We investigated the accuracy and completeness of four large language model (LLM) artificial intelligence tools. Most LLMs provided acceptable answers to commonly asked infection prevention questions (accuracy 98.9%, completeness 94.6%). The use of LLMs to supplement infection prevention consults should be further explored.

We evaluated one of the first secure large language models approved for protected health information, for identifying central line-associated bloodstream infections (CLABSIs) using real clinical notes. Despite no pretraining, the model demonstrated rapid assessment and high sensitivity for CLABSI identification. Performance would improve with access to more patient data.

Introduction: Social determinants of health can impact healthcare-associated infections. Hospital-onset bacteremia (HOB) may lead to poor outcomes, increased length of stay, and increased cost of care. We explored the association of social determinants of health and HOB. Methods: We retrieved blood culture data at Stanford Health Care from May 2019 to October 2023. We identified blood cultures ordered ≥4 days of admission. To evaluate the association between social determinants of health and HOB, we employed an unsupervised machine learning approach (K-Means clustering) to discern patterns in HOB rates based on the Social Vulnerability Index (SVI). The SVI indicates the relative vulnerability of every U.S. Census tract. It ranks the tracts on 16 measures of vulnerability across 4 themes: socioeconomic factors, household characteristics, racial and ethnic minority status, and housing/transportation aspects. The number of clusters was determined using the Elbow Method (Figure 1). Results: Out of 209,947 blood cultures from 23,938 unique patients with a California address, we identified 81,653 blood cultures collected after 4 days (40%). The K-Means clustering algorithm identified 3 distinct clusters within the Californian census tracts, suggesting heterogeneity in the relationship between SVI and HOB (Figure 2). Cluster 1 had a higher SVI (median 0.73, range 0.46 – 0.99), with logistic regression indicating a positive SVI-HOB association (OR 4.84, 95% CI 4.02 – 4.81, p <.001). Cluster 2, had a median SVI of 0.29 (range 0.0009 – 0.78), also showed a positive association between SVI and HOB (OR 1.67, 95% CI 1.4 – 1.89, p <.001), aligning with trends of higher infection risks in more vulnerable groups. In contrast, Cluster 3 had a median SVI of 0.22 (range 0.002 –0.84). In this cluster, the SVI showed a negative association with HOB (OR 0.24, 95% CI 0.18 – 0.31, p <.001). Cluster 3 was the cluster with the least number of subjects (15,000, versus 21,761 for Cluster 1 and 29,762 for Cluster 2). Most subjects in Cluster 3 resided in Santa Clara County, whereas those in Clusters 1 and 2 were spread across Santa Clara, San Mateo, Alameda, Merced, and Sacramento Counties (Figure 3). Conclusions: Advanced techniques can be used to explore the complex interplay between social determinants of health and healthcare-associated infections and could guide the development of community-specific strategies to improve outcomes.

Background: Artificial intelligence (AI) tools have demonstrated success in US medical licensing examinations; however, their utility in infection prevention and control (IPC) remains unknown. Methods: The program of hospital epidemiology handles consultation calls and records each question and answer. Using 2022 data, we selected 31 frequently asked questions. We utilized four AI tools, including Chat GPT-3.5 and 4.0, Bing AI, and OpenEvidence, to generate answers. We predefined scales (Table 1) to capture responses by three reviewers, including two hospital epidemiologists and one infection preventionist. The mean score of ≥ 3 and ≥ 4 was considered acceptable in accuracy and completeness, respectively. We reported the percentage of responses with acceptable accuracy and completeness out of assessed questions for each category. Results: Among 31 questions, 16 were associated with isolation duration, 9 with healthcare personnel (HCP) exposure, 4 with cleaning contaminated rooms, and 2 with patient exposure. Regarding accuracy, most AI tools performed worse in questions about isolation duration, ranging between 75% and 93.8%. All AI tools, except OpenEvidence, had a 100% accuracy rate for HCP and patient exposure. All AI tools had a 100% accuracy rate for contaminated room handling. The highest overall acceptable accuracy rate was observed in Chat GPT-3.5. Regarding completeness, most AI tools performed worse in questions about isolation duration, ranging between 44% and 75%. All AI tools, except OpenEvidence, had a 100% completeness rate for contaminated rooms and patient exposure. The highest overall acceptable completeness rate was observed in Bing AI (Table 2). Conclusions: All AI tools provided reasonable answers to commonly asked IPC-related questions, although, there were variations among different tools used. AI could be used to supplement the infection control program, especially if resources are limited.

