Timely interventions are essential

Implementation of a real-time computerized sepsis alert in nonintensive care unit patients* Amber M. Sawyer, PharmD; Eli N. Deal, PharmD; Andrew J. Labelle, MD; Chad Witt, MD; Steven W. Thiel, MD; Kevin Heard, BS; Richard M. Reichley, RPh; Scott T. Micek, PharmD; Marin H. Kollef, MD Objective: Early therapy of sepsis involving fluid resuscitation and antibiotic administration has been shown to improve patient outcomes. A proactive tool to identify patients at risk for developing sepsis may decrease time to interventions and improve patient outcomes. The objective of this study was to evaluate whether the implementation of an automated sepsis screening and alert system facilitated early appropriate interventions. Design: Prospective, observational, pilot study. Setting: Six medicine wards in Barnes-Jewish Hospital, a 1250-bed academic medical center. Patients: Patients identified by the sepsis screen while admitted to a medicine ward were included in the study. A total of 300 consecutive patients were identified comprising the nonintervention group (n 200) and the intervention group (n 100). Interventions: A real-time sepsis alert was implemented for the intervention group, which notified the charge nurse on the patient s hospital ward by text page. Measurements and Main Results: Within 12 hrs of the sepsis alert, interventions by the treating physicians were assessed, including new or escalated antibiotics, intravenous fluid administration, oxygen therapy, vasopressors, and diagnostic tests. After exclusion of patients without commitment to aggressive management, 181 patients in the nonintervention group and 89 patients in the intervention group were analyzed. Within 12 hrs of the sepsis alert, 70.8% of patients in the intervention group had received >1 intervention vs. 55.8% in the nonintervention group (p.018). Antibiotic escalation, intravenous fluid administration, oxygen therapy, and diagnostic tests were all increased in the intervention group. This was a singlecenter, institution- and patient-specific algorithm. Conclusions: The sepsis alert developed at Barnes-Jewish Hospital was shown to increase early therapeutic and diagnostic interventions among nonintensive care unit patients at risk for sepsis. (Crit Care Med 2011; 39:469 473) KEY WORDS: sepsis; shock; prediction Timely interventions are essential in the management of patients with sepsis rapidly progressing to severe sepsis and septic shock (1). Early goal-directed therapy involving fluid resuscitation and appropriate antibiotic administration has been shown to improve patient outcomes, including a significant decrease in mortality (2 4). Clinical evidence suggests that patients developing sepsis on general hospital wards may experience delays in treatment, *See also p. 588. From the Department of Pharmacy (AMS, END, STM), Barnes-Jewish Hospital, St Louis, MO; the Pulmonary and Critical Care Division (AJL, CW, SWT, MHK), Washington University School of Medicine, St Louis, MO; and Medical Informatics (KH, RMR), Barnes-Jewish Hospital, St Louis, MO. This study was funded in part by the Barnes- Jewish Hospital Foundation. The authors have not disclosed any potential conflicts of interest. For information regarding this article, E-mail: mkollef@dom.wustl.edu Copyright 2011 by the Society of Critical Care Medicine and Lippincott Williams & Wilkins DOI: 10.1097/CCM.0b013e318205df85 including fluid resuscitation, vasopressors, and antibiotic therapy as well as experience delays in intensive care unit transfer resulting in adverse outcomes (5 7). To improve early sepsis management, a real-time, computerized prediction tool (PT) using recursive partitioning regression tree analysis and an informaticsbased alert system was developed at Barnes-Jewish Hospital (8) (Fig. 1). The sepsis PT is an algorithm that includes routine laboratory and hemodynamic values to provide a simple screening method for the identification of impending sepsis that can be used in an automated fashion with an electronic medical record system. These parameters were selected because patients with sepsis usually present with a constellation of abnormal vital signs and laboratory findings (fever or hypothermia, tachycardia, tachypnea, abnormal white blood cell count, creatinine, liver function studies) and progression from a syndrome of abnormal vital signs and laboratory values to organ dysfunction and shock (9, 10). The goal of this study was to evaluate whether implementation