​Background: Antimicrobial resistance is rising in South and Southeast Asia. When new resistance mechanisms are reported, prescribing and purchasing may shift towards last-resort drugs, accelerating resistance and raising costs. Evidence on these market responses is limited. Methods: We analysed quarterly IQVIA sales data for 10,767 anti-infective products sold in 30 Indian metropolitan areas from 2008–13. Products were classified using the WHO Access, Watch, and Reserve (AWaRe) framework, which groups antibiotics into first-line (Access), higher-resistancerisk (Watch), and last-resort (Reserve) categories. We estimated event-study models around early 2010 reports of New Delhi metallo-β-lactamase-1 (NDM-1), comparing Watch and Reserve products with Access products while controlling for product, city, and quarter fixed effects. The estimates show how Watch and Reserve outcomes changed relative to Access. Findings: Reserve quantities rose disproportionately after early 2010. Post-shock average Reserve quantities were about 1.9 times the pre-period baseline, versus roughly 1.2 times for Access and Watch products. Reserve price coefficients showed modest upward drift before 2010, but 95% confidence intervals included zero. By end-2013, Reserve quantities were about 0.4 log points higher relative to Access. Reserve prices more than tripled relative to baseline and increased relative to Access, with a temporary interruption around 2011–12 before the differential widened again. Interpretation: Resistance shocks can increase both use and prices of last-resort antibiotics in private markets. The pattern is consistent with a demand surge into a thin Reserve segment with limited close substitutes and constrained short-run supply. Preparedness should combine rapid clinical guidance, stewardship, dispensing controls, and financing measures that protect access while conserving Reserve antibiotics.

Social contact networks based on synthetic populations are useful for studying the effects of population features and policy interventions on disease transmission. We present an adaptable and accessible method for generating geographically detailed synthetic populations and associated contact networks from public census data, and apply it to a selection of US metropolitan areas. We simulate a respiratory pathogen spreading in each population and find that network structure alone produces differences in infection risk among racial/ethnic subpopulations, as well as between geographic locations of differing socioeconomic status, particularly in urban centers. We then simulate a work and school closure policy intervention, and find an increase in geographic infection risk differences, and in some cities, in racial/ethnic risk differences as well. Different outcomes between cities are associated with demographic and geographic differences in household size, contact with school-age children, and employment industry. The results suggest that demography, socioeconomics, and policy interact in a context-dependent manner to shape epidemiological outcomes. We have made our methods available as open-source software that can be extended by other researchers.

In the early stages of the COVID-19 pandemic, uncertainty around the extent of SARS-CoV-2 spread hampered policymakers' understanding of the epidemic's extent. Mathematical models, which proved vital for aiding decision-making, relied primarily on reported cases that were unreliable due to significant underdetection and underreporting. While serological data was used to improve understanding of the epidemiology, it can be costly and difficult to implement without bias. To counter these issues, we integrated serological data from 7229 remnant serum samples collected in 15 Maryland emergency departments (EDs) in Maryland between August and December 2020 into a Bayesian modeling approach to derive an estimate of the incidence of infection and the case fatality rate during the pandemic's initial wave. We estimated that 5.2% (95% CI, 3.7-7.2%) of the population of Maryland had been infected by late fall 2020. The inferred reporting rate that was estimated started low (<10% in March 2020) and increased to 32% (95% HDI = 26-41%) by the fall, while the estimated infection fatality rate was likely initially higher but fell to 0.51% (95% HDI = 0.43-0.68%) after 1 September 2020. These results demonstrate how existing ED infrastructure can be leveraged to generate less biased, more accurate estimates of the true prevalence of a disease, improving the ability to make decisions and allocate resources under uncertainty.

