Katherine E. Goodman

Contact precaution policies are used to prevent the spread of pathogenic organisms. We aimed to test whether AI-assisted cameras could monitor aspects of compliance with these policies. Testing in both simulated and real patient care settings yielded exceptional sensitivity and good specificity, indicating potential to monitor adherence to contact precautions.

Background: Central line–associated blood stream infection (CLABSI) surveillance is mandated and publicly reported in United States hospitals but requires manual chart review. Generative artificial intelligence (AI) may facilitate CLABSI identification. Methods: We performed a retrospective cohort study of CLABSI surveillance using standardized prompts, clinical data, and Centers for Disease Control and Prevention definitions. In 11 hospitals, 24 infection control nurses and physicians retrospectively reviewed 220 CLABSI/non-CLABSI bacteremias. Three methods of review were compared to original facility expert review: (1) AI-assist review, (2) AI-alone review, and (3) repeat expert review. Disagreement between any method of review was adjudicated by a 2-expert physician referee panel. Results: AI-assisted review had a 93.2% (205/220; 95% confidence interval [CI], 89.0–96.1) overall accuracy (agreement with facility reporting or refereeing panel) versus AI-alone review, 90.0% (198/220; 95% CI, 85.3–93.6) overall accuracy or repeat expert review, 88.2% (97/110; 95% CI, 80.1–93.6) overall accuracy. Inter-site variability was greater for repeat expert review than AI-assist. AI-assisted review required a median of 14 minutes (interquartile range 6–25.5) versus 25 minutes (interquartile range 11–43) for repeat expert (P = .0001). AI-assisted review was reported as low effort in 52% of cases versus 21% of repeat expert cases (P < .0001). AI-assisted review was reported as somewhat or very objective in 72% of cases versus 61% of repeat expert cases (P = .092). Conclusions: CLABSI detection by AI methods was at least as accurate as traditional methods, better liked by experts, faster and less variable. AI-assisted review could improve healthcare-associated infection reporting.

Generative artificial intelligence (GenAI) is becoming widespread in society but has had limited impact on medicine. GenAI can rapidly digest large documents, making it particularly well suited for healthcare-associated infection (HAI) surveillance that is largely based on clinical notes. In this article, we review evidence on the application of GenAI for HAI surveillance and highlight the necessary steps for changes to practice. Small studies of simulated patients and retrospective studies of patient records have found GenAI can isolate individual criteria for HAI and make HAI determinations with accuracy similar to that of human experts. GenAI tools are being developed to assist in HAI detection. Use of GenAI for HAI review could reduce variability between facilities, improve efficiency of HAI review. and lead to more time for direct prevention work. For widespread adoption, GenAI HAI detection must be safe and effective in prospective clinical studies.

Among all 2024 emergency department visits for acute respiratory illness or conjunctivitis across a statewide healthcare system (n = 13 494), generative artificial intelligence–based surveillance with adjunctive human review rapidly and cost-effectively identified patients with potential avian influenza exposure. Generative artificial intelligence–based approaches may offer a promising opportunity for expanding public health surveillance in clinical settings.

Importance: An estimated half of all long-term care facility (LTCF) residents are colonized with antimicrobial-resistant organisms, and early identification of these patients on admission to acute care hospitals is a core strategy for preventing intrahospital spread. However, because LTCF exposure is not reliably captured in structured electronic health record data, LTCF-exposed patients routinely go undetected. Large language models (LLMs) offer a promising, but untested, opportunity for extracting this information from patient admission histories.Objective: To evaluate the performance of an LLM against human review for identifying recent LTCF exposure from identifiable patient admission histories.Design, Setting, and Participants: This cross-sectional, multicenter study used the history and physical (H&P) notes from unique, randomly sampled adult admissions occurring between January 1, 2016, and December 31, 2021, at 13 hospitals in the University of Maryland Medical System (UMMS) and the John Hopkins (Hopkins) health care system to compare the performance of an LLM (GPT-4-Turbo) using zero-shot learning and prompting against humans in identifying patients with recent LTCF exposure. LLM analyses were conducted from August to September 2024.Exposure: Recent (≤12 months) LTCF exposure documented in the H&P note, as adjudicated by (1) humans and (2) an LLM.Main Outcomes and Measures: LLM sensitivity and specificity with Clopper-Pearson 95% CIs. Secondary outcomes were note review time and cost. The LLM was also prompted to provide a rationale and supporting note-text for each classification.Results: The study included 359 601 eligible adult admissions, of which 2087 randomly sampled H&P notes were manually reviewed at UMMS (1020 individuals; median [IQR] age, 58 [41-71] years; 493 [48%] male) and Hopkins (1067 individuals; median [IQR] age, 58 [48-67] years; 561 [53%] male) for LTCF residence. Compared with human review, the LLM achieved a sensitivity of 97% (95% CI, 91%-100%) and a specificity of 98% (95% CI, 97%-99%) at UMMS, and 96% (95% CI, 86%-100%) and 93% (95% CI, 92%-95%) sensitivity and specificity, respectively, at Hopkins; specificity at Hopkins improved with prompt revision (96% [95% CI, 95%-97%]). Of 117 manually reviewed LLM rationales, all were factually correct and quoted note-text accurately, and some demonstrated inferential logic and external knowledge. The LLM identified 37 (1.8%) human errors. Human review time had a mean of 2.5 minutes and cost $0.63 to $0.83 per note vs a mean of 4 to 6 seconds and $0.03 per note for LLM review.Conclusions and Relevance: In this 13-hospital study of 2087 adult admissions, an LLM accurately identified LTCF residence from H&P notes and was more than 25 times faster and 20 times less expensive than human review.

