A QUANTITATIVE STUDY OF FACTORS INFLUENCING DECISION THRESHOLDS FOR ANTIBIOTIC INTIATION IN SUSPECTED SEPSIS

Abstract

TOPIC: Critical Care TYPE: Original Investigations PURPOSE: The imperative to provide rapid antibiotics to sepsis patients without a gold-standard diagnostic test to differentiate infectious from non-infectious etiologies requires clinicians to make treatment decisions under significant uncertainty. The threshold model of clinical decision-making describes the treatment threshold as the point at which clinicians are at equipoise regarding the decision to initiate treatment or obtain additional diagnostic data. Little is known about thresholds of infection likelihood and other factors influencing decisions to start antibiotics. We aimed to determine the infection likelihood threshold for antibiotic initiation in sepsis and test whether the threshold depends on illness severity and non-disease factors. METHODS: 153 clinicians from 3 institutions responded to an electronic survey instrument with 8 clinical vignettes with different time-varying probabilities of infection and illness severity represented by physiologic and laboratory variables. Physicians indicated when, relative to clinical parameters, they would start antibiotics. We calculated an infection likelihood score by assigning up to 5 points for derangements in temperature and white blood cell count and a severity of illness score by assigning up to 12 points for heart rate, SBP, respiratory rate, and neurologic status according to the MEWS rubric. The infection likelihood threshold associated with treatment initiation was estimated using a logistic regression model with a treat or no-treat outcome at each hour of each vignette and the following inputs: infection score, severity of illness score, infection by severity interaction, and random intercepts for physician and vignette. Using a standard approach, we considered the infection likelihood score at which half of the decisions were “treat” and half were “not treat” as the treatment threshold. We evaluated whether infection likelihood thresholds varied by severity and by physician factors (specialty and experience). RESULTS: Both infection and severity scores were significant predictors of the decision to start antibiotics (OR 2.2 95%CI: 2.1-2.2 for infection and OR 1.4, 95%CI:1.4-1.4 for severity of illness). Overall, the infection score threshold for initiating antibiotics was 3. However, this threshold demonstrated a significant interaction with severity of illness (p<0.01 for interaction term), such that antibiotics were started at lower infection scores when severity was higher (infection threshold = 5 for low severity, 2 for intermediate severity, and 0 for high severity). Specialty and years of experience were both significantly associated with the infection score threshold for starting antibiotics, both p<0.01). Infection thresholds were highest among infectious disease and lowest among emergency medicine clinicians, and infection thresholds increased with increasing years of experience. CONCLUSIONS: Decisions to start antibiotics in suspected sepsis are influenced by infection likelihood and illness severity. Infection likelihood thresholds for starting antibiotics are lower when illness severity is higher. Non-disease factors (physician specialty and experience) affected thresholds, suggesting potential unwarranted variation in antibiotic decisions. CLINICAL IMPLICATIONS: Understanding antibiotic decision-making is critical for developing clinical decision rules and practice guidelines. DISCLOSURES: No relevant relationships by Shih-Hsiung Chou, source=Web Response No relevant relationships by Marc Kowalkowski, source=Web Response No relevant relationships by Sable Skewes, source=Web Response No relevant relationships by Stephanie Parks Taylor, source=Web Response No relevant relationships by Gary Weissman, source=Web Response