Optimizing vancomycin dosage is critical for treating severe infections and combating antimicrobial resistance, yet it is hampered by slow, centralized laboratory testing. To address this clinical gap, we have developed a low-cost, disposable electrochemical sensor for the rapid quantification of vancomycin directly in undiluted human serum. Our platform integrates a selective molecularly imprinted polymer using phenol red as a redox-active functional monomer with a signal-amplifying highly porous gold nanostructure on a scalable printed circuit board. To address non-linear responses and matrix interference inherent to complex biological samples, the sensor output is processed by a Random Forest machine learning regression model. The sensor achieved a limit of detection of 0.848 μg mL-1 within a clinically relevant dynamic range (0-100 μg mL-1). As a preliminary proof-of-concept for clinical application, the sensor was tested using patient serum samples, demonstrating good correlation (R2 = 0.98) and agreement when compared against gold-standard liquid chromatography-tandem mass spectrometry (LC-MS/MS). This work presents a data-driven sensor system that offers a robust alternative to conventional methods, paving the way for real-time, personalized vancomycin therapy at the point of care.

BACKGROUND:Development of clinical decision support systems (CDSS) has been ongoing for over 60 years, more recently leveraging technologies such as artificial intelligence (AI) and machine learning (ML). Intelligent CDSS addressing different stages of the infection management process offer potential advantages in interpreting complex data and guiding clinical decision-making. OBJECTIVES:We outline the current applications of AI-driven CDSS across the continuum of bacterial infection management, from prevention and diagnosis to antibiotic prescribing and treatment individualization. We discuss the main limitations hindering their translation into clinical practice, as well as opportunities to improve their development to better meet clinical needs. METHODS:References for this review were identified through searches of PubMed, Google Scholar, bioRxiv and arXiv up to March 2025 by use of a combination of ML, decision-making and bacterial infection keywords. KEY FINDINGS:AI-CDSS studies increasingly leverage multimodal electronic health record (EHR) data, with most adopting lower-complexity models that perform well on structured data, particularly when supported by effective feature engineering. Despite efforts to develop accurate AI-driven systems, some of which achieve clinician-level accuracy in solving diagnostic and prescribing tasks, AI-CDSS have largely failed to integrate into clinical settings. Their adoption faces challenges related to the narrow scope of the defined medical task, failure to consider stakeholder workflow and lack of proper evaluation frameworks. CONCLUSION:There is a need to shift CDSS development towards a more adaptive and holistic approach that recognizes the continuous nature of the decision-making process in infection management. Comprehensive AI-powered platforms that can model infection dynamics could improve antibiotic stewardship and help tackle the global health emergency of antimicrobial resistance.

BACKGROUND:The ability to monitor host- and bacteria-specific biomarkers along with antimicrobial drug concentration at the site of infection offers potential for individualised approaches to antimicrobial therapy. Although urine collection is straightforward and directly linked to the infection site, the assessment of urinary tract infection (UTI) biomarkers during infection has not been extensively explored. The aim of this study is to evaluate the potential of monitoring urinary nitrite levels as a biomarker for antimicrobial pharmacodynamics in UTI treatment. METHODS:Resistant and susceptible E. coli strains were cultured in oxygen-free artificial urine, with amoxicillin added after 15 h. Colony-forming unit (CFU) counts, nitrite, and creatinine levels were measured at 5 timepoints over 66 h. Urine samples from 25 UTI patients and 25 non-UTI controls were analysed for bacterial growth, nitrite, and creatinine. Spearman rank correlation and Mann-Whitney U-tests were used for statistical analysis. RESULTS:Our in-vitro model demonstrates that measuring the bacteria-specific urinary biomarker nitrite during E. coli growth in artificial urine can effectively be applied to assess antimicrobial pharmacodynamics over the course of UTI treatment. In an in-vitro UTI model, nitrite concentration can differentiate between resistant and susceptible E. coli strains and correlates with CFU counts. Analysis of 25 clinical UTI samples is consistent with these findings, showing correlations between nitrite levels and CFU counts. CONCLUSIONS:Here we show that nitrite generation by E. coli may have clinical relevance as a biomarker for infection progression and antimicrobial treatment outcomes, offering a valuable tool for monitoring the pharmacodynamic responses to antimicrobial therapy in UTIs.

