281. Detecting bacterial sepsis among allogeneic HCT recipients with population-specific bedside tools. (31st December 2020)

• Record Type: Journal Article
• Authors: Lind, Margaret; Pergam, Steven A; Liu, Catherine; Phipps, Amanda; Mooney, Stephen; Althouse, Benjamin; Carone, Marco
• Is Part Of: Open forum infectious diseases. Volume 7:Number 1(2020) Supplement
• Page Start: S140
• Page End: S141
• Publication Date: 2020-12-31
• Subjects: Communicable diseases -- Periodicals; Medical microbiology -- Periodicals; Infection -- Periodicals; 616.9
• DOI: 10.1093/ofid/ofaa439.325
• Languages: English
• ISSNs: 2328-8957

Abstract

Abstract: Background: Diagnosing sepsis among allogeneic hematopoietic cell transplant (aHCT) recipients remains challenging. Existing criteria, for use in hospitalized patients, have limited predictive accuracy among aHCT recipients and their use may lead to missed events or antibiotic overuse. We developed bedside bacterial sepsis prediction tools (criteria and decision tree [DT]) for aHCT recipients and compared them against Systemic Inflammatory Response Syndrome (SIRS), quick Sequential Organ Failure Assessment (qSOFA) and National Early Warning Score (NEWS) criteria. Methods: Adult aHCT recipients transplanted between September 2010–2019 with ≥ 1 potential infection (PI) within 100 days post-transplantation were randomly assigned to model/validation (7/3) cohorts. Tools included demographic and clinical factors and were built against a bacterial sepsis endpoint (gram-negative, Staphylococcus aureus, or Streptococcus species bacteremia). The tools were developed using best subset selection with rare event logistic regression (criteria) and classification tree (DT) algorithms. Criteria scores were estimated using a beta/10 integer weighting approach and tool predictive performances were compared against existing criteria. Results: Between September 2010–2019, 1571 recipients with ≥ 1 PI contributed 7755 PIs and 238 sepsis events. The DT model included 7 terminal nodes based on 3 predictors: temperature, respiratory rate (RR), and sex. The criteria model contained 10 categories with 4 predictors: RR, temperature, pulse, and diastolic blood pressure (Figure 1). Our criteria and DT had AUCs of 71.1% (95% Confidence Interval (CI): 64.3, 77.9%) and 70.0% (CI: 63.7, 76.2%). SIRS had the highest AUC of existing criteria – 64.7% (CI: 57.1, 71.9%). Our criteria had the highest net benefit (for probabilities < 10%) and, at a 7+ cut-point, had a sensitivity of 73.8% (CI: 61.5–84.0%) and specificity of 55.0% (CI: 52.9, 57.1%) (Figure 2). Conclusion: We developed aHCT recipient-specific bedside bacterial sepsis prediction tools with higher AUCs than existing criteria. Tools targeted to high-risk populations may lead to fewer missed sepsis events and, in turn, reduce sepsis related mortality among this high-risk population. Disclosures: Steven A. Pergam, MD, MPH, Chimerix, Inc (Scientific Research Study Investigator)Global Life Technologies, Inc. (Research Grant or Support)Merck & Co. (Scientific Research Study Investigator)Sanofi-Aventis (Other Financial or Material Support, Participate in clinical trial sponsored by NIAID (U01-AI132004); vaccines for this trial are provided by Sanofi-Aventis)