of the sepsis screening and alert system facilitated early appropriate interventions for patients identified to be at risk for developing sepsis. METHODS Study Location. The study was conducted at Barnes-Jewish Hospital, a 1250-bed academic medical center in St Louis, MO. Six adult medicine wards were assessed from October 2008 through June 2009. Two of the medicine wards were assigned to be the intervention wards and the remaining four wards served as the control wards. The medicine wards are closed areas with patient care delivered by dedicated housestaff physicians under the supervision of a board-certified attending physician. The study was approved by the Washington University School of Medicine Human Studies Committee. Patients. Patients admitted to one of the control wards were included in the nonintervention group (NIG). The intervention group (IG) included all patients identified by the PT on the intervention wards who had a real-time sepsis alert generated. Patients were excluded from the study if they were not committed to aggressive therapy, defined as those with a pre-existing do not resuscitate order at the time of the alert or a documented decision for hospice or comfort care measures before the alert. 469

Study Design and Data Collection. This was a prospective pilot study that incorporated real-time data collection. The sepsis alerts were generated from a medical informatics system (Clinical Desktop; BJC HealthCare, St Louis, MO) using the PT developed to identify nonintensive care unit patients at risk for sepsis (8). The PT is an algorithm consisting of common laboratory values and hemodynamic parameters routinely monitored in hospitalized patients who were selected by the authors as a result of their availability and potential relevance to sepsis, including the shock index (heart rate divided by systolic blood pressure), mean arterial pressure, international normalized ratio, white blood cell count, hemoglobin, absolute neutrophil count, serum albumin, total bilirubin, and sodium (Fig. 1). The process of generating a sepsis alert began with nurses entry of patient hemodynamic values into the electronic medical record along with automatic transfer of laboratory data. Once the patient-specific data became electronically available, the values were automatically screened by the PT and if a patient was identified as a case, a sepsis alert was generated through the informatics system. The alert was then sent by secure e-mail to the primary investigator including patient name, location, and alerting values (Fig. 2). Additional steps to the PT process were added to formulate the active sepsis alert for the IG. The alert was sent automatically for identified patients on the two intervention wards through a text page to the charge nurse on those wards within 10 mins of the PT s identification (Fig. 2). A standardized education strategy was introduced to nurses and physicians on the intervention wards previous to implementing the active alert with the text pages. The investigators provided sepsis alert system education during staff meetings explaining the rationale, importance, and objectives of the study. Charge nurses were encouraged to respond immediately to the page, assess the patient, and inform the treating or on-call physician of the alert and the clinical status of the patient. It would then be at the physician s discretion as to what, if any, interventions were warranted. The primary outcome was the active administration of therapeutic and diagnostic interventions assessed within 12 hrs of the alert, including antibiotic escalation, intravenous fluid administration, oxygen therapy, vasopressor initiation, and diagnostics (microbiologic cultures and imaging). Secondary outcomes included rate of intensive care unit (ICU) transfer, rate of ICU transfer within 12 hrs of alert generation, inhospital mortality, hospital length of stay, and hospital length of stay after the alert. Data and Definitions. Baseline characteristics were recorded for each patient and included age, sex, race, comorbid conditions, and Acute Physiology and Chronic Health Figure 1. For each branch to the left indicates that the patient meets the conditions and to the right either the patient does not meet the condition or the data are missing. MAP, mean arterial pressure; INR, international normalized ratio; WBC, white blood cell count. Figure 2. Process of generating an automated sepsis alert. NIG, noninterventional group; IG, interventional