Introduction: Many central line-associated bloodstream infections are identified in patients nearing the end of life. Stanford Health Care recently introduced the General Inpatient Hospice program. This program offers inpatient hospice care for patients who, due to uncontrolled symptoms, cannot be discharged to a hospice facility or receive home hospice care. We investigated whether this program would impact blood cultures practices near the time of death. Methods: We performed a retrospective cohort study at Stanford Health Care using records of blood culture events from May 2019 to October 2023. We defined a blood culture near-death as those collected within 2 days before the date of death. We performed an interrupted time series linear regression before and after the implementation of the General Inpatient Hospice program on July 1, 2022 to assess blood culture intensity near-death. Blood culture intensity was defined as the proportion of cultures collected near-death in relation to the total number of blood cultures. Additionally, we calculated blood culture positivity rate, which was defined as the proportion of positive blood cultures among all those collected during our study period. Results: Out of 220,269 blood cultures from 24,955 unique patients, a total of 6,147 cultures (9%) were obtained near the time of death. Among these subjects, the median age was 65 years (range 20–102), with 43% identifying as being of White race-ethnicity and 57% as male. Of these cultures, 3044 were positive (49.5%), with Escherichia coli (618, 24%), Klebsiella pneumoniae (341, 13%), and Staphylococcus aureus (166, 10%) being the most common organisms. After the implementation of the General Inpatient Hospice program, the median enrollment was 12 patients (range 3–18) and the median mortality rate was 2.3% (range 2–3%). The blood culture intensity near death decreased by 0.81%, a change that was not statistically significant (95% CI -2.4% to 0.8%, p=.32; Figure 1). Subsequently, the blood culture intensity showed a non-significant increasing trend of 0.05% (95% CI -0.1% to 0.2%, p=0.53). The blood culture positivity rate near the time of death increased by 16% following the intervention, but this increase was not statistically significant (95% CI – 11.8% to 43.3%, p=.26; Figure 2), and it was followed by a non-significant downtrend of 1.9% (95% CI -3.9% to 1.4%, p=.36). Conclusion: We found no significant association between the implementation of an inpatient hospice program and blood culture practices near the time of death, likely due to low patient enrollment.

Introduction: There is a need to optimize blood culture (BC) utilization among hospitalized adults. A previous study showed an evidence-based BC algorithm improved BC utilization in a medical intensive care unit (MICU) at a large academic center. Our aim is to evaluate the impact of an intervention based on this algorithm on blood culture utilization in the MICU at Stanford Health Care, a referral center known for its high volume of solid organ transplants. Methods: We conducted a before-after study evaluating the impact of a BC diagnostic stewardship program in the MICU at Stanford Health Care, a 20-bed unit with an average of 20% lung transplant patients per day. All patients ≥18 years of age admitted to the unit during the study period were included. We adopted a previously published evidence-based algorithm detailing syndromes with low and high risk for bacteremia, which was referenced during patient rounds. Additionally, education and feedback to providers about BC utilization and indication inappropriateness was performed during leadership meetings every other month. We performed an interrupted time series analysis using historical data 17 months before the implementation of the blood culture stewardship program on April 1, 2022 and 8 months after. We assessed changes on BC utilization adjusted to patient days and crude mortality rate during the same period as a balancing measure. Results: Before the implementation of the program, the median BC utilization was 216 per 1,000 patient days (range 150–250 per 1,000 patient days). Following the introduction of the program, there was a significant decrease in blood culture utilization by 5% (95% CI -10% to -1%, p=.02; Figure 1). Post-intervention, the blood culture stabilized, with no significant increase observed (0.09% increase, 95% CI -0.7% to 1%, p=.81). The mortality rate, prior to the implementation of the program, had shown a significant downward trend of 1.4% over time (95% CI -3% to -0.2%, p=.02). After the intervention, a nonsignificant decrease of 8% was observed (95% CI -44% to 27%, p=.62; Figure 2), followed by a nonsignificant upward trend of 6% (95% CI -1% to 13%, p=.10). Conclusion: We observed a significant reduction in BC utilization after implementing the BC diagnostic stewardship program in a MICU frequented by a high number of lung transplant patients.

