SEPSIS is a life-threatening condition that occurs when the body responds to an infection in a manner that injures its own tissues and organs. Sepsis can lead to shock, multiple organ failure, and death, especially if it is not recognized and treated promptly. Sepsis remains the primary cause of death from infection despite medical advances that include vaccines, antibiotics, and acute care; millions of people around the world die of sepsis each year.1
Often, sepsis is not recognized as a medical emergency until a patient's condition starts to decline. This means that all health care providers must have a role in identifying early sepsis. Nurses and physicians alike should be monitoring their patients for signs of systemic inflammatory response syndrome (SIRS) and organ dysfunction. Current initial, life-saving treatment options for sepsis do not require special skills or equipment, and they can help reduce patient mortality.2
The "Surviving Sepsis Campaign: International Guidelines for Management of Severe Sepsis and Septic Shock: 2012" recommends routine screening of potentially septic patients. In fact, the Surviving Sepsis Campaign Guidelines Committee suggested that the greatest improvement in outcomes can be made through education and process changes targeting those who care for patients with severe sepsis in non-intensive care unit settings and across the spectrum of acute care.3 Therefore, early warning systems based on physiological parameters may provide timely recognition of patients with potential or established critical illness regardless of their location and, ultimately, support timely and aggressive management of these patients.4
BACKGROUND
Because of its aggressive, multifactorial nature, sepsis is a rapid killer.5 Mortality associated with severe sepsis and septic shock remains unacceptably high, with around 30% of patients dying within the first month of diagnosis and 50% dying within 6 months. Therefore, the rapid diagnosis and management of sepsis are vital to successful treatment.6
Thus, early identification of sepsis and reduction in sepsis mortality rates were the primary factors behind the development in 2011 of a quality improvement (QI) initiative-the Sepsis Alert Program-at Spectrum Health, a not-for-profit health care system with headquarters in Grand Rapids, Michigan. Spectrum Health includes 9 hospitals, nearly 2000 licensed beds, and more than 130 ambulatory and service sites in the West Michigan region. It is the area's largest health care employer, with 1500 medical staff members and 19 100 employees.
In 2010, Spectrum Health identified sepsis mortality rates that were unacceptably high throughout its Grand Rapids hospitals; they had reached 16.23%. In addition, the organization did not have a standardized tool to identify and care for this patient population. The sepsis team at Spectrum Health was asked to develop a tool to identify patients before they become septic, or earlier in the course of their decline, and to provide standardized treatment.
Program creation
From Spectrum Health's large sepsis team, a core team of 15 members was formed to work on the new Sepsis Alert Program. Members included a clinical nurse specialist, physicians, critical care nurse rounders (CCNRs), nurse educators, nurse managers, nursing directors, an information services architect, an application systems analyst, a project manager, and QI specialists. This core team was responsible for developing the program framework, implementing the program, and making ongoing changes to the sepsis alert process.
In preparation for program design and development, the core team viewed webinars on sepsis intervention and treatment and reviewed the "Surviving Sepsis Campaign: International Guidelines for Management of Severe Sepsis and Septic Shock: 2008."7 The team also reviewed information about St. John Sepsis agent, a proven algorithm that continuously monitors key clinical indicators and attempts to recognize a potentially septic pattern.8
Following this research, the team identified the primary goal of the Sepsis Alert Program: to rapidly identify and provide intervention and treatment to hospitalized patients in the early stages of sepsis, before they could progress to severe sepsis or septic shock. The team members decided that they would develop a computer-generated alert system, modeled after the St. John Sepsis agent, which would identify patients with early-stage sepsis and automatically flag the nursing and physician staff. Using a computer-based algorithm to assist in the diagnosis of complicated illnesses has been of interest to many physicians and hospitals; this approach has the ability to accurately and precisely manage clinical interventions, often more consistently than a bedside clinician.9-11
Program development focused on Spectrum Health's medical-surgical units rather than its intensive care units. It was hoped this alert/warning system would clearly identify septic patients, lead to earlier interventions, prevent transfers to a higher level of care, and reduce mortality.