OBJECTIVES:To evaluate the implementation and effectiveness of a novel home infusion central line-associated bloodstream infection (CLABSI) and home infusion-onset bloodstream infection (HiOB) dashboard and prevention toolkit. DESIGN:Mixed methods study. SETTING AND PARTICIPANTS:Five home infusion agencies participating in the first CLABSI prevention collaborative. METHODS:Agencies uploaded CLABSI and HiOB data to a comparative dashboard. The dashboard started in December 2022 and accepted retrospective data from June 2021. A CLABSI prevention toolkit was made available in June 2024. Using an interrupted time series, we present CLABSI and HiOB rates before and after dashboard and toolkit implementation. We surveyed and interviewed participants about the tools and toolkit, using directed content analysis to analyze the interviews. RESULTS:After dashboard implementation, there was a decrease in CLABSI (-0.23/10,000 home-CVC days, 95% CI -0.28 to -0.18) and HiOB (-0.25/10,000 home-CVC days, 95% CI: -0.31 to -0.18) over time. With toolkit implementation, there was a further decrease in CLABSI (-0.17/10,000 home-CVC days, 95% CI: -0.30 to -0.044) and HiOB (-0.23/10,000 home-CVC days, 95% CI: -0.37 to -0.089) over time. Themes were associated with use of the tools (accessible, adaptable, patient-centered tools; user-friendly education to enhance understanding; barriers identified; tool mismatches; and strategies for tool delivery) and toolkit implementation (structural barriers, user-centered design, collaborative engagement and communication, toolkit used to enhance workforce competency, and concerns related to consistency). CONCLUSIONS:Implementation of a dashboard and a CLABSI prevention toolkit were each associated with both CLABSI and HiOB reduction in a collaborative of home infusion agencies.

The US Centers for Disease Control and Prevention (CDC), in 2019, designated Methicillin-resistant Staphylococcus aureus (MRSA) as a serious antimicrobial resistance threat. The risk of acquiring MRSA and suffering life-threatening consequences due to it remains especially high for hospitalized patients due to a unique combination of factors, including: co-morbid conditions, immuno suppression, antibiotic use, and risk of contact with contaminated hospital workers and equipment. In this paper, we present a novel generative probabilistic model, GenHAI, for modeling sequences of MRSA test results outcomes for patients during a single hospitalization. This model can be used to answer many important questions from the perspectives of hospital administrators for mitigating the risk of MRSA infections. Our model is based on the probabilistic programming paradigm, and can be used to approximately answer a variety of predictive, causal, and counterfactual questions. We demonstrate the efficacy of our model by comparing it against discriminative and generative machine learning models using two real-world datasets.

The world's governments have agreed on actions to address the challenge of antibiotic resistance. This raises the question of what level of national action is associated with improved outcomes, including both slower growth and lower levels of antibiotic resistance. Answering this question is challenged by variation in data availability and quality as well as disruptive events such as the COVID-19 pandemic. We investigate the association between level of national action and temporal trends in multiple indicators related to health system capacity, antibiotic use (ABU), absolute rates of resistance (ABR) and a Drug Resistance Index (DRI). Using the Global Database for Tracking Antimicrobial Resistance (TrACSS) to construct an index of national action, we apply cross-sectional regression across 73 countries to estimate the association between the level of action in 2016 and trends in national indicators (2000-2016). We find that national action is consistently associated with improved linear or categorical trends in all groups of indicators. Reductions are associated with a relatively high action index (range 0-4) for ABU (median 2.8, 25-75% quartile 2.6-3.3), ABR (3.0, 2.4-3.4), and DRI (3.5, 3.1-3.6). These associations are robust to the inclusion of other contextual factors related to socio-economic conditions, human population density, animal production and climate. Since 2016, a majority of both Low- and Middle-Income Countries (LMICs) and High-Income Countries (HICs) report increased action on repeated questions, while one third of countries report reduced action. The main limitations in interpretation are heterogeneity in data availability and in when actions have been implemented. Our findings highlight the importance of national action to address the domestic situation related to antibiotic resistance and indicate the value of both incremental changes in reducing adversity of outcomes and the need for high levels of action in delivering reduced levels of resistance.