We assessed the impact of metagenomic next-generation sequencing (mNGS) on patient care using previously established criteria. Among 37 patients receiving mNGS testing, 16% showed results that had a positive clinical impact. While mNGS results may offer valuable supplementary information, results should be interpreted within the broader clinical context and evaluation.

Background: Many hospitals have implemented admission SARS-CoV-2 testing to evaluate for the need for transmission-based precautions. However, a positive test in an asymptomatic patient may represent (1) active infection, signifying infectiousness; (2) false positivity; or (3) past infection with prolonged viral shedding. We used a strand-specific SARS-CoV-2 reverse real-time polymerase chain reaction (rRT-PCR) assay to assess infectivity among asymptomatic patients with a positive SARS-CoV-2 PCR admission test. Methods: We used a 2-step rRT-PCR specific to the minus strand of the SARS-CoV-2 envelope gene. We reviewed records of patients with a positive SARS-CoV-2 PCR who were also tested for the strand-specific SARS-CoV-2 PCR within 2 days of admission at Stanford Health Care during July 2020–April 2022. We restricted our analysis to each patient’s first test. We calculated the percentage of detectable minus strand-specific tests among asymptomatic patients over time and gathered descriptive statistics for age, sex, and immunocompromised state. Results: In total, 848 admitted patients had strand-specific SARS-CoV-2 assays performed. Of 532 patients with a strand-specific assay done within 2 days of admission, 242 (45%) were asymptomatic. Among asymptomatic patients, the mean age was 56 years (range, 19–99), 133 (55%) were male, 50 (21%) had immunocompromising conditions, and 30 (12%) were admitted for a surgical procedure. In total, 21 (9%; range, 4%–25% per quarter) had detectable minus strand-specific assays (Fig. 1). Conclusions: Most asymptomatic patients tested for SARS-CoV-2 on admission were not infectious. Hospitals using SARS-CoV-2 PCR admission testing may need to re-evaluate the continued use of this practice.

Fig. 1. Minus strand-specific SARS-CoV-2 assay percentage positivity per quarter among asymptomatic patients tested within 2 days of admission. The peak positivity in November 2021–January 2022 quarter coincided with the SARS-CoV-2 omicron variant surge in our county.

Disclosure: None

Background: SARS-CoV-2 viral load decreases over time after illness onset. However, immunocompromised patients may take longer for viral load decrease or have a more erratic viral-load trajectory. We used strand-specific assay data from admitted patients to evaluate viral-load trajectories after illness onset. Methods: We reviewed records of hospitalized patients with a positive SARS-CoV-2 PCR and tested using the strand-specific SARS-CoV-2 PCR during July 2020–April 2022. At Stanford Healthcare, we use a 2-step reverse real-time polymerase chain reaction (rRT-PCR) assay specific to the minus strand of the SARS-CoV-2 envelope gene to assess infectivity. Restricting our analysis to each patient’s first strand-specific assay, we used logistic regression models to compare patients with single versus multiple assays. Among patients with multiple tests, we compared those who had an upward trajectory in cycle threshold (Ct) values (a surrogate of decreasing viral load) versus those who did not. We analyzed presence of symptoms, immunocompromised state, immunosuppression reason, and severe COVID-19 leading to ICU care in univariate and multivariate models that further adjust for additional covariates. Significant differences were assessed using logistic regression odds ratios and an α level of 0.05. Results: In total, 848 inpatients were included. Among them, 703 were tested only once and 145 were tested 2–6 times. The longest duration of minus-strand detection was 163 days. In univariate analyses, patients with a single minus-strand assay had lower odds of being symptomatic (OR, 0.55), of being immunocompromised (OR, 0.58), and of being admitted to the ICU with severe COVID-19 (OR, 0.49). In the multivariate analysis, being admitted to the ICU with severe COVID-19 was the only significant variable associated with having >1 test (OR, 2.44). Among patients who had multiple strand-specific SARS-CoV-2 assays, 119 had upward minus-strand trends of Ct values (as expected) and 26 did not. Being immunocompromised was associated with nonrising minus-strand CT values (OR, 33.3) when holding all other covariates in the model constant. Conclusions: Immunocompromised patients with COVID-19 tend to actively replicate for longer and have unexpected viral trajectories compared to immunocompetent patients. Among immunocompromised patients, suspension of transmission-based precautions may require a case-by-case evaluation.

