To evaluate the ability of large language models (LLMs) with targeted feedback to classify medications as antimicrobial or non-antimicrobial and their implication in antimicrobial stewardship.

Cross-sectional evaluation using a two-phase process: initial unguided classification and feedback-informed reclassification.

Medication-level analysis of health system prescribing data.

A data set of 7,239 unique medication entries from health systems in the Collaboration to Harmonize Antimicrobial Registry Measures (CHARM) project.

Four LLMs, including ChatGPT-3.5, Copilot GPT-4o, Claude Sonnet 4, and Gemini 2.5 Flash, classified all entries against a manual reference standard. Models then received feedback on 20% of misclassified cases for reclassification. Metrics included accuracy, macro F1-score (95% confidence intervals via bootstrap resampling), positive predictive value, negative predictive value, processing time, and error reduction rates (ERRs). McNemar’s test assessed accuracy between phases and model differences.

Baseline accuracy varied among LLMs. Postfeedback, all accuracies improved significantly (P < .001) to Gemini (99.6%), Claude Sonnet 4 (99.4%), ChatGPT-3.5 (81.0%), and Copilot (79.7%). Gemini achieved the highest macro-F1 (98.9%, 95% CI: 98.4–99.3) and ERR (69.2%). Processing time were fastest for Copilot (42 s), followed by ChatGPT-3.5 (47 s), Gemini (1,080 s), and Claude (3,475 s). Manual classification of this task is estimated to take 18 hours without LLM. Misclassification was most common among antiseptics, antiparasitics, and drugs with antimicrobial components for non-infectious uses (eg, sulfasalazine).

Top-performing LLMs achieved accuracy levels suitable for automating initial antimicrobial classification in stewardship workflows. Performance variability underscores the need for careful selection and continued human oversight in clinical applications.

Four LLMs were evaluated for antimicrobial classification using 7,239 medications. Claude Sonnet 4 and Gemini achieved>99% accuracy, while ChatGPT-3.5 and Copilot showed substantial limitations. Top performers could automate stewardship workflows with appropriate oversight.

Evaluating the COVID-19 pandemic restrictions and their effect on rates of new respiratory viral infections in patients undergoing hematopoietic stem cell transplantation (HSCT), and if there were changes to the morbidity and mortality rates compared to historical controls.

Retrospective chart review.

University-based tertiary care center.

All patients who underwent HSCT, including chimeric antigen receptor T-cell therapy, between 7/1/2013 and 12/19/2021 were included in the study. No eligible patients were excluded.

We found a significant difference in new respiratory infections as measured by a respiratory viral pathogen (RVP) polymerase chain reaction-based assay during transplant admissions between the pre- and early COVID-19 eras, with complete absence of new respiratory viral infections in the early COVID-19 era. The late-COVID-19 era, coincident with availability of severe acute respiratory syndrome coronavirus 2 vaccines and relaxation of some hospital-based restrictions, had a similar incidence of new RVP positivity compared to the pre-COVID-19 era. For allogeneic HSCT outcomes alone, rates of pediatric intensive care unit admission (66.7% vs 32.2%, p = .01), intubation (57.1% and 18.6%, p < .01), and oxygen requirement (66.7% vs 40.7%, p = .04) were found to be statistically different with new RVP positivity. Other outcomes examined, including death at 100/180/365 days post-transplant, length of stay, and acute graft-versus-host disease incidence, were similar, regardless of new RVP positivity.

In order to reduce morbidity, these findings argue for a continuation of the strict protective isolation practices initially employed during the pandemic for HSCT patients, particularly during viral emergence following the COVID-19 lockdown.

Methicillin-resistant Staphylococcus aureus (MRSA) is a pathogen commonly implicated in pneumonia for which vancomycin is often used to cover empirically. Our study examined the impact of a pharmacy-based automatic MRSA polymerase chain reaction (PCR) nares intervention.

We included patients prescribed vancomycin for pneumonia and defined three cohorts: pre-intervention (Cohort 1), post-test implementation (Cohort 2), and post-test implementation with electronic medical record (EMR) enhancements and active stewardship involvement (Cohort 3). The primary outcome was total duration of vancomycin treatment. We performed a desirability of outcome ranking with response adjusted for duration of antibiotic risk (DOOR/RADAR) analysis to prioritize clinical outcomes including adverse events, escalation of care, and death.