On the basis of recommendations from the 2008 Surviving Sepsis Campaign and the St. John Sepsis agent, the core team defined 17 vital signs, patient assessments, and laboratory values that would be used to trigger an electronic sepsis alert (Table 1). At this point, the information services architect built 3 sepsis PowerPlans for the patient electronic medical record (EMR) system (ie, care planning tools that would support the management of orders, outcomes, and interventions as they relate to predefined plans of care).12 The 3 PowerPlans were based on risk stratification and coded as yellow for sepsis, orange for severe sepsis, and red for septic shock (Table 2). The PowerPlans also contained recommendations for antibiotics based on the suspected source of sepsis: health care/community-acquired pneumonia, ventilator-associated pneumonia, nosocomial meningitis, community meningitis, health care-associated urinary tract infection, community-associated urinary tract infection, intra-abdominal infection, skin/soft-tissue infection, neutropenic fever, and unknown source.
A sepsis MPage was also created by the information services architect-a Web-based platform that enabled the organization to create custom workflows within the EMR and pull data in real time as needed.13 The MPage was specifically designed to show patient assessment data and interventions completed if a physician activated a sepsis PowerPlan. Finally, the initial sepsis alert software was built for the EMR by the application systems analyst.
As the software was being developed, the core team turned its attention to how the sepsis alert would be used in daily practice. The team initially decided that the Sepsis Alert Program would send a message to the pagers of all rapid response (RAP) team members every time a patient triggered a sepsis alert and that the entire team would respond. Because the RAP team is made up of a physician, a pharmacist, a respiratory therapist, and a CCNR, it was believed that patient assessment and intervention (eg, the administration of antibiotics) would be faster.
Pilot program
In May 2011, the core team piloted the new Sepsis Alert Program at one hospital on a digestive diseases medical-surgical unit. The alert was turned off after only 3 weeks. The major problem was alert fatigue due to lack of specificity with the alert. The core team heard many complaints from the RAP members, who felt that they were being pulled away from their other job duties to respond to sepsis alerts on patients who were not septic. After reviewing the data available, 6 major changes were planned for the alert process:
* Tighten the parameters on several of the vital sign and organ dysfunction triggers.
* Exclude patients on dialysis, in comfort care, and in surgery.
* Change the initial response from the entire RAP team to just the CCNR on duty.
* Develop a Sepsis Alert Tool for the CCNRs to complete after each sepsis alert.
* Allow the CCNRs to order a lactic acid level based on their nursing assessment, with cosign of the order by the core team physician.
* Adjust the blood glucose-level parameter to exclude patients who were receiving total parenteral nutrition because their blood glucose level might be falsely elevated.
The core team's decision to use CCNRs as "first responders" to sepsis alerts was made to directly address the RAP team members' concerns. At Spectrum Health, the CCNRs are a team of critical care nurses who are assigned to round through the medical-surgical and progressive care units during their shift. These CCNRs are available 24/7/365 to respond to patient emergencies, and they do not have a patient assignment. Thus, these critical care nurses were already assisting with declining patients and critical situations. The core team believed that a trained CCNR could quickly assess the conditions surrounding a sepsis alert and determine whether the emergent intervention of a RAP team was required.
Program rollout
After these changes were made to the Sepsis Alert Program, the core team turned on the alert process in the background on 3 medical-surgical units that were designated to eventually "go live": a neurology unit, a geriatric unit, and the digestive diseases unit used in the original pilot program. Several months of behind-the-scenes evaluation followed. Medical record reviews were conducted on the patients who triggered a sepsis alert during those months as the QI specialist looked for trends and unforeseen problems. Every effort was made to ensure that the second attempt would be a success.
Finally, September 18, 2012, was selected as the rollout date for the 3 designated units. Although this rollout was, in essence, another pilot program in relation to the entire health system, the team chose not to call it a pilot. It was decided that the alert system would not be turned off again; instead, additional changes would be made in the live environment.