The presence of antibiotics in surface waters poses risks to aquatic ecosystems and human health due to their toxicity and influence on antimicrobial resistance. After human consumption and partial metabolism, antibiotic residues are excreted and undergo complex accumulation and decay processes along their pathway from wastewater to natural river systems. Here, we use a global contaminant fate model to estimate that of the annual human consumption of the 40 most used antibiotics (29,200 tonnes), 8,500 tonnes (29%) are released into the river system and 3,300 tonnes (11%) reach the world's oceans or inland sinks. Even when only domestic sources are considered (i.e. not including veterinary or industrial sources), we estimate that 6 million km of rivers worldwide are subject to total antibiotic concentrations in excess of thresholds that are protective of ecosystems and resistance promotion during low streamflow conditions, with the dominant contributors being amoxicillin, ceftriaxone, and cefixime. Therefore, it is of concern that human consumption alone represents a significant risk for rivers across all continents, with the largest extents found in Southeast Asia. Global antibiotic consumption has grown rapidly over the last 15 years and continues to increase, particularly in low- and middle-income countries, requiring new strategies to safeguard water quality and protect human and ecosystem health.

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: Hepatic steatosis is a precursor to more severe liver disease, increasing morbidity and mortality risks. In the Emergency Department, routine abdominal imaging often reveals incidental hepatic steatosis that goes undiagnosed due to the acute nature of encounters. Imaging reports in the electronic health record contain valuable information not easily accessible as discrete data elements. We hypothesized that large language models could reliably detect hepatic steatosis from reports without extensive natural language processing training. Methods: We identified 200 adults who had CT abdominal imaging in the Emergency Department between August 1, 2016, and December 31, 2023. Using text from imaging reports and structured prompts, 3 Azure OpenAI models (ChatGPT 3.5, 4, 4o) identified patients with hepatic steatosis. We evaluated model performance regarding accuracy, inter-rater reliability, sensitivity, and specificity compared to physician reviews. Results: The accuracy for the models was 96.2% for v3.5, 98.3% for v4, and 98.8% for v4o. Inter-rater reliability ranged from 0.99 to 1.00 across 10 iterations. Mean model confidence scores were 2.9 (SD 0.8) for v3.5, 3.9 (SD 0.3) for v4, and 4.0 (SD 0.07) for v4o. Incorrect evaluations were 76 (3.8%) for v3.5, 34 (1.7%) for v4, and 25 (1.3%) for v4o. All models showed sensitivity and specificity above 0.9. Conclusions: Large language models can assist in identifying incidental conditions from imaging reports that otherwise may be missed opportunities for early disease intervention. Large language models are a democratization of natural language processing by allowing for a user-friendly, expansive analyses of electronic medical records without requiring the development of complex natural language processing models.

BACKGROUND:Human antibiotic consumption is a major contributing factor to antimicrobial resistance. Understanding the dynamics of the antibiotic market can help improve antibiotic stewardship efforts and encourage innovation. METHODS:We used quarterly pharmaceutical sales value and volume data from IQVIA MIDAS to estimate aggregate and per capita real annual spending (inflation adjusted) on antibiotics in 62 countries from 2013 to 2023, with unit values defined by the ratio of sales value and quantity. We evaluated trends by broad classes of antibiotics and country income groups and conducted multivariate regression analyses to identify associations with factors such as income and health spending. RESULTS:Between 2013 and 2023, aggregate and per capita real spending on antibiotics decreased from $49.61 billion to $30.68 billion and from $12.08 to $7.92, respectively. Real spending per unit of antibiotic, which is an indicator of price but not necessarily the final consumer price, also declined from $0.85 (2013) to $0.45 (2023). Spending decreased across country income groups and converged, driven by more rapid reductions in high-income countries as compared with other regions. In 2020, spending decreased sharply due to the COVID-19 pandemic, followed by a small rebound. In multivariate analysis, income, health spending, median age, and clean water access were associated positively with spending on antibiotics per 1000 people, while schooling attainment, availability of doctors, and higher state capacity were associated negatively. CONCLUSIONS:Global spending on antibiotics declined and overall converged among countries from 2013 to 2023. More investment is necessary toward reducing antibiotic use and developing new effective antibiotics.