Antibiotic optimisation through therapeutic drug monitoring (TDM) is a key strategy in tackling antimicrobial resistance. However, current quantification methods rely on laboratory-based equipment, delaying timely dose adjustments. We developed a novel point-of-care lateral flow assay (LFA) for the quantification of vancomycin in serum. Unlike conventional LFAs with a single test and control line, this assay incorporates two test lines: one for vancomycin binding and one for avidin capture of excess conjugate. Gold nanoparticles were conjugated to biotinylated bovine serum albumin-vancomycin, enabling competitive binding with anti-vancomycin antibodies and avidin capture. Signal ratios from smartphone-acquired images were quantified using an automated image analysis algorithm. Applied to vancomycin in serum, the LFA demonstrated a quantitative detection range of 2.88 to 45,000 ng/mL, with good accuracy, reproducibility, and high recovery in spiked samples. This approach offers a promising solution for decentralised, rapid TDM in clinical settings, especially where timely dose optimisation is critical.

The increasing prevalence of beta-lactamase-producing Enterobacteriaceae presents a critical challenge in clinical microbiology, complicating infection management due to resistance to most first- and second-line beta-lactam antibiotics. Current antimicrobial susceptibility testing (AST) approaches may not reliably detect and rapidly characterise beta-lactamase production, potentially leading to suboptimal treatment. In this study, we introduce a novel iridium oxide-based biosensor that detects in vitro beta-lactamase production in Escherichia coli (E.coli) type strains and genetically modified E.coli isolates within 10 min. The biosensor demonstrated high reproducibility (6.88% coefficient of variation). Clinical validation using 26 E. coli isolates from urinary tract infection patients showed a sensitivity of 100%, specificity of 57%, and accuracy of 88%. This biosensor-based method offers a rapid, reproducible, and cost-effective method for phenotypic detection of antimicrobial resistance, with the potential to significantly reduce diagnostic turnaround times and support individualised AST strategies.

Antimicrobial resistance (AMR) is a global health challenge that requires cross-disciplinary collaboration to mitigate its impact on human health. We discuss some of the topical advances in the field, highlighting the AMR collection , which brings attention to the problem of AMR and suboptimal antimicrobial use in human medicine.

Artificial intelligence (AI) is reshaping infectious disease diagnostics by supporting clinical decision making, optimising laboratory and clinical workflows, and enabling real-time disease surveillance. AI approaches improve pathogen detection, antimicrobial stewardship, and treatment monitoring, enhancing diagnostic accuracy, efficiency, and scalability. The role of AI in combating antimicrobial resistance is particularly significant, enabling rapid pathogen identification and personalised treatment. Despite progress over the past two decades, widespread AI adoption in infectious disease diagnostics faces challenges. In high-income countries, fragmented data ecosystems, incomplete datasets, and algorithmic bias hinder clinical integration. Meanwhile, low-income and middle-income countries contend with limited digital infrastructure, unstandardised data, and financial constraints, exacerbating disparities in diagnostic access. Further barriers include concerns over interoperability, data privacy, cybersecurity, and the regulation of AI implementation. This paper examines the role of AI in infectious disease diagnostics, highlighting both opportunities and limitations. It underscores the need for coordinated investments in digital infrastructure, harmonised data-sharing frameworks, and clinician engagement to support equitable, sustainable adoption. Addressing these challenges will enable health-care systems to harness the potential of AI to improve infectious disease detection, prevention, and management of infectious diseases, thereby strengthening global health resilience.

