and b, blue lines represent median number of susceptible results per panel for the personalised approach (PDAST), shading represents interquartile range. In plots (c, d) blue lines represent percentage of specimens providing at least one susceptible result. Red dashed lines represent median number of susceptible results per specimen (a, b) and percentage of specimens with at least one susceptible result (c, d) with the standard panel. Source data are provided as a Source Data file.

Fig. 5. Microsimulation study results-per-antimicrobial agent analysis.

The total number of susceptible antimicrobial susceptibility testing (AST) results provided by the personalised (PDAST) approach and standard approach for all 12 antimicrobial agents studied. WHO Access category agent axis labels are green in colour, while those for Watch category agents are orange. Source data are provided as a Source Data file.

Fig. 6. Results provided for the antimicrobial agent inpatients were prescribed.

Bar size corresponds to the number of prescriptions, with susceptible (S) results in green and resistant (R) results in red. Counts for the standard approach are displayed to the left of the central vertical line, and counts for the personalised approach are displayed to the right of the line. Source data are provided as a Source Data file.

Fig. 7. Flow chart for the study population.

The initial dataset was cleaned, then split into model development and microsimulation datasets. The model development dataset was subsequently split into training-testing dataset used for hyperparameter tuning and feature selection, and a validation dataset used once to measure predictive performance. Created in BioRender. Howard, A. (2024) BioRender.com/z80x717.

Fig. 8. Design of the microsimulation study.

For each specimen in the microsimulation study dataset, a personalised panel was composed using prediction modelling and a test prioritisation function to maximise the probability of susceptible (‘S’) results for Access agents, and failing that, susceptible results for other agents—the results for agents in this panel were then populated by actual results for those agents in the dataset. The number of susceptible results for Access agents and all agents that would have been provided by using this panel was then compared against a standard panel based on international UTI treatment guidelines. Created in BioRender. Howard, A. (2024) BioRender.com/k67g601.