Prior to this September rollout date, while the alert process was operating in the background, nurses on the neurology and geriatric units were educated about sepsis, the sepsis alert, and their role in the process. The nursing staff on the digestive diseases unit were updated about the sepsis alert changes and were given a shorter education course about sepsis because they had received sepsis education prior to the original pilot. The physician staff were also given an update about the changes to the alert and what their role would be when a sepsis alert was triggered. Posters were created for the nursing units describing the sepsis alert and the process to be followed after a patient had triggered a sepsis alert. Finally, the CCNRs were updated about the alert changes and what their role would be, and they were briefed about the roles of the physicians, bedside nurses, and charge nurses.
PROGRAM STRUCTURE
Current alert process
Under the Sepsis Alert Program, an electronic sepsis alert is triggered when a patient admitted to one of Spectrum Health's 3 "live" units has a minimum of 2 signs of SIRS and 1 sign of organ dysfunction within a 24-hour period. When the patient meets these criteria, the computer program sends an alert to the CCNR phone, the unit charge nurse phone, and the RAP physician pager, along with an e-mail message to the QI specialist. The program also generates an order for a sepsis assessment and activates the sepsis MPage in the EMR. Once a patient has triggered a sepsis alert, another alert will not be triggered for 24 hours.
Roles and responsibilities
When a sepsis alert is triggered, the bedside nurse meets with the charge nurse and the CCNR and helps complete an initial assessment of the patient. The bedside nurse performs any interventions ordered by the RAP physician and also explains care to the patient and family.
The charge nurse meets with the bedside nurse and the CCNR and helps complete an initial assessment of the patient. The charge nurse also assists in performing any interventions ordered by the RAP physician.
The CCNR meets with the unit's charge nurse and the patient's bedside nurse to evaluate the situation. Specifically, the CCNR reviews the sepsis alert trigger(s), reviews current laboratory results, and assesses the patient. If the CCNR believes that the patient needs intervention or physician evaluation, the RAP physician is called for a bedside evaluation and consultation. After the sepsis alert event is over, the CCNR makes a note in the "Nursing Progress Note" section of the EMR: "Sepsis assessment triggered, PowerPlan initiated" or "Sepsis assessment triggered, PowerPlan not initiated." If a PowerPlan is not initiated, the CCNR will document any interventions the RAP physician ordered. The CCNR also completes the Sepsis Alert Tool (see Supplemental Digital Content Figure 1, available at: http://links.lww.com/JNCQ/A43) and submits it to the QI specialist for data analysis.
The nursing assistant obtains a new set of vital signs and gathers any needed supplies. If necessary, the nursing assistant also transports STAT blood work to the laboratory.
At Spectrum Health, the RAP physicians are a team of hospitalists specializing in internal medicine. They are in the hospital 24/7/365 to complete admissions and respond to emergencies. When a sepsis alert is triggered, the RAP physician on duty will respond to the alert if requested by the CCNR. The RAP physician will assess the patient along with the CCNR, order any interventions, and activate a sepsis PowerPlan if necessary. The RAP physician will then hand sepsis care over to the attending service, communicating to that team what has occurred.
OUTCOMES MEASUREMENT
Many discussions were held about how to track the effectiveness of the Sepsis Alert Program. The core team brainstormed about which metrics should be tracked and who should be responsible for tracking these metrics. As changes were made to the alert process after the pilot program, it became obvious that the CCNRs, as first responders, would be overseeing the alert process from beginning to end and would be involved in every sepsis alert. Also, they could assess the situation with an unbiased opinion because they would not know individual patients. Thus, the team concluded that not only the CCNRs needed to be the first responders but they also needed to activate the first step in data collection.
Sepsis alert tool
At the same time, after the pilot program, the core team decided to use the Model for Improvement, developed by Associates in Process Improvement and promoted by the federal Health Resources and Services Administration.14 Ultimately, the need for a Sepsis Alert Tool (see Supplemental Digital Content Figure 1, available at: http://links.lww.com/JNCQ/A43) emerged from this process.