Background Blood cultures are the gold standard for diagnosing bacterial bloodstream infections, but test results are only available 24-48 h after sampling. We aimed to develop and evaluate models using health-care data to predict bloodstream infections in patients admitted to hospital. Methods In this retrospective cohort study, we used routinely collected blood biomarkers and demographic data from patients who underwent blood sample collection for testing via culture between March 3, 2014, and Dec 1, 2021, at Imperial College Healthcare NHS Trust (London, UK) as model features. Data up to 14 days before blood sample collection were provided to long short-term memory (LSTM) or static logistic regression models. The primary outcome was prediction of blood culture results, defined as a pathogenic bloodstream infection (ie, isolation of pathogenic bacteria of interest) or no bloodstream infection (ie, no growth or contamination). Data collected up to Feb 28, 2021 (n=15 212) comprised the training set and were evaluated against a temporal hold-out test set comprising patients who were sampled after March 1, 2021 (n=5638). Findings Among 20 850 patients with available data, pathogenic bacteria were observed in the cultured blood samples of 3866 (18.5%) patients. 2920 (62.2%) of 4897 patients who had their blood samples taken more than 48 h after admission to hospital had pathogenic bloodstream infections, and so were defined as having hospital-acquired bloodstream infections. Including data from the 7 days before admission (7-day window approach) and using five-fold cross validation in the training set gave an area under receiver operator curve (AUROC) of 0.75 (IQR 0.68-0.82) and an area under the precision recall curve (AUPRC) of 0.58 (0.46-0.77) for static models and an AUROC of 0.92 (0.91-0.93) and AUPRC of 0.75 (0.72-0.76) for the LSTM model. In the hold-out test set performances were: AUROC of 0.74 (95% CI 0.70-0.78) and AUPRC of 0.48 (0.43-0.53) for static models and AUROC of 0.97 (0.96-0.97) and AUPRC of 0.65 (0.60-0.70) for LSTM. Removal of time series information resulted in lower model performance, particularly for hospital-acquired bloodstream infections. Dynamics of C-reactive protein concentration, eosinophil count, and platelet count were important features for prediction of blood culture results. Interpretation Deep learning models accounting for longitudinal changes could support individualised clinical decision making for patients at risk of bloodstream infections. Appropriate implementation into existing diagnostic pathways could enhance diagnostic stewardship and reduce unnecessary antimicrobial prescribing.

Therapeutic drug monitoring (TDM) is essential for optimizing antibiotic dosing, particularly in critically ill patients. However, conventional methods, such as LC/MS, have long turnaround times, limiting timely dose adjustment. We developed a novel competitive lateral flow assay (LFA) format with dual test lines previously validated for vancomycin and adapted it for Meropenem quantification. Using BlaR-CTD, a beta-lactam receptor protein, in place of an antibody and biotinylated BSA-Meropenem conjugated to gold nanoparticles, the LFA produced a concentration-dependent change in test line intensities. A custom image analysis algorithm showed strong correlation with Meropenem concentrations (R 2 = 0.9537). The platform demonstrates the potential for rapid, point-of-care antibiotic monitoring across diverse healthcare settings. While further optimization is needed for low-concentration accuracy, this proof-of-concept supports broader applicability to other beta-lactams.

BACKGROUND:Voriconazole is the first-line therapy for invasive aspergillosis (IA). To determine the minimum inhibitory concentration of Aspergillus, a voriconazole pharmacokinetic-pharmacodynamic (PK-PD) model linked to galactomannan response was developed and evaluated, and its clinical correlation for IA treatment was elucidated. METHODS:Adult patients with probable or definite IA and at least one serum voriconazole measurement were included. A two-compartment voriconazole PK model was linked to a previously described PD model of galactomannan response. PK and PD parameters were estimated using a nonparametric adaptive grid technique. The relationship between the ratio of voriconazole exposure that induced half-maximum galactomannan response (EC50) and the observed terminal galactomannan concentration was evaluated. The factors associated with the PK-PD parameters and mortality were also determined. RESULTS:Between January 2013 and December 2022, 41 patients were prescribed voriconazole for IA. The 30-day mortality rate was 17%. A high correlation was found for the observed-predicted Bayesian posterior estimates of voriconazole and galactomannan levels. Moreover, a nonlinear relationship was identified between AUC:EC50 and terminal galactomannan. The factors associated with higher AUC:EC50 were intravenous administration and intubation. In the survival analysis, higher EC50 tended to be associated with mortality, higher AUC was significantly associated with increased mortality, and higher AUC:EC50 tended to be associated with higher mortality. After adjusting for the intravenous route, higher AUC and AUC:EC50 were not associated with mortality. CONCLUSIONS:Individual EC50 estimation can provide insights into in vivo host and organism responses. Elevated EC50 showed comparable and unfavorable trends to higher minimum inhibitory concentration. Thus, determining EC50 might help guide individualized target serum voriconazole levels.

Monitoring host- and pathogen-specific biomarkers alongside drug levels at the site of infection offers promise for personalised antimicrobial therapy. Here, we highlight the potential of longitudinally monitoring biomarkers in urine using existing technologies to enable individualised PK-PD optimisation in urinary tract infections. Stadler et al. propose using existing technologies to link urinary biomarkers and antimicrobial drug levels for personalised treatment of urinary tract infections. This approach aims to enable real-time pharmacokinetic-pharmacodynamic monitoring and optimise individual antibiotic dosing.

