Early prediction of septic shock in hospitalized patients

Steven W. Thiel, Jamie M. Rosini, William Shannon, Joshua A. Doherty, Scott T. Micek, Marin H. Kollef

Research output: Contribution to journalArticlepeer-review

63 Scopus citations


BACKGROUND: Hospitalized patients who develop severe sepsis have significant morbidity and mortality. Early goal-directed therapy has been shown to decrease mortality in severe sepsis and septic shock, though a delay in recognizing impending sepsis often precludes this intervention. OBJECTIVE: To identify early predictors of septic shock among hospitalized non-intensive care unit (ICU) medical patients. DESIGN: Retrospective cohort analysis. SETTING: A 1200-bed academic medical center. PATIENTS: Derivation cohort consisted of 13,785 patients hospitalized during 2005. The validation cohorts consisted of 13,737 patients during 2006 and 13,937 patients from 2007. INTERVENTION: Development and prospective validation of a prediction model using Recursive Partitioning And Regression Tree (RPART) analysis. METHODS: RPART analysis of routine laboratory and hemodynamic variables from the derivation cohort to identify predictors prior to the occurrence of shock. Two models were generated, 1 including arterial blood gas (ABG) data and 1 without. RESULTS: When applied to the 2006 cohort, 347 (54.7%) and 121 (19.1%) of the 635 patients developing septic shock were correctly identified by the 2 models, respectively. For the 2007 patients, the 2 models correctly identified 367 (55.0%) and 102 (15.3%) of the 667 patients developing septic shock, respectively. CONCLUSIONS: Readily available data can be employed to predict non-ICU patients who develop septic shock several hours prior to ICU admission.

Original languageEnglish
Pages (from-to)19-25
Number of pages7
JournalJournal of hospital medicine
Issue number1
StatePublished - Jan 2010


  • Prediction
  • Sepsis
  • Shock


Dive into the research topics of 'Early prediction of septic shock in hospitalized patients'. Together they form a unique fingerprint.

Cite this