Joshua C. Eby

Background: Accurate clinical outcome prediction using electronic health records (EHRs) is crucial for patient care and resource allocation. EHRs include both structured data and rich, unstructured clinical notes. However, prior machine learning methods struggle with the multimodality, long context of notes, and severe class imbalance in clinical tasks. Objective: This study aimed to introduce and evaluate KAMELEON (Knowledge-Augmented Multimodal EHR Learning for Outcome Prediction), a unified, 2-stage hybrid framework that integrates diverse EHR modalities and external biomedical knowledge to enhance clinical risk prediction. Methods: This study used the publicly available, deidentified Medical Information Mart for Intensive Care-III dataset, which includes structured and unstructured data for over 40,000 intensive care unit patients. The 2 tasks studied were 30-day readmission (403/10,031, 4% positive rate) and in-hospital mortality prediction (2423/17,903, 13% positive rate). Train-test splits were patient-disjoint (80:20). Performance was evaluated against general and medical large language models (LLMs) and structured baselines. Key metrics included the area under the receiver operating characteristic curve (AUROC), area under the precision-recall curve (AUPRC), and macro F1-score. Results: The KAMELEON framework consistently outperformed all existing baselines. 30-day readmission: the KAMELEON-balanced random forests model achieved an AUROC of 0.85 and a sensitivity (recall) of 0.79. Ablation analysis shows the critical role of the LLM-generated reasoning, with its removal causing the AUROC to drop from 0.85 to 0.7 and sensitivity to fall by over 80%. In-hospital mortality: the KAMELEON-extreme gradient boosting model achieved an AUROC of 0.92 and an AUPRC of 0.650. Unstructured-only models showed limited ability to discern mortality, with AUROC values near chance (around 0.51-0.53). Conclusions: To our knowledge, KAMELEON represents one of the first systematic frameworks to enhance LLMs for health care prediction through graph-guided knowledge retrieval combined with structured machine learning. The framework demonstrates superior performance across both prediction tasks, highlighting the synergistic value of combining diverse data modalities and LLM reasoning for robust clinical risk estimation.

BACKGROUND Accurate clinical outcome prediction using electronic health records (EHRs) is crucial for patient care and resource allocation. EHRs include both structured data and rich, unstructured clinical notes. However, prior machine learning methods struggle with the multimodality, long context of notes, and severe class imbalance in clinical tasks. OBJECTIVE This study aimed to introduce and evaluate KAMELEON (Knowledge-Augmented Multimodal EHR Learning for Outcome Prediction), a unified, 2-stage hybrid framework that integrates diverse EHR modalities and external biomedical knowledge to enhance clinical risk prediction. METHODS This study used the publicly available, deidentified Medical Information Mart for Intensive Care-III dataset, which includes structured and unstructured data for over 40,000 intensive care unit patients. The 2 tasks studied were 30-day readmission (403/10,031, 4% positive rate) and in-hospital mortality prediction (2423/17,903, 13% positive rate). Train-test splits were patient-disjoint (80:20). Performance was evaluated against general and medical large language models (LLMs) and structured baselines. Key metrics included the area under the receiver operating characteristic curve (AUROC), area under the precision-recall curve (AUPRC), and macro F1-score. RESULTS The KAMELEON framework consistently outperformed all existing baselines. 30-day readmission: the KAMELEON-balanced random forests model achieved an AUROC of 0.85 and a sensitivity (recall) of 0.79. Ablation analysis shows the critical role of the LLM-generated reasoning, with its removal causing the AUROC to drop from 0.85 to 0.7 and sensitivity to fall by over 80%. In-hospital mortality: the KAMELEON-extreme gradient boosting model achieved an AUROC of 0.92 and an AUPRC of 0.650. Unstructured-only models showed limited ability to discern mortality, with AUROC values near chance (around 0.51-0.53). CONCLUSIONS To our knowledge, KAMELEON represents one of the first systematic frameworks to enhance LLMs for health care prediction through graph-guided knowledge retrieval combined with structured machine learning. The framework demonstrates superior performance across both prediction tasks, highlighting the synergistic value of combining diverse data modalities and LLM reasoning for robust clinical risk estimation.