The median duration of vancomycin therapy decreased between Cohorts 1 and 2 (55 versus 47 hours, p = 0.15), with a greater decrease detected among the subset of Cohort 2 patients from whom the PCR test was actually collected (42 hours, p = 0.035). Duration of therapy was lowest in Cohort 3 (22.3 hours, p < 0.001). In the DOOR/RADAR analysis, no meaningful difference in clinical outcomes was detected between Cohorts 1 and 2 (p = 0.55). Cohort 3 had substantially better ranked outcomes when compared to Cohorts 1 and 2 (p < 0.001).

Vancomycin treatment duration was shorter in patients after a pharmacy-directed, EMR-based MRSA PCR intervention was implemented, with a greater response after implementation of additional EMR enhancements and active stewardship involvement, improving test collection rate and result response. This study will help inform future interventions pairing automated EMR-based interventions with stewardship team efforts.

Concern for Pseudomonas aeruginosa in diabetic foot infections (DFIs) is common and a driving factor for empiric antibiotic prescribing, particularly among immunocompromised hosts. This study sought to identify the prevalence of Pseudomonas aeruginosa in DFIs among kidney transplant recipients and the use of antipseudomonal agents for empiric antibiotic coverage.

Retrospective cohort study.

Large, urban, tertiary care transplant center.

39 kidney transplant recipients who developed DFIs at our institution between 2015–2023.

We performed a retrospective review of 47 DFIs in 39 kidney transplant recipients hospitalized for a DFI with a positive tissue culture.

A total of 79 organisms were isolated. Pseudomonas represented 6.33% of organisms (5/79) and was present in 5/47 DFIs (10.6%). Staphylococcus aureus was the most commonly isolated organism (30.4%, 24/79). Compared to those with non-Pseudomonas infections, patients with cultures positive for Pseudomonas were temporally closer to transplant (50.6 days +/- 214.9 vs 1047.6 days +/- 766.4, p = 0.041) and more likely to have been treated for rejection with the past 12 months (2/5, 40% vs 0/42, 0%, p = 0.009). Eighty three percent of patients received empiric antibiotic coverage with an antipseudomonal agent.

Our study found a low overall prevalence of Pseudomonas in DFIs among kidney transplant recipients. Despite this, empiric coverage with antipseudomonal antibiotics was common, highlighting a potential antimicrobial stewardship target in solid organ transplantation.

To investigate the epidemiology and clinical outcomes of Candida auris (C. auris), reclassified Candidozyma auris, in a Detroit tertiary care center, focusing on mortality, length of stay (LOS), glucose variability, and demographic factors.

Inpatient admissions in a Detroit tertiary care center, June 2023–March 2025.

Among 9,025 C. auris tests, 725 (8.03%) were positive. A subpopulation of 242 hospitalized patients was analyzed for outcomes.

Clinical and demographic data, including infection status, mortality, and LOS, were analyzed. Glucose variability was assessed by the coefficient of variation. ANOVA and Kruskal–Wallis tests evaluated group differences, pairwise Z-tests compared mortality, and a competing-risks survival analysis estimated probabilities of discharge and in-hospital death.

Mortality was highest among colonized/infected patients (31.3%) and significantly associated with race (p = 0.0072) and ethnicity (p = 0.0069). High glucose variability (>190 mg/dL) was linked to increased mortality (27.63%) and prolonged LOS (46.14 days; p < 0.01). Infected patients had the longest LOS (67.9 days), followed by colonized/infected (39.9) and colonized-only (30.2; p < 0.001).

C. auris disproportionately affects African American patients and those with high glucose variability, contributing to higher mortality and longer hospitalization. Infection status and glycemic instability were the strongest predictors of LOS, while racial and ethnic disparities influenced mortality. Strengthened infection control, early identification, and optimized glucose management may improve outcomes in high-risk populations.

This quality improvement project aims to contribute to the body of knowledge on effective and sustainable surgical site infection (SSI) prevention strategies in a teaching community hospital setting.