The Sepsis Alert Tool currently focuses on answers to the following questions:
* Is this a change from the patient's baseline?
* Does the patient have other organic causes for symptoms of SIRS and organ dysfunction?
* Did you need to contact the RAP physician about this patient? Why or why not?
* Does the patient have a known or suspected infection?
* Were any interventions performed?
Definition of successful alert
Initially, the core team defined a sepsis alert as "successful" if it triggered on a patient who was truly septic and required the initiation of a sepsis PowerPlan. However, the team soon realized that this goal was not realistic because the sepsis alert was triggering on many patients who were not septic but did require some sort of intervention. The team then redefined a successful sepsis alert to include any of several outcomes: if the patient required some sort of intervention from a physician or the nursing staff; if the patient was experiencing a change from the baseline; or if the patient had a discharge diagnosis of sepsis, severe sepsis, or septic shock.
Data collection
In response to a decision by the core team, a QI specialist is responsible for the Sepsis Alert Program's data collection. When a CCNR sends a completed Sepsis Alert Tool, the improvement specialist records that information in a spreadsheet. Then additional data are collected about each sepsis alert.
Every time a sepsis alert is triggered, an e-mail message with the alert information is sent to the QI specialist, who logs it into the spreadsheet. This e-mail contains the date/time of alert, the patient's name, the medical record number, and a statement about what triggered the alert. Using this initial information logged into the spreadsheet, the QI specialist then conducts patient medical record reviews several times a week to capture more data about all of the sepsis alerts triggered to date. Data are collected about the following for each patient:
* What was the patient's age and sex?
* What was the patient's hospital length of stay?
* What was the admission source? (If admitted through the emergency department [ED], how soon after admission was the sepsis alert triggered?)
* Did the patient have a surgical procedure?
* Did the patient require transfer to a higher level of care?
* Did the patient expire?
* On what day of the week was the alert triggered?
* What was the patient's discharge diagnosis? (This data point is gathered bimonthly because it is not available until the patient has been discharged.)
PRELIMINARY RESULTS AND DISCUSSION
After several months, enough Sepsis Alert Tool documents and sepsis alert data were gathered to trigger an overall evaluation of the program. Specific patient trends were identified. The data also showed that further changes were needed, including modifications to the sepsis alert process and the EMR and enhanced education for the nursing staff.
General observations
In the first 8 months of the Sepsis Alert Program, there were 995 sepsis alerts triggered on 617 patients admitted to the 3 units. An alert was triggered on average 4 times a day (maximum = 11; minimum = 0). Alerts were evenly spread throughout the day but not throughout the week; Wednesdays, Thursdays, and Saturdays tended to be higher alert days. Only 22% of the sepsis alerts (n = 217) were triggered by patients who had gone through a surgical procedure.
During data review, it was found that patients admitted to 1 of the 3 units via the ED were 3 times more likely to trigger a sepsis alert than those who were admitted directly to an inpatient unit (total ED alerts, n = 738 [74%]; total direct admission alerts, n = 257 [26%]). Thirty-eight percent of the patients (n = 280) admitted through the ED triggered a sepsis alert within the first 23 hours after admission, with 20% (n = 149) in the first 6 hours. This led to discussions about patient placement, timeliness of admission orders, and transitions of care between the ED and inpatient units.
As noted previously, each patient needed to meet at least 3 criteria to trigger a sepsis alert: 2 signs of SIRS and 1 sign of organ dysfunction within a 24-hour period. However, some patients met the guidelines for up to 6 or 7 criteria during 1 sepsis alert. When the data were analyzed for all 617 patients for the 17 possible criteria that could trigger an alert (Table 1), the top 5 triggers were as follows:
1. Pulse (n = 776)
2. White blood cell count (n = 587)
3. Respiratory rate (RR) (n = 383)
4. Glucose level (n = 355)
5. Level of consciousness (LOC) (n = 352)
In-depth examinations of the data for 2 of the top 5 triggers eventually led to a change in either the sepsis alert process or nursing practice.