group. Evaluation scores (11). Other baseline characteristics recorded included temperature ( 38 C or 36 C) and white blood cell count ( 16,000 cells/mm 3 or 4000 cells/mm 3 ) 24 hrs before or after generation of the sepsis alert; antibiotic therapy and diagnostics 48 hrs before generation of the sepsis alert; and use of intravenous maintenance fluids, oxygen therapy, and vasopressors at the time of the sepsis alert. Interventions were defined a priori as occurring within 12 hrs after the sepsis alert and included: antibiotic escalation, defined as any new antibiotic received or optimization of the dose of current antibiotic(s) as previously described at our institution (12, 13); administration of fluids, defined as any intravenous fluid 500 ml administered over 30 mins or the initiation of intravenous maintenance fluids at a rate 50 ml/hr for 3 hrs; oxygen therapy, defined as any form of supplemental oxygen or ventilation initiated or current requirements escalated; and diagnostic tests, including microbiologic cultures (blood, urine, sputum, wound) and radiographic imaging. The intravenous fluid thresholds were selected because they represented obtainable thresholds from our database that were felt to be clinically relevant by the investigators. The definitions for sepsis, severe sepsis, and septic shock were taken from the Surviving Sepsis Campaign (4). Statistical Analysis. Our sample size calculation was based on our prior experience (3, 5) and assumed a postalert intervention rate of 60% in the NIG and a postalert intervention 470

rate of 75% in the IG. We calculated that approximately 304 patients would need to be enrolled to achieve statistical power of 80% with a two-sided significance level of.05. We planned on enrolling patients into the study in a ratio of one intervention patient to two nonintervention patients. Chi square and Fisher s exact tests were performed for all dichotomous variables. Student s t test was performed for continuous variables. All tests were twotailed and a p value of.05 was considered statistically significant. Analyses were performed using SPSS 16.0 software package (SPSS Inc, Chicago, IL). RESULTS Patients. Baseline demographics were similar in both groups and initial treatment before the sepsis alert was evenly distributed (Table 1). Of the 300 patients identified by the alerts, 181 (90.5%) in the NIG and 89 (89.0%) in the IG were committed to aggressive medical management and constituted the study population. No nurses or physicians were contacted as a result of the sepsis alerts for any patient in the NIG. The number of patients actually developing sepsis, severe sepsis, and septic shock was similar in the IG and the NIG (Table 2). All episodes of sepsis, severe sepsis, and septic shock occurred within 48 hrs of generating the sepsis alert. Outcomes. Within 12 hrs of generating the sepsis alert, 70.8% of patients in the IG had received 1 interventions vs. 55.8% in the NIG (p.018). Significant increases were seen in antibiotic escalation (36.0% vs. 23.8%; p.035), intravenous fluid administration (38.2% vs. 23.8%; p.013), and oxygen therapy (20.2% vs. 8.3%; p.005) (Fig. 3). Diagnostics, including both microbiologic cultures and radiographic imaging, were also greater in the IG. Patients in both groups had similar rates of ICU transfer; however, patients in the IG were more likely to be transferred to the ICU within 12 hrs of the sepsis alert (9.0% vs. 4.4%) (Fig. 4). Hospital mortality, total hospital length of stay, and length of stay after generation of the sepsis alert were also similar between groups (Table 2). DISCUSSION The results of this pilot study suggest that an automated sepsis alert using a validated PT could influence the management of patients predicted to develop sepsis on general hospital wards. By implementing an informatics-based sepsis Table 1. Baseline characteristics Characteristic Nonintervention Group (n 181) Intervention Group (n 89) p Age, years (mean SD) 52.6 17.4 50.4 17.7.347 Male, no. (%) 92 (50.8) 49 (55.1).513 White, no. (%) 104 (57.5) 46 (51.7).369 Comorbidities, no. (%) Cardiovascular 86 (47.5) 49 (55.1).244 Renal 38 (21.0) 24 (26.9).273 Dialysis 17 (9.4) 8 (8.9).914 Liver 67 (37.0) 35 (39.3).713 Solid organ transplant 8 (4.4) 4 (4.5).978 Underlying malignancy 28 (15.5) 11 (12.4).494 HIV-positive 5 (2.8) 4 (4.5).456 APACHE II score (mean SD) 17.6 6.6 17.7 7.2.572 24 hrs before alert Temperature 36 C or 38.0 C and WBC 29 (16.0) 13 (14.6).763 4000 or 16,000 cells/mm 