Background: Clinical trials for assessing the effects of infection prevention and control (IPC) interventions are expensive and have shown mixed results. Mathematical models can be relatively inexpensive tools for evaluating the potential of interventions. However, capturing nuances between institutions and in patient populations have adversely affected the power of computational models of nosocomial transmission.Methods: In this study, we present an agent-based model of ICUs in a tertiary care hospital, which directly uses data from the electronic medical records (EMR) to simulate pathogen transmission between patients, HCWs, and the environment. We demonstrate the application of our model to estimate the effects of IPC interventions at the local hospital level. Furthermore, we identify the most important sources of uncertainty, suggesting areas for prioritization in data collection.Results: Our model suggests that the stochasticity in ICU infections was mainly due to the uncertainties in admission prevalence, hand hygiene compliance/efficacy, and environmental disinfection efficacy. Analysis of interventions found that improving mean HCW compliance to hand hygiene protocols to 95% from 70%, mean terminal room disinfection efficacy to 95% from 50%, and reducing post-handwashing residual contamination down to 1% from 50%, could reduce infections by an average of 36%, 31%, and 26%, respectively.Conclusions: In-silico models of transmission coupled to EMR data can improve the assessment of IPC interventions. However, reducing the uncertainty of the estimated effectiveness requires collecting data on unknown or lesser known epidemiological and operational parameters of transmission, particularly admission prevalence, hand hygiene compliance/efficacy, and environmental disinfection efficacy.

Background:Predicting seasonal and emerging waves of respiratory viruses is crucial for effective public health responses. Despite significant efforts in developing coronavirus disease 2019 (COVID-19) forecast models, there remains a need for improvement in model performances. Methods:We developed and evaluated a machine learning model to forecast COVID-19 hospitalizations by extending the Neural Basis Expansion Analysis for Time Series Forecasting (N-BEATS) architecture. Specifically, we integrated a temporal convolutional network to incorporate exogenous variables and added additional residual blocks to create a variance-forecasting network component for probabilistic predictions. We compared the performance of our model to the ensemble models from the COVID-19 Forecast Hub. Additionally, we implemented the model in a large academic medical center, applying transfer learning to adapt the model to local hospitalization data. Results:Our model demonstrated a 34.0% improvement in mean absolute error over the performance-weighted ensemble and 37.0% over the unweighted ensemble in predicting total US hospitalizations. Similar trends were obtained using mean absolute percent error and symmetric mean absolute percent error. In a real-world implementation, the model provided actionable forecasts for hospital leadership to optimize resource allocation and surge preparation. Conclusions:The enhanced architecture significantly improves the forecasting of COVID-19 hospitalizations, particularly in anticipating peaks and resurgences. Its successful implementation in a hospital system highlights its potential for aiding decision-making and resource planning during pandemics and other respiratory disease outbreaks.

Healthcare-associated infections (HAIs) from multi-drug resistant organisms (MDROs) pose a significant challenge for healthcare systems. Patients can arrive at hospitals already infected ("importation") or acquire infections during their stay ("nosocomial infection"). Many cases, often asymptomatic, complicate rapid identification due to testing limitations and delays. Although recent advancements in mathematical modeling and machine learning have aimed to identify at-risk patients, these methods face challenges: transmission models often overlook valuable electronic health record (EHR) data, while machine learning approaches typically lack mechanistic insights into underlying processes. To address these issues, we propose NeurABM, a novel framework that integrates neural networks and agent-based models (ABM) to leverage the strengths of both methods. NeurABM simultaneously learns a neural network for patient-level importation predictions and an ABM for infection identification. Our findings show that NeurABM significantly outperforms existing methods, marking a breakthrough in accurately identifying importation cases and forecasting future nosocomial infections in clinical practice.

