Rapid and accurate diagnostics of bacterial infections are necessary for efficient treatment of antibiotic-resistant pathogens. Cultivation-based methods, such as antibiotic susceptibility testing (AST), are limited by bacterial growth rates and seldom yield results before treatment needs to start, increasing patient risk and contributing to antibiotic overprescription. Here, we present a deep-learning method that leverages patient data and available AST results to predict antibiotic susceptibilities that have not yet been measured. After training on three million AST results from 30 European countries, the method achieved an average accuracy of 93% across bacterial species and antibiotics. It predicted susceptibility with an average major error rate below 5% for quinolones, cephalosporins, and carbapenems, and below 8% and 14% for aminoglycosides and penicillins, respectively. Furthermore, the model predicted resistance with an average very major error rate below 10% for cephalosporins, carbapenems, and aminoglycosides, but with higher very major error rates for penicillins and quinolones. We combined the method with conformal prediction and demonstrated accurate estimation of the predictive uncertainty at the patient level. Our results suggest that artificial intelligence-based decision support may offer new means to meet the growing burden of antibiotic resistance.IMPORTANCEImproved diagnostic tools are vital for maintaining efficient treatment of antibiotic-resistant bacteria and for reducing antibiotic overconsumption. In our research, we introduce a new deep learning-based method capable of predicting untested antibiotic resistance phenotypes. The method uses transformers, a powerful artificial intelligence (AI) technique that efficiently leverages both antibiotic susceptibility tests (AST) and patient data simultaneously. The model produces predictions that can be used as time- and cost-efficient alternatives to results from cultivation-based diagnostic assays. Significantly, our study highlights the potential of AI technologies to address the increasing prevalence of antibiotic-resistant bacterial infections.

Abstract Antimicrobial resistance is a public health challenge, driving the need for rapid, cost-effective diagnostic support tools. Artificial intelligence (AI) may enable prediction of susceptibility to untested antibiotics from known susceptibility results, but prospective clinical validation is required before routine use. We evaluated an AI-based decision support method, trained on invasive isolates from the European Surveillance System (TESSy), for prediction of antibiotic susceptibility in clinical Escherichia coli urine isolates. The evaluation included 99 E. coli isolates from urine samples with diversity in age, sex, and antibiotic susceptibility. Predictions were evaluated for 14 antibiotics using patient metadata and susceptibility results for 4–8 antibiotics as input. Prediction uncertainty was handled using conformal prediction, allowing abstention when confidence was insufficient. EUCAST disk diffusion test results were used as reference and genomic sequence data was used to explore mechanisms of the AI performance. Without conformal prediction, 84% of predictions were correct when susceptibility results of six antibiotics were used to predict susceptibility to eight additional antibiotics. Across all predictions generated using susceptibility results for six antibiotics as input, the major and very major error rates were 19% and 12%, respectively. Prediction errors varied between antibiotics and were associated with certain phenotypic and genotypic resistance patterns. Conformal prediction reduced errors but increased abstentions; at confidence levels of 90%, 95%, and 97.5%, the model abstained in 9.6%, 14%, and 22% of instances. The method showed promising performance, but its clinical use remains limited and may require diagnostic data beyond susceptibility test results and demographic variables. Importance Antibiotic susceptibility testing is essential for guiding treatment of bacterial infections, but results are often incomplete when initial treatment decisions are made. This study evaluates a novel diagnostic concept: using artificial intelligence to extend the information obtained from partial susceptibility test results, rather than replacing routine susceptibility testing. The method predicts susceptibility to untested antibiotics from patient metadata and existing phenotypic susceptibility results. In this prospective evaluation of clinical Escherichia coli urine isolates, we assessed both prediction performance and uncertainty control by conformal prediction, which allows the method to abstain when predictions are insufficiently reliable. By linking prediction errors to phenotypic and genotypic resistance patterns, the study identified both potential and current limitations of AI-based susceptibility prediction. These findings move AI-based antimicrobial resistance prediction from retrospective model development toward prospective clinical evaluation, a necessary step before implementation.

To identify the spread of nosocomial infections and halt outbreak development caused by Escherichia coli that carry multiple antibiotic resistance factors, such as extended-spectrum beta-lactamases (ESBLs) and carbapenemases, is becoming demanding challenges due to the rapid global increase and constant and increasing influx of these bacteria from the community to the hospital setting. Our aim was to assess a reliable and rapid typing protocol for ESBL-E. coli, with the primary focus to screen for possible clonal relatedness between isolates. All clinical ESBL-E. coli isolates, collected from hospitals (n = 63) and the community (n = 41), within a single geographical region over a 6 months period, were included, as well as clinical isolates from a polyclonal outbreak (ST131, n = 9, and ST1444, n = 3). The sporadic cases represented 36 STs, of which eight STs dominated i.e. ST131 (n = 33 isolates), ST648 (n = 10), ST38 (n = 9), ST12 and 69 (each n = 4), ST 167, 405 and 372 (each n = 3). The efficacy of multiple-locus variable number tandem repeat analysis (MLVA) was evaluated using three, seven or ten loci, in comparison with that of pulsed-field gel electrophoresis (PFGE) and multi locus sequence typing (MLST).

Extended-spectrum β-lactamase producing Escherichia coli (ESBL-E. coli) were isolated from infants hospitalized in a neonatal, post-surgery ward during a four-month-long nosocomial outbreak and six-month follow-up period. A multi-locus variable number tandem repeat analysis (MLVA), using 10 loci (GECM-10), for 'generic' (i.e., non-STEC) E. coli was applied for sub-species-level (i.e., sub-typing) delineation and characterization of the bacterial isolates. Ten distinct GECM-10 types were detected among 50 isolates, correlating with the types defined by pulsed-field gel electrophoresis (PFGE), which is recognized to be the 'gold-standard' method for clinical epidemiological analyses. Multi-locus sequence typing (MLST), multiplex PCR genotyping of bla CTX-M, bla TEM, bla OXA and bla SHV genes and antibiotic resistance profiling, as well as a PCR assay specific for detecting isolates of the pandemic O25b-ST131 strain, further characterized the outbreak isolates. Two clusters of isolates with distinct GECM-10 types (G06-04 and G07-02), corresponding to two major PFGE types and the MLST-based sequence types (STs) 131 and 1444, respectively, were confirmed to be responsible for the outbreak. The application of GECM-10 sub-typing provided reliable, rapid and cost-effective epidemiological characterizations of the ESBL-producing isolates from a nosocomial outbreak that correlated with and may be used to replace the laborious PFGE protocol for analyzing generic E. coli.