Retrospective and prospective chart review.

305-bed acute care urban teaching hospital.

All abdominal hysterectomies were included in the data analysis from January 1, 2018 to June 30, 2025.

The project consisted of two time periods: baseline and monitoring periods. During the baseline period, an SSI prevention bundle was in place, but no individual bundle elements were monitored. At the end of the baseline period, due to rising SSI rates after abdominal hysterectomy, a multi-disciplinary workgroup was formed to evaluate bundle complexity and identify deficiencies in practice. Staff were re-educated and audit tools were created. In the monitoring period, compliance with individual bundle elements was measured and results were disseminated to stakeholders.

There were 782 abdominal hysterectomy procedures in the baseline period and 497 in the monitoring period. 13 patients had an SSI in the baseline period versus one patient in the monitoring phase (p = 0.003). Compared with the baseline, the SSI rate decreased significantly during the monitoring period (0.2 SSI versus 1.7 per 100 abdominal hysterectomies, p = 0.002).

This project demonstrated significant improvements in outcome (SSI rates) when care processes were regularly monitored, and compliance was shared with key players. Future quality improvement projects will be focused on prioritizing the most important care processes that need to be continuously monitored.

To evaluate the clinical and implementation outcomes of an antimicrobial stewardship program (ASP) intervention to improve antibiotic therapy for a rapid diagnostic test (RDT) for bloodstream infections (BSIs).

Retrospective pre and postintervention study.

Single pediatric tertiary center from August 2022 to May 2024.

Patients presenting with a positive blood culture accompanied by the Verigene (VG) Gram-positive Blood Culture Nucleic Acid (BC-GP) assay. Implementation outcomes surveys were completed by infectious diseases (ID) clinicians and non-ID clinicians.

We implemented a 24 hours a day, 7 days a week (24/7) ASP intervention to improve response to BC-GP assay results. The primary clinical end point was time to optimal antimicrobial therapy (OAT). Secondary endpoints included duration of bacteremia, hospital length of stay (LOS), and mortality. We assessed our intervention’s acceptability, appropriateness, and feasibility using validated implementation outcomes surveys.

Among 211 pre and 91 postimplementation BC-GP results, the median time to OAT decreased from 20.3 hours (95% CI:14.2–26.4) to 1.3 hours (95% CI: 0.0–2.7), p = .002. No significant differences were found in duration of bacteremia, LOS, or mortality. Implementation surveys from 23 ID and 47 non-ID clinicians demonstrated over 80% agreement on intervention appropriateness and acceptability. ID clinicians rated feasibility measures lower than non-ID clinicians (3.58 vs 4.51 on a five-point Likert scale, p < .001).

24/7 ASP intervention paired with RDTs for Gram-positive BSIs is associated with reduced time to OAT. Feasibility differences between ID and non-ID clinicians highlight implementation challenges and the need for future strategy evaluation.

To examine how artificial intelligence (AI) has been applied to infection prevention and control in healthcare, identify barriers and risks affecting implementation, and develop a structured checklist to support safe adoption.

Scoping review conducted in line with Joanna Briggs Institute methodology and reported according to PRISMA-ScR.

PubMed, Scopus, and Web of Science were searched for primary studies (2014–2024) describing real-world AI applications for IPC. Studies reporting implementation experiences, outcomes, or risks were included. Data on study design, AI type, IPC function, integration level, barriers, and outcomes were extracted and synthesized thematically to derive a 41-item decision-support checklist.

Of 2,143 records screened, 100 studies met inclusion. Most were published since 2022, with the United States and China leading output. Machine learning dominated (75%), mainly for predictive analytics (53%), HAI detection (13%), and hand hygiene monitoring (13%). Only 15% of tools were integrated into existing digital infrastructures. Barriers centred on data quality (45%), technical and data related (16%), and economic/technical constraints (16%). Reported risks clustered around operational failures (35%), technical errors (33%), and data security (12%). Evidence was heavily skewed toward high-income countries, with limited prospective validation or implementation science.