3 48 hrs before alert Antibiotics 94 (51.9) 42 (47.2).464 Radiographs 112 (61.9) 51 (57.3).470 Microbiology cultures 109 (60.2) 54 (60.6).943 Time of alert Maintenance IV fluids 44 (24.3) 19 (21.3).589 Oxygen 51 (28.2) 22 (24.7).548 Vasopressors 1 (0.5) 1 (1.1).607 APACHE, Acute Physiology and Chronic Health Evaluation; WBC, white blood cells; IV, intravenous. Table 2. Secondary outcomes Outcome Nonintervention Group (n 181) Intervention Group (n 89) p Hospital LOS, days, median (IQR) 7 (5 14) 9 (5 15).805 Hospital LOS after alert days, median (IQR) 5 (3 10) 6 (3 12).724 Hospital mortality, no. (%) 21 (11.6) 9 (10.1).714 Septic shock a within 12 hrs of alert, no. (%) 6 (3.3) 5 (5.6).368 Sepsis, no. (%) 42 (23.2) 11 (12.4).035 Severe sepsis, no. (%) 44 (24.3) 27 (30.3).290 Septic shock, a no. (%) 17 (9.4) 13 (14.6).200 LOS, length of stay; IQR, interquartile range. a Requirement for vasopressors to maintain mean arterial pressures 60 mm Hg despite intravenous fluid administration. Values expressed as median and interquartile range (IQR). screening and alert tool, comprised of routine laboratory and hemodynamic parameters, we were able to significantly increase the rate of interventions within 12 hrs of identifying patients at risk for the onset of sepsis. The potential importance of these findings is suggested by a recent multicentered study performed by the Surviving Sepsis Campaign, which demonstrated that one-third of patients with sepsis originate from hospital wards and that these patients have a greater overall mortality compared with patients with sepsis originating in the emergency department or the ICU (14). Track and trigger scoring systems to detect clinical deterioration of patients on general hospital wards are used globally at many institutions for the triage of acutely ill patients admitted to these wards. Scoring systems such as the modified early warning system provide a tool for bedside evaluation based on physiological parameters (15). If the assessment (tracking) results in a score that is indicative of deterioration, escalation of care may need to be initiated (trigger). Recommendations have been made by the Department of Health in London for establishing critical care teams trained to respond quickly to a deteriorating patient on a hospital ward and to make appropriate interventions (16, 17). The intention was to extend critical care skills beyond the ICU to all wards to ensure timely ICU admission or support on the wards, to 471

Figure 3. New therapies and diagnostics obtained within 12 hrs of the sepsis alert in the intervention group (black bars) and the nonintervention group (white bars). Figure 4. Intensive care unit (ICU) transfer and hospital mortality for patients in the intervention group (black bars) and the nonintervention group (white bars). avert ICU admissions, and to enhance the skills of nonspecialists treating acutely ill patients (17). There is limited evidence to suggest that clinical response teams contribute to improved patient outcomes. One wardrandomized trial reported reduced inhospital mortality after implementing a critical care outreach service on surgical and medical wards (18), whereas other nonrandomized trials have reported reductions in ICU admission and length of stay (19, 20). However, conflicting evidence was seen from the hospital-randomized Medical Emergency Team Implementation trial performed in Australia, showing no improvement in outcomes including cardiac arrest, ICU admission, and mortality (21). Presence of a dedicated medical emergency team led to more emergency team calls without substantially affecting patient outcomes. Similarly, a national survey of hospitals in England found that critical care outreach services that have been widely introduced throughout England were associated with no clear evidence of improvement inpatient outcomes (22). For both of these negative studies, a major potential limitation was the inability to identify early signs of patient deterioration when interventions might be most helpful. The importance of this is noted in a recent report suggesting that 2900 US hospitals now have rapid response systems in place without clear demonstration of their efficacy (23). Although early warning scoring systems and critical care teams are important for providing acute care to ward patients and making decisions to transfer patients to a higher level of care, early identification of patient deterioration still needs improvement. Currently, scoring systems