Earlier accurate recognition and treatment of sepsis is the goal of every hospital critical care unit. Tift Regional Medical Center in Georgia recently hired a nurse to serve as a dedicated sepsis coordinator and is getting ready to implement a tool that will offer real-time analytics of structured and unstructured patient data to empower its clinical teams with alerts.
Kaine Brown, M.D., chief medical information officer and medical director of Affinity Hospital Medicine at Tift Regional Medical Center in Tifton, Ga., recently spoke with Healthcare Innovation about their goals in adding real-time artificial intelligence analysis to their arsenal.
Sepsis is the deadliest and most expensive condition treated in hospital critical care units, with septic shock carrying a 34 percent mortality rate. Brown used the term “fog of war,” to describe the challenge critical care clinicians face in catching sepsis early. “There are so many variables that we look at every day — from vital signs to laboratory values to the chief complaint of the patient and their description of their symptoms. Putting it all together is similar to putting the pieces of a puzzle together.” Sometimes the vital signs for a patient in sepsis and a patient who is not in sepsis can be exactly the same, he said.
Research studies have shown that earlier identification and treatment of sepsis reduces mortality significantly, Brown explained. “When you think about a stroke, we often hear that time is brain; when you think about a heart attack, you often hear that time is muscle; and with sepsis time really is life.”
Tift has chosen to leverage Wolters Kluwer’s POC Advisor tool to analyze the hospital’s EHR data with the goal of identifying patients at risk of sepsis more than six hours earlier than EHR-based sepsis alerts. POC Advisor’s Alert Response dashboard will also provide a comprehensive view of Tift’s clinician actions to promote continual improvement, educate staff and reduce unwarranted care variations.
Although Tift does not currently have sepsis alerts built into its EHR, it does have a sepsis order set designed to help it to meet a septic shock management bundle for reporting to CMS. “We track data on the usage of that order set,” Brown said. “There are certain features of that bundle that we track data around, such as obtaining certain laboratory values and performing certain clinical tasks such as administering antibiotics and intravenous fluids. We can drill down into each case to say, yes, we met the bundle or no, we did not meet the bundle. And if we did not, seek to understand why not, and then we provide education to providers, so that we can meet the very complex bundle. It requires a team effort rather than depending on the individual clinician to meet that bundle, and I think that is one of the reasons it is so difficult to meet because it's a team effort and not the actions of one individual.”
Currently Tift is doing retrospective reviews of charts to educate its physicians and nursing staff to improve the quality of care, Brown said. “However, moving forward once we go live with POC Advisor, we're looking to have real-time alerts that would include things like changes in vital signs or laboratory values,” he added. “Some of those alerts would go immediately to the physician and some would go to the nursing staff, who are specifically trained on interpreting that alert, and in discussing that alert with the physician taking care of the patient. So we're hopeful that with these alerts on this dashboard, we can look at this data in more real time as opposed to retrospective view and a delayed interpretation of the vital signs and the laboratory results.”
The solution’s analytic dashboards also will showcase CMS quality performance metrics to reduce practice variation by identifying gaps in CMS SEP-1 bundle compliance.
Wolters Kluwers stresses that its tool not only looks at labs and vital signs, it also pulls unstructured data from clinical notes for assessment. “About 80 percent of the clinical data actually lies within unstructured textual notes, narrations, progress notes, consultant and radiology reports, pathology reports, etc.,” said Itay Klaz, M.D., medical director for clinical surveillance, compliance and data solutions for the company.
“They're not brought in as distinct categorizations of standard data, but more like free-flowing text. And if you are not able to discern the critical elements to identify symptoms as early as possible out of this flow of free text and unstructured data, then you are probably likely to fail in recognizing not only which ones are sepsis, but also which ones are false positives, or potentially ones that can be attributed to other comorbid or baseline conditions of the patient, but do not necessarily define sepsis in the moment.”
Klaz stressed that reducing the false positive rate is very important. “We understand the now decade-old concept of alert fatigue and provider burnout from getting alerts that are not very helpful or actionable,” he said. “If you are a clinician with very limited time and with very sick patients, and you have to deal with so much noise from unnecessary alerts, then even the necessary and true ones can disappear within the crowd of alerts, and that's a major issue for every provider in every organization we work with.”
Tift is planning a staged rollout, Brown said. “We will start on several of the floors in our medical units, and then we will migrate to the emergency department and the intensive care unit,” he said. The roll-out will take place over the next six months. The recently hired dedicated sepsis coordinator is helping with the development of the various protocols, and discerning to which person the alerts will go. “She will also help with the interpretation of the data,” he added, “and hopefully help with some of the real-time decision-making with regard to interventions that can improve outcomes.”
Brown reiterated that the alert fatigue that Klaz mentioned is very real and “we do not want to ignore these alerts. That is why it’s important that when we do receive alerts, they are very accurate. Because of the technology that Wolters Kluwer is using with natural language processing and artificial intelligence, their product will have many less false-positive alerts,” he said. “Basically, the product won’t be crying wolf, so to speak, because of its accuracy. And that's extremely important.”