AI offers clear promise for IPC, particularly in early detection and compliance monitoring, but its translation into practice remains constrained by data fragmentation, limited integration, and uneven readiness across settings. Our evidence-informed checklist provides IPC teams with a structured tool to assess feasibility, governance, and resource needs before adoption, supporting safer and sustainable innovation.

Clostridioides difficile is the primary cause of healthcare-associated infectious diarrhea in hospitalized patients. The most common laboratory testing methods for C. difficile infection (CDI) are toxin detection via enzyme immunoassay (EIA) and polymerase chain reaction (PCR), which detect a toxogenic strain. This study examines the impact of Rhode Island’s largest hospital system changing from PCR-only to two-step CDI testing.

A retrospective cohort study of 2,173 adult inpatients was conducted. Patients were grouped into two cohorts: those tested for toxigenic C. difficile via PCR-only (June 2019–May 2021, n = 1,194) and those tested with the two-step algorithm (June 2021–May 2023, n = 979). Cluster analysis identified patient risk groups for hypothesis generation, and complications such as death, colectomy, intensive care unit ICU transfer, and 30-day readmission were compared across these groups.

In the moderate-risk group, there was a significant reduction in ICU transfers and readmission rates with the two-step testing by 5% and 7%, respectively. There were no other significant differences in complications between testing groups. Anti-CDI antibiotics were discontinued in 15% (n = 106) of EIA-negative patients in the two-step testing group. Moderate-risk patients were less likely to have treatment discontinued than severe-risk patients (OR = 2.00, p = 0.016).

The two-step testing algorithm did not negatively affect patient outcomes and led to a modest decrease in anti-CDI treatment, supporting the safety of two-step CDI testing in hospitalized patients.

During the severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) pandemic, recommended infection prevention practices for preventing transmission in acute healthcare settings included, at a minimum, universal source control with masking and eye protection within six feet of others and using a respirator when caring for individuals with COVID-19.

A mixed methods study consisting of observations and self-reported infection prevention behaviors among critical care nurses identified high adherence to mask use in a medical intensive care unit (ICU) during the COVID-19 pandemic.

Surveyed nurses reported more barriers to proper use eye protection than with masking. We observed nurses wearing required eye-protection only 20% of the time when within 6 feet of others. Use of eye protection increased in the presence of patients while decreasing near other healthcare workers. In self-reported assessments, these nurses described decreased protective behaviors at work and at home after vaccination for COVID-19. However, self-reported masking in public remained high and was not impacted by vaccination status. Finally, nurses most frequently perceived high transmission risk of SARS-CoV-2 in the community and patient care settings, with lowest risk from co-workers and household members.

Perceived risks of exposure to SARS-CoV-2 likely impact infection prevention behaviors. Differences in perceived risk between patient and peer groups provide insight into strategies for improving infection prevention behaviors in both pandemic and endemic multidrug-resistant organism settings.

The presence of insects in hospital environments poses a potential risk for the dissemination of pathogenic bacteria, including multidrug-resistant species. Despite strict sanitation protocols, some arthropod populations persist in less regulated areas, potentially acting as mechanical vectors of bacterial contamination.

This study aimed to analyze the bacterial diversity associated with insects collected in hospital settings and assess their potential role in spreading pathogens relevant to public health.

A descriptive observational approach was employed to identify and classify bacterial taxa associated with hospital-collected insects. High-throughput sequencing was used for taxonomic classification at the phylum, family, and genus levels.

Proteobacteria was the predominant phylum, additionally we found families such as Moraxellaceae and Mycobacteriaceae, known to include clinically relevant species. The genera Acinetobacter and Mycobacterium were particularly abundant in some samples, raising concerns about their potential role in nosocomial infections. Other identified bacteria included Pseudomonas, Escherichia, and Shigella, albeit at lower frequencies. The persistence of these bacteria in hospital environments suggests that insects may contribute to their dissemination.

The findings highlight the need for enhanced arthropod control measures in healthcare facilities as part of routine biosecurity protocols. The presence of multidrug-resistant bacteria associated with hospital-dwelling insects reinforces their role in pathogen transmission, emphasizing the importance of comprehensive vector management strategies to mitigate nosocomial infection risks.