are usually initiated and calculated by the ward staff and could either be performed too late or not at all (15). In a recent study by Donohue et al (24) looking at track and trigger system failures in a critical care outreach service, they reported the track and trigger scoring system was used most often to quantify the patients clinical deterioration rather than to identify the initial signs of deterioration. The computerized system used in our study provides a potential solution to issues such as this by using an automatic screening tool that prompts further assessment by the generated alert that is delivered real-time to the bedside care provider. The sepsis PT developed at our institution is essentially an early tracking method that provides an algorithm that can be used as an automated screening tool for hospitalized patients with impending sepsis (8). It provides a simple method that can be used with an informatics-based system to predict the onset of sepsis before overt clinical signs and symptoms are recognized. Our method of implementation included a text page sent directly to the nursing staff to prompt patient evaluation and physician notification, although other forms of communication could be used. The main limitation of the PT is its low positive predictive value. When the prediction tool was developed from a series of patients from 2005 and validated against cohorts from 2006 and 2007, the positive predictive value of identifying a patient that transferred to the ICU secondary to severe sepsis or septic shock was found to be 19.5% with a negative predictive value of 95.8% (8). However, refinements in the accuracy of such PTs over time may improve their diagnostic accuracy. Such refinements could include the use of health information technology bundles with remote physician coverage as was recently shown to improve resource use and outcomes among ICU patients in an academic-affiliated community hospital (25). Limitations. The main limitation of our study is the relatively small number 472

of patients making up the sepsis cases in our model validation study (8) and the IG in the present study. Our study is also underpowered to detect a mortality difference given the observed mortality rate of approximately 10% in this patient population. The small sample size may also explain the diagnostic heterogeneity we observed in the incidence of sepsis, severe sepsis, and septic shock between the study groups. Similarly, the low overall predictive accuracy for the PT suggests that this model may not be good enough for routine clinical use (8). This is supported by the lack of demonstrable clinical benefit resulting from implementation of the PT in the IG. The narrow group of patients with sepsis evaluated in these studies also suggests that the PT may be too limited in terms of its scope for identifying a broad group of at-risk patients. To improve the predictive accuracy of the PT, we are in the process of refining and recalibrating our PT using a more general hospital ward population, including surgical patients, and adding real-time vital signs captured continuously with a locally developed wireless monitoring device. Another important limitation of our study is that we examined all interventions equally in determining the effectiveness of the PT. It is likely that certain interventions such as fluid resuscitation of shock and appropriate antibiotic therapy are more important than others. A PT that allows earlier use of such targeted interventions may be more clinically relevant than one that simply allows any intervention to be applied. Additionally, our study was not blinded, which may have allowed unforeseen biases to influence the results. Finally, we cannot exclude an education effect from having occurred as a result of our training of the nurses and physicians on the intervention wards. This may have accounted, at least in part, for the greater use of specific therapies such as antibiotic escalation and intravenous fluids compared with the nonintervention wards. CONCLUSION Implementation of a real-time computerized sepsis alert on hospital medical wards resulted in an increase in early interventions, including antibiotic escalation, intravenous fluids, oxygen therapy, and diagnostics in patients identified to be at risk for sepsis. However, there was no improvement in patient outcomes or lengths of stay with use of this computerized alert. Based on this pilot study, a larger, randomized controlled trial is planned at