Objectives Given the support for methicillin-resistant Staphylococcus aureus (MRSA) antimicrobial stewardship in the 2021 Surviving Sepsis Campaign Guidelines, we sought to measure the use of vancomycin in the emergency department (ED) in the years preceding these recommendations. Methods A retrospective cohort study was conducted of all patients aged ≥ 18 years presenting to 5 emergency departments within a university-based health system who were given intravenous (IV) vancomycin during their ED index visit. The primary outcome assessed the proportion of patients with MRSA-positive blood cultures who received IV vancomycin in the ED. We also measured associations between clinical attributes associated with any MRSA infection. Results Of the 20,212 unique ED visits for patients who received IV vancomycin, 63% (n = 12,755) had at least 1 MRSA risk factor. Only 2.4% (n = 494) and 14.1% (n = 2850) of patients receiving IV vancomycin in the ED were found to have MRSA bacteremia or any MRSA-positive culture, respectively. A total of 3160 patients met Sepsis-3 criteria and received IV vancomycin, though 65% (n = 2064) had no MRSA risk factors. For any patient with culture-proven MRSA, 63.8% (n = 315) and 43.4% (n = 1236) received an MRSA antimicrobial in the ED. MRSA risk factors were not associated with MRSA bacteremia (≥1 MRSA risk factor: odds ratio, 1.3, 95% CI, 0.9-1.8) or an MRSA-positive culture of any type (odds ratio, 0.9, 95% CI, 0.7-1.1). Conclusion Within our hospital system, MRSA was an infrequent cause of bacteremia for patients presenting to the ED with sepsis or septic shock. Although vancomycin is frequently used in the ED, many patients with culture-proven MRSA did not receive MRSA antimicrobials. Notably, one-third of patients with culture-proven MRSA had no MRSA risk factors. MRSA risk factors were not predictive of culture-proven MRSA, thus highlighting the complexity of antimicrobial stewardship in the ED without validated clinical decision rules.

At a United States hospital, sequencing of ICU rectal surveillance cultures indicated 5% ESBL-E colonization. Of confirmed ESBL isolates, 91% were Escherichia coli or Klebsiella pneumoniae; 6% carried non-bla CTX-M genes. Only 53% of third-generation cephalosporin-resistant Enterobacterales harbored ESBL genes, underscoring the limitations of phenotypic approaches as ESBL surrogates, particularly for non-E. coli/K. pneumoniae species.

Over a 2-year period, we identified Transmission from Room Environment Events (TREE) across the Johns Hopkins Health System, where the subsequent room occupant developed the same organism with the same antimicrobial susceptibilities as the patient who had previously occupied that room. Overall, the TREE rate was 50/100,000 inpatient days.

Objective: To evaluate the economic costs of reducing the University of Virginia Hospital's present "3-negative" policy, which continues methicillin-resistant Staphylococcus aureus (MRSA) contact precautions until patients receive 3 consecutive negative test results, to either 2 or 1 negative. Design: Cost-effective analysis. Settings: The University of Virginia Hospital. Patients: The study included data from 41,216 patients from 2015 to 2019. Methods: We developed a model for MRSA transmission in the University of Virginia Hospital, accounting for both environmental contamination and interactions between patients and providers, which were derived from electronic health record (EHR) data. The model was fit to MRSA incidence over the study period under the current 3-negative clearance policy. A counterfactual simulation was used to estimate outcomes and costs for 2- and 1-negative policies compared with the current 3-negative policy. Results: Our findings suggest that 2-negative and 1-negative policies would have led to 6 (95% CI, -30 to 44; P < .001) and 17 (95% CI, -23 to 59; -10.1% to 25.8%; P < .001) more MRSA cases, respectively, at the hospital over the study period. Overall, the 1-negative policy has statistically significantly lower costs ($628,452; 95% CI, $513,592-$752,148) annually (P < .001) in US dollars, inflation-adjusted for 2023) than the 2-negative policy ($687,946; 95% CI, $562,522-$812,662) and 3-negative ($702,823; 95% CI, $577,277-$846,605). Conclusions: A single negative MRSA nares PCR test may provide sufficient evidence to discontinue MRSA contact precautions, and it may be the most cost-effective option.