Monotherapy with vancomycin or daptomycin remains guideline-based care for methicillin-resistant Staphylococcus aureus bacteremia (MRSA-B) despite concerns regarding efficacy. Limited data support potential benefit of combination therapy with ceftaroline as initial therapy. We present an assessment of outcomes of patients initiated on early combination therapy for MRSA-B. This was a single-center, retrospective study of adult patients admitted with MRSA-B between July 1, 2017 and April 31, 2023. During this period, there was a change in institutional practice from routine administration of monotherapy to initial combination therapy for most patients with MRSA-B. Combination therapy included vancomycin or daptomycin plus ceftaroline within 72 h of index blood culture and monotherapy was vancomycin or daptomycin alone. The primary outcome was a composite of persistent bacteremia, 30-day all-cause mortality, and 30-day bacteremia recurrence. Time to microbiological cure and safety outcomes were assessed. All outcomes were assessed using propensity score-weighted logistic regression. Of 213 patients included, 118 received monotherapy (115 vancomycin, 3 daptomycin) and 95 received combination therapy with ceftaroline (76 vancomycin, 19 daptomycin). The mean time from MRSA-positive molecular diagnostic blood culture result to combination therapy was 12.1 h. There was no difference between groups for the primary composite outcome (OR 1.58, 95% CI 0.60, 4.18). Time to microbiological cure was longer with combination therapy (mean difference 1.50 days, 95% CI 0.60, 2.41). Adverse event rates were similar in both groups. Early initiation of ceftaroline-based combination therapy did not improve outcomes for patients with MRSA-B in comparison to monotherapy therapy.

Addison S Hicks, Joshua C Eby; The Value of Outpatient Parenteral Antimicrobial Therapy: Better Care in Twenty-Seven Minutes, Clinical Infectious Diseases,

Quality is central to value-based care, and measurement is essential for assessing performance and understanding improvement over time. Both value-based care and methods for quality measurement are evolving. Infectious diseases (ID) has been less engaged than other specialties in quality measure development, and ID providers must seize the opportunity to engage with quality measure development and research. Antimicrobial stewardship programs are an ideal starting point for ID-related quality measure development; antimicrobial stewardship program interventions and best practices are ID specific, measurable, and effective, yet they are grossly undercompensated. Herein, we provide a scheme for prioritizing research focused on development of ID-specific quality measures. Maturation of quality measurement research in ID, beginning with an initial focus on stewardship-related conditions and then expanding to non-stewardship topics, will allow ID to take control of its future in value-based care and promote the growth of ID through greater recognition of its value.

Background: Directing COVID-19 diagnostic testing to healthcare workers (HCWs) who are likely to be infected has potential to reduce staffing shortages and decrease opportunity for in-hospital transmission; however, HCWs with COVID-19 may exhibit a range of symptoms. We assessed the burden of symptoms in relation to cycle threshold (Ct) values as a surrogate for viral shedding in vaccinated healthcare workers. Methods: We retrospectively reviewed employee health records of COVID-19–vaccinated employees who tested positive for SARS-CoV-2 between December 2020 and January 2022 at 2 academic hospital systems. We reviewed demographic data, reasons for testing including symptoms, exposure history, medical history, vaccination dates, Ct values, and genotypes when available. We compared mean Ct values between symptomatic and minimally symptomatic cases using independent sample t tests. Patients were defined as minimally symptomatic if they had no symptoms or a single symptom that is not cough, fever, or anosmia at the time of testing. Patients were defined as more symptomatic if they reported >1 symptom or cough, fever, or anosmia. Results: In total, 298 HCWs tested positive for COVID-19. Most positive cases were female (73%), white (78%), and had patient-facing roles (77%). Genotypic testing (n = 109) revealed that most genotypes belonged to the SARS-CoV-2 delta variant (AY lineages, B1.617.2). More cases were minimally symptomatic (62%) than were more symptomatic (38%). None required hospitalization during the study period. Mean Ct values (n = 141) showed no significant difference between more symptomatic and minimally symptomatic cases (19.8 vs 20.6; P = .40) (Fig. 1). Also, there was no significant difference in mean Ct value, comparing those with vaccination 90 days prior to positive (20.52 vs 19.88; P = .537). Conclusions: Our study shows no significant difference in cycle threshold values between minimally symptomatic and more symptomatic infections in vaccinated HCWs. In addition, HCWs exhibit high viral load even when infected within 90 days after vaccination. When considering whether to attend work, HCWs should be aware that mild symptoms and recent vaccination do not necessarily reflect low transmissibility and that they should follow CDC guidance regarding when to return to work. Disclosures: None