our institution to determine whether patient outcomes can be improved with use of a sepsis alert. REFERENCES 1. Ferrer R, Artigas A, Levy MM, et al: Improvement in process of care and outcome after a multicenter severe sepsis educational program in Spain. JAMA 2008; 299:2294 2303 2. Rivers E, Nguyen B, Havstad S, et al: Early goal-directed therapy in the treatment of severe sepsis and septic shock. N Engl J Med 2001; 345:1368 1377 3. Micek ST, Roubinian N, Heuring T, et al: Before after study of a standardized hospital order set for the management of septic shock. Crit Care Med 2007; 34:2707 2713 4. Dellinger RP, Levy MM, Carley JM: Surviving Sepsis Campaign: International guidelines for management of severe sepsis and septic shock: 2008. Crit Care Med 2008; 36: 296 327 5. Thiel SW, Asghar M, Micek ST, et al: Hospital-wide impact of a standardized order set for the management of bacteremic severe sepsis. Crit Care Med 2009; 37:819 824 6. Lundberg JS, Perl TM, Wiblin T, et al: Septic shock: An analysis of outcomes for patients with onset on hospital wards versus intensive care units. Crit Care Med 1998; 26: 1020 1024 7. Young MP, Gooder VJ, McBride K, et al: Inpatient transfers to the intensive care unit: Delays are associated with increased mortality and morbidity. J Gen Intern Med 2003; 18:77 83 8. Thiel SW, Rosini JM, Shannon WA, et al: Early prediction of septic shock in hospitalized patients. J Hosp Med 2010; 5:19 25 9. Rangel-Frausto MS, Pittet D, Costigan M, et al: The natural history of the systemic inflammatory response syndrome (SIRS). A prospective study. JAMA 1995; 273:117 123 10. Brun-Buisson C, Doyon F, Carlet J, et al: Incidence, risk factors, and outcome of severe sepsis and septic shock in adults. A multicenter prospective study in intensive care units. JAMA 1995; 274:968 974 11. Knaus WA, Draper EA, Wagner DP, et al: APACHE II: A severity of disease classification system. Crit Care Med 1985; 13: 818 829 12. Kollef MH, Morrow LE, Niederman MS, et al: Clinical characteristics and treatment patterns among patients with ventilatorassociated pneumonia. Descriptive findings from the Assessment of Local Antimicrobial Resistance Measures (ALARM) study. Chest 2006; 129:1210 1218 13. Micek ST, Welch EC, Khan J, et al: Empiric combination antibiotic therapy is associated with improved outcome in gram-negative sepsis: A retrospective analysis. Antimicrob Agents Chemother 2010; 54:1742 1748 14. Levy MM, Dellinger RP, Townsend SR, et al: The Surviving Sepsis Campaign: Results of an international guideline-based performance improvement program targeting severe sepsis. Crit Care Med 2010; 38:367 374 15. Subbe CP, Kruger M, Rutherford P, et al: Validation of a modified early warning score in medical admissions. Quarterly J Med 2001; 94:521 526 16. McQuillan P, Pilkington S, Allan A, et al: Confidential inquiry into quality of care before admission to intensive care. BMJ 1998; 316:1853 1858 17. National confidential enquiry into patient outcomes and death: An acute problem? A report of the National Confidential Enquiry Into Patient Outcomes and Death. London, 2005 18. Priestley G, Watson W, Rashidian A, et al: Introducing critical care outreach: A wardrandomised trial of phased introduction in a general hospital. Intensive Care Med 2004; 30:1398 1404 19. Pittard AJ: Out of our reach? Assessing the impact of introducing critical care outreach service. Anaesthesiology 2003; 58:882 885 20. Ball C, Kirkby M, Williams S: Effect of the critical care outreach team on patient survival to discharge from hospital and readmission to critical care: Non-randomised population based study. BMJ 2003; 327: 1014 1016 21. Hillman K, Chen J, Cretikos M, et al: Introduction of the medical emergency team (MET) system: A cluster-randomised control trial. Lancet 2005; 365:2091 2097 22. Gao H, Harrison DA, Parry GJ, et al: The impact of the introduction of critical care outreach services in England: A multicentre interrupted time-series analysis. Crit Care 2007; 11:R113 23. Schneider ME: Rapid response systems now established at 2,900 hospitals. Hospitalist News 2010; 3:1 24. Donohue LA, Endacott R: Track, trigger and teamwork: Communication of deterioration in acute medical and surgical wards. Intensive Crit Care Nursing 2010; 26:10 17 25. McCambridge M, Jones K, Paxton H, et al: Association of health information technology and teleintensivist coverage with decreased mortality and ventilator use in critically ill patients. Arch Intern Med 2010; 170: 648 653 473