Abstract Background The impact of antimicrobial resistance (AMR) on death at the patient level remains challenging to estimate. We aimed to characterize AMR-attributable deaths in a large UK centre through death certification. Methods This was a retrospective study of all patients who died in University College London Hospitals NHS Foundation Trust in 2022 (N=758). Clinical records of participants with positive microbiological samples within 28 days of death were reviewed by two independent investigators. AMR-attributable deaths were defined using a newly proposed standardized patient-level definition after consensus agreement between the two investigators. A third investigator acted as a tiebreaker in cases of disagreement. Results Overall, infection was the underlying cause of death for 11.7% (89/758) of participants and was implicated in the pathway that led to death in 41.1% (357/758) of cases. The most common infection syndromes leading to death as documented on the medical certificate of confirmation of death (MCCD) were respiratory tract infections, sepsis and COVID-19. In total, 4.2% (32/758) of all deaths were AMR-attributable. AMR-attributable deaths were more commonly recorded in patients admitted under haematology (7.5%, 9/120) and in younger patients (median age 67 versus 72, P=0.01). The median time from the index sample collection until death was 4.5 days (IQR 2–10.5 days). The majority of AMR-attributable deaths (56.3%, 18/32) were caused by treatment failure and subsequent treatment delay caused by intrinsic resistance mechanisms, primarily by Enterococcus faecium (38.9%, 7/18), Enterobacterales carrying repressed chromosomal AmpCs (27.7%, 5/18) and Pseudomonas aeruginosa (22.2%, 4/18). On the contrary, a minority of AMR-attributable deaths (43.7%, 14/32) were caused by acquired resistance mechanisms, primarily derepressed AmpCs (28.6%, 4/14) and ESBLs (21.4%, 3/14). The median time to effective treatment was 32 h 15 min and did not differ significantly between the two groups. Only 62.5% (20/32) of AMR-attributable deaths as judged by study investigators had infection recorded on the MCCD. AMR was not recorded as a cause of death in any of the patients, including the 14 patients who were deemed to have suffered deaths due to acquired resistance by the study investigators. Conclusions Infection and AMR were significant causes of death in this cohort, yet there significantly underreported during death certification. In a low incidence setting for AMR, delays in treatment and AMR-attributable mortality due to intrinsic resistance mechanisms (expected phenotypes) were more common compared with acquired resistance mechanisms.

Background: The optimization of antimicrobial dosing plays a crucial role in improving the likelihood of achieving therapeutic success while reducing the risks associated with toxicity and antimicrobial resistance. Probenecid has shown significant potential in enhancing the serum exposure of phenoxymethylpenicillin, thereby allowing for lower doses of phenoxymethylpenicillin to achieve similar pharmacokinetic/pharmacodynamic (PK/PD) targets. We developed a triple quadrupole liquid chromatography mass spectrometry (TQ LC/MS) analysis of, phenoxymethylpenicillin, benzylpenicillin and probenecid using benzylpenicillin-d7 and probenecid-d14 as IS in single low-volumes of human serum, with improved limit of quantification to support therapeutic drug monitoring. Methods: Sample clean-up was performed by protein precipitation using acetonitrile. Reverse phase chromatography was performed using TQ LC/MS. The mobile phase consisted of 55% methanol in water + 0.1% formic acid, with a flow rate of 0.4 mL min-1. Antibiotic stability was assessed at different temperatures. Results: Chromatographic separation was achieved within 2 minutes, allowing simultaneous measurement of phenoxymethylpenicillin, benzylpenicillin and probenecid in a single 15 μL blood sample. Validation indicated linearity over the range 0.0015-10 mg L-1, with accuracy of 96-102% and a LLOQ of 0.01 mg L-1. All drugs demonstrated good stability under different storage conditions. Conclusion: The developed method is simple, rapid, accurate and clinically applicable for the quantification of phenoxymethylpenicillin, benzylpenicillin and probenecid in tandem.