Disclosures: None

Background: Mitigating the risk of nosocomial respiratory disease transmission in the healthcare facilities of low- and middle−income countries (LMICs) poses unique challenges because mechanical ventilation and mixed−mode strategies are often unavailable. Carbon dioxide (CO2) can serve as a proxy for ventilation and, hence, airborne infectious disease transmission risk in naturally ventilated spaces. We assessed the adequacy of ventilation in Liberian hospitals. Methods: We sampled 3 hospitals, both urban and rural, in Montserrado County, Liberia. Moreover, 3 CO2 meters were concurrently utilized to measure CO2 levels at a 1-meter height in every patient-care room in each facility. We recorded temperature, humidity, room dimensions, and number of people in the rooms. From these variables, we calculated absolute ventilation using the ASHRAE equation to determine areas with the highest risk of nosocomial respiratory disease transmission. We also recorded qualitative observations about the sampled spaces. Results: From August 2022 to February 2023, 39 rooms in 3 healthcare facilities were sampled. Initial quantitative findings show that only 8 rooms (21%) met the WHO-recommended ventilation rate of 60 L per second per person. The average ventilation rate per person in the adequately ventilated settings was 86 L per second per patient, compared to 19 liters per second per patient in inadequately ventilated rooms. Additionally, 467 ppm mean CO2 was noted in well-ventilated rooms compared to 895 ppm mean CO2 in inadequately ventilated rooms.

Initial qualitative observations showed that facilities with lower CO2 readings tended to be older constructions that likely had been constructed with airborne disease such as tuberculosis in mind. Willingness to open windows was limited by lack of window screens for malaria prevention, and there was a pervasive fallacy that air conditioning was a source of ventilation. Correspondingly, of the 31 inadequately ventilated rooms, 22 (71%) had operating air conditioning units compared with 4 (50%) of the 8 adequately ventilated rooms. Overall, of the 13 rooms without air conditioning, 7 (54%) were more frequently characterized by open windows compared to only 5 of 26 (28%) of rooms that did have air conditioners. Conclusions: Being prepared for the next respiratory disease outbreak and creating more resilient healthcare systems in LMICs requires a frameshift of prevention strategies. Measuring CO2 provides a simple strategy for identifying areas at highest risk for nosocomial respiratory disease transmission, which can be prioritized for low-cost environmental interventions, such as provision of window screens, as part of routine infection prevention and control efforts.

Disclosure: None

We used a strand-specific RT-qPCR to evaluate viral replication as a surrogate for infectiousness among 242 asymptomatic inpatients with a positive severe acute respiratory coronavirus virus 2 (SARS-CoV-2) admission test. Only 21 patients (9%) had detectable SARS-CoV-2 minus-strand RNA. Because most patients were found to be noninfectious, our findings support the suspension of asymptomatic admission testing.

Of the 2,668 patients admitted with coronavirus disease 2019 (COVID-19), 4% underwent prolonged isolation for >20 days. Reasons for extended isolation were inconsistent with Centers for Disease Control and Prevention (CDC) guidelines in 25% of these patients and were questionable in 54% due to an ongoing critically ill condition at day 20 without CDC-defined immunocompromised status.

Severe acute respiratory coronavirus virus 2 (SARS-CoV-2) real-time reverse-transcription polymerase chain reaction (rRT-PCR) strand-specific assay can be used to identify active SARS-CoV-2 viral replication. We describe the characteristics of 337 hospitalized patients with at least 1 minus-strand SARS-CoV-2 assay performed >20 days after illness onset. This test is a novel tool to identify high-risk hospitalized patients with prolonged SARS-CoV-2 replication.