Complement receptor 3 (CR3; CD11b/CD18; α _m β ₂ integrin) is a conserved phagocytic receptor. The active conformation of CR3 binds the iC3b fragment of complement C3 as well as many host and microbial ligands, leading to actin-dependent phagocytosis. There are conflicting reports about how CR3 engagement affects the fate of phagocytosed substrates. Using imaging flow cytometry, we confirmed that binding and internalization of iC3b-opsonized polystyrene beads by primary human neutrophils was CR3-dependent. iC3b-opsonized beads did not stimulate neutrophil reactive oxygen species (ROS), and most beads were found in primary granule-negative phagosomes. Similarly, Neisseria gonorrhoeae (Ngo) that does not express phase-variable Opa proteins suppresses neutrophil ROS and delays phagolysosome formation. Here, binding and internalization of Opa-deleted (Δopa) Ngo by adherent human neutrophils was inhibited using blocking antibodies against CR3 and by adding neutrophil inhibitory factor, which targets the CD11b I-domain. Neutrophils did not produce detectable amounts of C3 to opsonize Ngo. Conversely, overexpressing CD11b in HL-60 promyelocytes enhanced Δopa Ngo phagocytosis, which required CD11b I domain. Phagocytosis of Ngo was also inhibited in mouse neutrophils that were CD11b-deficient or treated with anti-CD11b. Phorbol ester treatment upregulated surface CR3 on neutrophils in suspension, enabling CR3-dependent phagocytosis of Δopa Ngo. Neutrophils exposed to Δopa Ngo had limited phosphorylation of Erk1/2, p38, and JNK. Neutrophil phagocytosis of unopsonized Mycobacterium smegmatis , which also resides in immature phagosomes, was CR3-dependent and did not elicit ROS. We suggest that CR3-mediated phagocytosis is a silent mode of entry into neutrophils, which is appropriated by diverse pathogens to subvert phagocytic killing.

Objectives: This study was performed to assess whether an intervention for critically appraising influenza vaccine exemption requests from healthcare personnel (HCP) affected (1) the overall rate of influenza vaccine exemption within a healthcare institution and/or (2) the rates of postintervention vaccine acceptance among those who inconsistently request exemption from annual vaccination and those who consistently request exemption from vaccination.Design: Retrospective, before-and-after intervention study.Setting: We conducted the study at a single academic medical center.Participants: This study included 29,663 HCP.Methods: Between 2010 and 2019, HCP were permitted to request an exemption from influenza vaccination without critical appraisal of exemption requests. After January 2019, medical center policy required critical appraisal of exemption requests. Of those employed 3 or more years who requested an exemption at least once during the preintervention period (n = 1,177), those with unchanging exemption reasons annually were termed “consistent exempters.” Those who changed reasons or accepted vaccination n ≥ 1 times were termed “inconsistent exempters.”Results: The overall exemption rate from influenza vaccine decreased from 3.8% to 1.2% (P < .001; N = 29,663) after the intervention. Of those requesting exemption at least once before the intervention, 329 (28.0%) of 1,177 were consistent exempters and 878 (72.0%) were inconsistent exempters. Of inconsistent exempters employed after the intervention, 442 (88.9%) of 497 accepted vaccine postintervention compared with 118 (59.6%) of 198 consistent exempters (P < .001). Of all exempters who changed from exemption to acceptance after the intervention, 442 (78.9%) of 560 were inconsistent exempters.Conclusions: Critical appraisal of HCP exemption requests promotes influenza vaccine acceptance, and acceptance by inconsistent exempters drives the effect of the intervention. Analysis of changes in annual exemption requests represents a novel objective method for describing those on the spectrum of vaccine hesitancy.

Respirators provide protection from airborne particles, and the N95 filtering facepiece respirator (FFR) is the most commonly used type in health care. Recent FFR shortages have led to increased interest in reusable elastomeric half-mask respirators (EHMRs),^1,2 which have a flexible interface and larger straps than FFRs. Data from simulation and industrial settings suggests that EMHRs may provide higher respiratory protection than FFRs.³ Case reports in previous severe acute respiratory syndrome coronavirus pandemics have suggested transmission during cardiopulmonary resuscitation (CPR) despite FFR use,⁴ and simulation studies have demonstrated an unacceptable leak during CPR in up to 40% FFR users.⁵ Owing to shortages of FFR models, individuals who could not be fit in an available FFR model were fit for an EMHR. We assessed whether EHMRs provide improved fit during simulated CPR compared with FFRs.

Our study describes the first in vivo transposon sequencing (Tn-seq) analysis of B. pertussis and identifies genes predicted to be essential for in vivo growth in a murine model of intranasal infection, generating key resources for future investigations into B. pertussis pathogenesis and vaccine design.