Respiratory rate
When evaluating the RR results, it was discovered that the majority of patients triggered at a rate of 22 or 24, and it also seemed that the patients almost always breathed at an evenly numbered rate. The problem was finally identified as an issue not with the sepsis alert but with the nursing assistants' and nursing technicians' accuracy taking an RR during vital signs checks. This issue was taken to the nursing education department, and additional vital signs education was planned for all nursing assistants and nursing technicians.
Level of consciousness
Change in LOC is a good indicator of organ dysfunction, but it is a subjective assessment. Two nurses could evaluate the same patient and come up with 2 different assessments. When nurses chart "confused," "restless," "agitated," "listless," "unresponsive," or "obtunded" in their neurological assessments, those are triggers for an alert firing under the Sepsis Alert Program.
While exploring the data, it was discovered that many of the false alerts were triggered from LOC reporting. For instance, the geriatric medical-surgical unit has a large population of patients with dementia, Alzheimer disease, and baseline confusion. Also, many of these patients experience sundowning (ie, confusion at night). The sepsis alert software is not yet sophisticated enough to distinguish between chronic and acute confusion. Thus, to help fix this problem, the LOC alert trigger was modified to include only results from the last 12 hours instead of the last 24 hours. This change reduced some of the false alerts. However, there are still many that occur and further work is in process.
Physician intervention
There was early concern from the RAP physicians about their role in the process in relation to primary care services. The RAP physicians believed that they might become involved in a patient case when they should not do so. To help with this process, education was conducted with the RAP physicians and the primary care teams on the 3 units involved. Both groups were encouraged to view a sepsis alert as an emergent situation; after the sepsis alert process was over, the RAP physician would contact the primary team and turn sepsis care over to it.
Patients in decline
Only 17% (n = 102) of the 617 patients who triggered a sepsis alert during the first 8 months of the program had a discharge diagnosis of sepsis, severe sepsis, or septic shock. This number was not encouraging until it was discovered that 40% of the 995 sepsis alerts (n = 394) had required an intervention of some kind. These interventions included evaluation by a physician, intravenous fluids, laboratory work, medication administration, respiratory support, and/or transfer to a higher level of care.
After further evaluation of the data and discussions with the core team, it was determined that the Sepsis Alert Program was, in part, serving as an alert program that identified patients in decline. The population of patients with sepsis happened to fit into this larger group. After discussing the data and further exploring the alerts, it was determined that the sepsis alert was having a positive effect on patient care: patients who needed additional medical attention were being identified more quickly than usual.
Keeping the original goal of the Sepsis Alert Program in mind, the core team redirected its focus to refining the alert process so that it would more accurately identify the sepsis population. After a follow-up review of the 2012 Surviving Sepsis Campaign guidelines,3 the team realized that many of the campaign's vital sign and laboratory value parameters were not the same as those adopted by Spectrum Health. Work is underway to make the sepsis alert criteria fit more closely with the Surviving Sepsis Campaign guidelines, with the hope that this will more accurately identify patients with sepsis.
Project limitations
The Sepsis Alert Program is still under development and is currently limited by low sensitivity and specificity. The core team has not been able to expand the program to additional nursing units because the increased sepsis alert volume does not allow CCNRs enough time in their shifts to adequately assess each patient with sepsis alert and complete their other job responsibilities. This inability to add additional nursing units currently limits program evaluation to a comparatively small patient population with a limited variation in medical diagnoses.
CONCLUSION
Spectrum Health's Sepsis Alert Program has been moderately successful to date in rapidly identifying patients whose condition is declining. Additional work is needed to make the sepsis alert process more sensitive and specific and to implement the program on additional units. Although the program is considered "a work in process" at this time, it is having a positive effect on patient care. Other health care organizations might find it useful to establish a similar program, tailored to their specific needs and care environment.
REFERENCES