| dc.description.abstract | Sepsis is an elusive and costly syndrome that is one of the leading causes of death globally. Annually, there are approximately 19 million cases of sepsis that result in more than 5 million deaths. The Agency for Healthcare Research and Quality (AHRQ) ranked sepsis as the most expensive condition ($23.7 billion) for patients treated in hospitals in the United States (U.S.). Nurses are critical in the early identification of sepsis and implementation of therapeutic interventions known as the “sepsis bundle”. Previously, sepsis was described as a systemic, pro-inflammatory response to an infection. Sepsis was defined as two or more systemic inflammatory response syndrome (SIRS) criteria with a suspected infection, severe sepsis was defined as sepsis with organ failure and septic shock was defined as severe sepsis with shock. For several decades SIRS criteria with organ failure criteria have been used to develop measurement systems for detection of sepsis. A recent study comparing SIRS criteria to the sepsis-related organ failure assessment (SOFA) score demonstrated that SOFA had greater prognostic accuracy of mortality in patients with an infection than SIRS. This led to sepsis definition changes in 2016. The term “severe sepsis” was dropped and sepsis was defined as a life-threatening organ dysfunction caused by a dysregulated host response to an infection leading to tissue injury and organ failure. Many clinicians were concerned that this new definition might lead to late detection of sepsis. What was unknown was how well SIRS with organ failure criteria compared with SOFA in detection of sepsis. Many clinicians in the U.S. working in a TeleICU had been using SIRS with organ failure criteria to support early identification of sepsis. Using human factors science concepts, their practice was studied and an electronic sepsis alert (sepsis prompt) was developed. Thus, the overall objective of this dissertation was to conduct a retrospective study using a large U.S. data repository to determine if an electronic prompt, that uses SIRS and organ failure (OF) criteria, can detect sepsis. Another objective of this study was to determine the prognostic accuracy of the SOFA score and the sepsis prompt in discriminating in-hospital mortality among patients with sepsis in the intensive care unit. Among 2,020,489 patients admitted to ICUs associated with a TeleICU from January 1, 2010, to December 31, 2015, at 459 hospitals throughout the U.S., we identified 912,509 (45%) eligible patients at 183 hospitals. We compared the performance of the SOFA score and sepsis prompt criteria in detecting sepsis. Of those in the primary cohort, a secondary cohort was derived based on presence of sepsis resulting 186,870 (20.5%) patients. To assess performances of the SOFA score and the sepsis prompt (a Fuzzy Logic SIRS and OF algorithm) to detect sepsis, we calculated diagnostic performance of an increase in the SOFA score of 2 or more and criteria met for the Fuzzy Logic SIRS and OF algorithm. For predictive validity, training of baseline risk models was performed on training sets with prediction and performance analytics completed on test sets for each cohort for the outcomes of mortality and sepsis. Results were expressed as the fold change in outcome over deciles of baseline risk of death or risk of sepsis, area under the receiver operating characteristic curve (AUROC), and sensitivity, specificity, and negative and positive predictive values. In the primary cohort (912,509) there were 86,219 (9.4%) who did not survive their hospital stay and 186,870 (20.5%) with suspected sepsis of whom 34,617 (18.5%) did not survive hospitalization. The Fuzzy Logic SIRS/OF (crude AUROC 0.67, 99% CI: 0.66-0.67 and adjusted AUROC 0.77, 99% CI: 0.77-0.77) outperformed SOFA (crude AUROC 0.61, 99% CI: 0.61-0.61 and adjusted AUROC 0.74, 99% CI: 0.74-0.74) in discrimination of sepsis in both crude and adjusted AUROC (in-between differences AUROC 0.06; z-value 49.06 and AUROC 0.03; z-value 36.22, respectively). In the primary cohort, Fuzzy Logic SIRS/OF (crude AUROC 0.67, 99% CI: 0.67-0.68 and adjusted AUROC 0.78, 99% CI: 0.77-0.78) outperformed SOFA (crude AUROC 0.64, 99% CI: 0.64-0.64 and adjusted AUROC 0.76, 99% CI: 0.76-0.76) in prognostic accuracy of mortality in both crude and adjusted AUROC (in-between differences AUROC 0.03; z-value 24.68 and AUROC 0.02; z-value 14.74, respectively). In the secondary cohort, Fuzzy Logic SIRS/OF (crude AUROC 0.57, 99% CI: 0.57-0.58 and adjusted AUROC 0.69, 99% CI: 0.68-0.70) outperformed SOFA (crude AUROC 0.56, 99% CI: 0.56-0.56 and adjusted AUROC 0.68, 99% CI: 0.67-0.68) in prognostic accuracy of mortality in both crude and adjusted AUROC (in-between differences AUROC 0.01; z-value 6.86 and AUROC 0.01; z-value 7.53, respectively). The results of this study demonstrated that among adult ICU patients, the predictive validity for sepsis and in-hospital mortality of a complex algorithm based on Fuzzy Logic applied to expanded SIRS criteria with organ failure criteria was better than SOFA for detection of sepsis and for prognostic accuracy of mortality. The findings of this study support the use of a computer-enhanced algorithm that includes a combination of expanded SIRS with organ failure criteria as a tool to assist nurses and healthcare providers in early identification of sepsis. | |