Background: In recent years, new broad-spectrum antibiotics targeting Gram-negative organisms have been introduced, including cefiderocol, ceftazidime-avibactam, ceftolozane-tazobactam, eravacycline, imipenem-relebactam, omadacycline, and meropenem-vaborbactam. This study aimed to describe new antibiotic use across a large national cohort. Methods: We performed a retrospective cohort study of hospital discharges from June 2022 to May 2023 using the Premier Healthcare Database. Antibiotic utilization was ascertained from daily charges. Clinical indication(s) were inferred from International Classification of Diseases, 10th revision, diagnosis codes. Antibiotic therapy was considered definitive if continued >3 days. Piperacillin-tazobactam was used as a comparator. Results: Across 832 hospitals, 3 890 557 admissions (61.9% of all admissions) included an antibiotic prescription. New antibiotics were prescribed in 9768 admissions (0.25% of antibiotic-prescribing admissions) across 537 hospitals. Ceftolozane-tazobactam was prescribed in 4157 admissions (42.6% of 9768), ceftazidime-avibactam in 3660 (37.5%), eravacycline in 1213 (12.4%), cefiderocol in 1060 (10.9%), meropenem-vaborbactam in 456 (4.7%), omadacycline in 104 (1.1%), and imipenem-relebactam in 99 (1.0%). In contrast, piperacillin-tazobactam was prescribed in 731 719 (18.8%) and colistin in 570 (0.01%) admissions. Forty-six percent (n = 4647/9768) of new antibiotics were started in the first 3 days of hospital admission, and 70% (n = 6799/9768) were used as definitive therapy. Sepsis (76%), pneumonia (46%), and urinary tract infection (39%) were the most common clinical indications. On average, patients treated with new antibiotics had 8 more comorbid conditions than patients receiving piperacillin-tazobactam. Conclusions: Ceftazidime-avibactam and ceftolozane-tazobactam remain the most frequently prescribed new antibiotics, with uptake of subsequently approved agents trailing. New antibiotics are most commonly used as treatment for sepsis among patients with multiple comorbidities.

Background: Clinicians often start unnecessarily broad-spectrum empiric gram-negative antibiotics out of the concern that delaying effective therapy could lead to a worse clinical outcome. This study examined the consequences of delayed initiation of broad-spectrum gram-negative antibiotics. Methods: In a retrospective cohort of adult inpatients from 928 US hospitals, we compared clinical outcomes after (1) empiric narrow-spectrum antibiotics escalated to broad-spectrum antibiotics (delayed broad-spectrum therapy [DBT]) and (2) empiric broad-spectrum antibiotics continued as post-empiric therapy (early broad-spectrum therapy [EBT]) using Win Ratios. DBT and EBT patients were matched on hospital, admitting diagnosis, and propensity scores incorporating 28 clinical variables. The outcome of interest was a ranked composite of mortality, readmission, and adverse drug events. Results: Out of 746 880 inpatients, 82 276 (11%) received DBT and 664 604 (89.0%) received EBT. Among the 67 046 with DBT who were matched to 67 046 with EBT, mortality was 8.7% after DBT and 9.5% after EBT (P = .022), readmission was 10.5% after DBT and 11.8% after EBT (P < .0001), and the rate of adverse drug events was 8.4% after DBT and 7.2% after EBT (P < .0001). Among matched patients, clinical outcomes were superior after DBT compared with EBT (win-ratio 1.06; P < .0001). Conclusions: On average, among a large sample of adult inpatients who ultimately received broad-spectrum antibiotic therapy, delaying initiation of a broad-spectrum antibiotic was not associated with worse outcomes. Although broad-spectrum empiric therapy is undoubtedly sometimes warranted, this finding challenges the common belief that is it safer to err towards overly broad-spectrum empiric antibiotic therapy.