Rationale & Objective: The prevalence of community-acquired acute kidney injury (CA-AKI) in the United States and its clinical consequences are not well described. Our objective was to describe the epidemiology of CA-AKI and the associated clinical outcomes. Study Design: Retrospective cohort study. Setting & Participants: 178,927 encounters by 139,632 adults at 5 US emergency departments (EDs) between July 1, 2017, and December 31, 2022. Predictors: CA-AKI identified fi ed using KDIGO (Kidney Disease: Improving Global Outcomes) serum creatinine (Scr)-based criteria. Outcomes: For encounters resulting in hospitalization, the in-hospital trajectory of AKI severity, dialysis initiation, intensive care unit (ICU) admission, and death. For all encounters, occurrence over 180 days of hospitalization, ICU admission, new or progressive chronic kidney disease, dialysis initiation, and death. Analytical Approach: Multivariable logistic regression analysis to test the association between CA-AKI and measured outcomes. Results: For all encounters, 10.4% of patients met the criteria for any stage of AKI on arrival to the ED. 16.6% of patients admitted to the hospital from the ED had CA-AKI on arrival to the ED. The likelihood of AKI recovery was inversely related to CA-AKI stage on arrival to the ED. Among encounters for hospitalized patients, CA-AKI was associated with in-hospital dialysis initiation (OR, 6.2; 95% CI, 5.1-7.5), ICU admission (OR, 1.9; 95% CI, 1.7-2.0), and death (OR, 2.2; 95% CI, 2.0-2.5) compared with patients without CA-AKI. Among all encounters, CA-AKI was associated with new or progressive chronic kidney disease (OR, 6.0; 95% CI, 5.6-6.4), dialysis initiation (OR, 5.1; 95% CI, 4.5-5.7), subsequent hospitalization (OR, 1.1; 95% CI, 1.1-1.2) including ICU admission (OR, 1.2; 95% CI, 1.11.4), and death (OR, 1.6; 95% CI, 1.5-1.7) during the subsequent 180 days. Limitations: Residual confounding. Study implemented at a single university-based health system. Potential selection bias related to exclusion of patients without an available baseline Scr measurement. Potential ascertainment bias related to limited repeat Scr data during follow-up after an ED visit. Conclusions: CA-AKI is a common and important entity that is associated with serious adverse clinical consequences during the 6-month period after diagnosis. Complete information references. Correspondence M.R. (mehmann1@jhmi.edu) Am 83(6):762-771. online doi: j.ajkd.2023.10.009 (c) Kidney

Background: External comparisons of antimicrobial use (AU) may be more informative if adjusted for encounter characteristics. Optimal methods to define input variables for encounter-level risk-adjustment models of AU are not established. Methods: This retrospective analysis of electronic health record data included 50 US hospitals in 2020-2021. We used NHSN definitions for all antibacterials days of therapy (DOT), including adult and pediatric encounters with at least 1 day present in inpatient locations. We assessed 4 methods to define input variables: 1) diagnosis-related group (DRG) categories by Yu et al., 2) adjudicated Elixhauser comorbidity categories by Goodman et al., 3) all Clinical Classification Software Refined (CCSR) diagnosis and procedure categories, and 4) adjudicated CCSR categories where codes not appropriate for AU risk-adjustment were excluded by expert consensus, requiring review of 867 codes over 4 months to attain consensus. Data were split randomly, stratified by bed size as follows: 1) training dataset including two-thirds of encounters among two-thirds of hospitals; 2) internal testing set including one-third of encounters within training hospitals, and 3) external testing set including the remaining one-third of hospitals. We used a gradient-boosted machine (GBM) tree-based model and two-staged approach to first identify encounters with zero DOT, then estimate DOT among those with >0.5 probability of receiving antibiotics. Accuracy was assessed using mean absolute error (MAE) in testing datasets. Correlation plots compared model estimates and observed DOT among testing datasets. The top 20 most influential variables were defined using modeled variable importance. Results: Our datasets included 629,445 training, 314,971 internal testing, and 419,109 external testing encounters. Demographic data included 41% male, 59% non-Hispanic White, 25% non-Hispanic Black, 9% Hispanic, and 5% pediatric encounters. DRG was missing in 29% of encounters. MAE was lower in pediatrics as compared to adults, and lowest for models incorporating CCSR inputs (Figure 1). Performance in internal and external testing was similar, though Goodman/Elixhauser variable strategies were less accurate in external testing and underestimated long DOT outliers (Figure 2). Agnostic and adjudicated CCSR model estimates were highly correlated; their influential variables lists were similar (Figure 3). Conclusion: Larger numbers of CCSR diagnosis and procedure inputs improved risk-adjustment model accuracy compared with prior strategies. Variable importance and accuracy were similar for agnostic and adjudicated approaches. However, maintaining adjudications by experts would require significant time and potentially introduce personal bias. If findings are confirmed, the need for expert adjudication of input variables should be reconsidered.