Abstract In the face of increasing antimicrobial tolerance and resistance there is a global obligation to optimise oral antimicrobial dosing strategies including narrow spectrum penicillins, such as penicillin-V. We conducted a randomised, crossover study in healthy volunteers to characterise the influence of probenecid on penicillin-V pharmacokinetics and estimate the pharmacodynamics against Streptococcus pneumoniae. Twenty participants took six doses of penicillin-V (250 mg, 500 mg or 750 mg four times daily) with and without probenecid. Total and free concentrations of penicillin-V and probenecid were measured at two timepoints. A pharmacokinetic model was developed, and the probability of target attainment (PTA) calculated. The mean difference (95% CI) between penicillin-V alone and in combination with probenecid for serum total and free penicillin-V concentrations was significantly different at both timepoints (total: 45 min 4.32 (3.20–5.32) mg/L p < 0.001, 180 min 2.2 (1.58–3.25) mg/L p < 0.001; free: 45 min 1.15 (0.88–1.42) mg/L p < 0.001, 180 min 0.5 (0.35–0.76) mg/L p < 0.001). There was no difference between the timepoints in probenecid concentrations. PTA analysis shows probenecid allows a fourfold increase in MIC cover. Addition of probenecid was safe and well tolerated. The data support further research into improved dosing structures for complex outpatient therapy and might also be used to address penicillin supply shortages.

Antimicrobial resistance (AMR) and healthcare associated infections pose a significant threat globally. One key prevention strategy is to follow antimicrobial stewardship practices, in particular, to maximise targeted oral therapy and reduce the use of indwelling vascular devices for intravenous (IV) administration. Appreciating when an individual patient can switch from IV to oral antibiotic treatment is often non-trivial and not standardised. To tackle this problem we created a machine learning model to predict when a patient could switch based on routinely collected clinical parameters. 10,362 unique intensive care unit stays were extracted and two informative feature sets identified. Our best model achieved a mean AUROC of 0.80 (SD 0.01) on the hold-out set while not being biased to individuals protected characteristics. Interpretability methodologies were employed to create clinically useful visual explanations. In summary, our model provides individualised, fair, and interpretable predictions for when a patient could switch from IV-to-oral antibiotic treatment. Prospectively evaluation of safety and efficacy is needed before such technology can be applied clinically.

Beta-lactamase-producing Enterobacteriaceae present a significant therapeutic challenge. Current developments in phenotypic diagnostics focus primarily on rapid minimum inhibitory concentration (MIC) determination. There is a requirement for rapid phenotypic diagnostics to improve antimicrobial susceptibility tests (AST) and aid prescribing decisions. Phenotypic AST are limited in their ability to characterise beta-lactamase-producing Enterobacteriaceae in detail. Despite advances in rapid AST, gaps and opportunities remain for developing additional diagnostic approaches that facilitate personalised antimicrobial prescribing. In this perspective, we highlight the state-of-the-art in beta-lactamase detection, identify gaps in current practice, and discuss barriers for innovation within this field.

Background: Vancomycin is commonly prescribed in late onset sepsis (LOS) in neonatal intensive care (NICU). Despite variation in vancomycin population pharmacokinetics, a paucity of evidence exists to support dose optimisation. This study explored the relationship between trough vancomycin concentrations and estimated area-under-the-concentration-time-curve (AUC) to minimum inhibitory concentration (MIC) ratios in real-world practice. Methods: Patients treated with vancomycin for LOS in two tertiary NICUs between October 2022 and February 2023 were included. Electronic patient record data on demographics, microbiology, dosing, therapeutic drug monitoring (TDM), and outcomes were extracted; these were used to estimate individual patient AUC and AUC:MIC ratios using Bayesian forecasting. Trough and AUC estimates were compared. Target attainment was estimated using an AUC:MIC>400, and toxicity using AUC>600 mg·h/L. Estimates for target attainment were evaluated at different MICs. Results: 32 patients, with 41 discrete treatment episodes, were analysed. Median gestational age at birth was 26.5 (IQR 25-30) weeks. Ten patients (31%) were female and median weight was 0.87 (IQR 0.7-1.4) kg. Trough concentrations correlated poorly with AUC estimates (r 2 =0.38). Dose adjustment using troughs did not improve AUC/MIC target attainment. Acute kidney injury (AKI) occurred in 4/41 (10%) treatment episodes; peak median AUC was 1170.4 (IQR 839.1-1493.7) mg·h/L compared to 582.1 (IQR 485.4-699.3) mg·h/L in those without AKI. For individual episodes, AUC/MIC targets at day 2 would be met for vancomycin in 30/41 (73%) for organisms with an MIC of 1 mg/L, 1/41 (2%) for MIC 2 mg/L, and 0/41 (0%) for MIC 4 mg/L. Conclusion: Using trough based TDM correlated poorly with AUC-based estimates for target attainment. Dose adjustment using trough-based TDM fails to improve drug-exposure, especially with MIC >1mg/L.