Amy Carpenter
October 2024—For emergency physician Chadd Kraus, DO, DrPH, of Lehigh Valley Health Network, Allentown, Pa., caring for patients who may have sepsis often goes one of three ways. It’s a “slam-dunk” case, a “sorry, not sepsis” case (a lot was done only to learn it wasn’t sepsis), or it’s a case of subtle or occult sepsis, “which is that we don’t have a great answer. It’s someone in between.”
A good test to him is “increased signal, less noise, because I’m getting a whole bunch of data streamed at me.” Above all is sensitivity. “Less about predictive values, less about specificity. We care a lot about sensitivity.” The reason, he said, is that he has three jobs as an emergency physician: resuscitate, rule out, and not be wrong.
“I don’t have to know that your upper gastric pain is gastritis or GERD, but I do have to know it’s not acute coronary syndrome,” he said.
Dr. Kraus spoke this summer in an ADLM session on sepsis with Alison Woodworth, PhD, D(ABCC), clinical director of global laboratory services, CTI Clinical Trial and Consulting, Cincinnati, and Nathan Ledeboer, PhD, D(ABMM), of the Medical College of Wisconsin. (For Dr. Ledeboer’s comments, see part one.) Dr. Woodworth in her talk spoke about the state of testing for sepsis today.
“We don’t have that ‘slam-dunk’ troponin-type diagnostic marker for sepsis, but I think we’re getting closer to identifying a strategy,” she said.
“We know that sepsis is a severe disorder and responsible for the greatest mortality rate in hospitalized patients,” she said. “So why is it that we don’t have that single biomarker? For heart attack as equally severe, we have troponin. For stroke, we have CT scan.”
One problem: “We haven’t had a consistent definition of sepsis throughout the course of history,” Dr. Woodworth said. The first official definition of sepsis was published in 1992, and in 2001 the sepsis-2 definition was released (with treatment strategies and biomarkers). Sepsis-3 came in 2016. “This pivoted to create a clinical diagnosis,” she said, and expanded treatment strategies. The Surviving Sepsis Campaign’s guidelines put forth additional recommendations based on the evidence for diagnosis and treatment.
Sepsis-1 is “often referred to as the gold standard,” Dr. Woodworth said, and “is the first time we heard the term SIRS [systemic inflammatory response syndrome].” Sepsis is defined as SIRS plus infection—viral, bacterial, fungal, or parasitic.
For sepsis-3, “there was a large population-based study looking at mortality for patients with the signs and symptoms of sepsis,” she said. From that came the definition of sepsis as a life-threatening organ dysfunction caused by a dysregulated host response to infection, as well as clinical parameters, among them the sequential organ failure assessment (SOFA) risk score. The sepsis-3 task force “went on to describe sepsis severity, which included septic shock,” she said, a syndrome associated with hypoperfusion and organ dysfunction and shutdown.
The Centers for Medicare and Medicaid Services, too, stepped in with its SEP-1 measure, bringing with it a different use of lactate, different definitions of septic shock and severe sepsis, and treatment requirements based on severity. “More recently,” Dr. Woodworth said, “these initiatives have been escalated,” with hospitals asked to report for reimbursement purposes more than 60 data elements related to sepsis diagnosis and treatment. “This has led to significant confusion because what’s in the literature versus what hospitals are being asked to report is not necessarily always in agreement.” The CMS released its final rule in 2023, and the SEP-1 measure was added to its Hospital Value-Based Purchasing Program for fiscal year 2026.
The Saving Sepsis Campaign’s 2021 evidence-based screening guideline update rolled back recommendations for the use of laboratory tests like lactate, as well as its recommendation for the use of quick-SOFA. “So, ‘Use some way to figure out if your patient has sepsis, but don’t use what we said in the literature two weeks ago,’” as Dr. Woodworth puts it. “And, yes, you can measure lactate,” but the evidence is low quality and the recommendation is weak, per the update. The 2021 update also contained new recommendations on the timing of antibiotic initiation.
Plenty of requirements for treatment, “and yet no direction on how to diagnose,” Dr. Woodworth said.
Two of the cases Dr. Kraus presented illustrate how difficult diagnosis in the ED can be.
In one case, a 70-year-old male came to the ED via ambulance after having a witnessed syncopal episode. He had an indwelling foley catheter for benign prostatic hyperplasia and was alert but confused on initial evaluation in the ED. “His heart rate is 112, he is in atrial fibrillation, he’s hypotensive, febrile, tachypneic, and hypoxic. This looks like sepsis. We start supplemental oxygen and give him IV fluids, acetaminophen, and ceftriaxone.”
A nurse later tells the emergency physician the patient is struggling to breathe and may need intubation. A point-of-care ultrasound is performed; it shows right-heart strain suggestive of a pulmonary embolism. “In this case, sepsis treatment probably did harm, and care was delayed for treatment of the PE.” A case of “sorry, not sepsis.”
In another case, a 19-year-old male presented to the ED with fever, sore throat, and general malaise, “like every college kid in November, December, January, and February,” Dr. Kraus said. A strep throat test was negative. The patient was a little tachycardic and slightly febrile but otherwise looked good. “He received dexamethasone and is told, ‘It’s pharyngitis, go home, you’re fine.’”
The patient returned 24 hours later and now “looks worse. . . . Now they have sepsis.” Blood cultures came back positive the next day, and it’s a missed case of Lemierre syndrome. “That one is a little more subtle because you see a whole bunch of 18- to 20-year-olds with sore throats. Usually the sore throat has a viral etiology—except those that don’t.” Subtle sepsis is a challenge. Meanwhile, in the face of challenging cases of sepsis, the ED is packed with patients with a variety of other time-sensitive conditions.
In the early stage of sepsis there is an interplay between upregulation of anti- and proinflammatory cytokines, and this interplay can become dysregulated, Dr. Woodworth said.
“This dysregulation can lead to activation of leukocytes and endothelial cell dysfunction, constriction in microvascular flow, and coagulopathy. If left untreated, this leads to tissue injury, organ dysfunction, and ultimately death.”
The sepsis-3 definition comes into the later portion of the sepsis process and looks at sepsis severity, clinical signs, and the patient’s response. “But when we’re talking about patients presenting to the ED, it’s our responsibility to target patients in the early phase of the sepsis pathobiological process,” Dr. Woodworth said. “That’s one of the challenges and one of the reasons why we don’t have troponin for sepsis yet—because there’s been a misunderstanding of where we should be targeting biomarkers.”
From a laboratory perspective, she said, the ideal sepsis biomarker should be specific for sepsis and distinguish sepsis from noninfectious causes of SIRS. “We want something that is sensitive and can pick up sepsis very early in the pathobiological process so we can initiate goal-directed therapy, treat our patient, and improve outcomes.”
The biomarker should be easy for emergency physicians to interpret, be useful for risk stratification and prognosis and for monitoring resuscitation and antimicrobial therapy, and have a rapid turnaround time.
“The standard-of-care laboratory tests”—lactate, procalcitonin, and C-reactive protein—“are what everybody is using. We’re not quite in a place where we’re using next-generation biomarkers like cytokines,” she said.
There’s no role for lactate in diagnosing early sepsis, and it’s not specific because it can be elevated in various settings, Dr. Woodworth noted. But it continues to be recommended in the Surviving Sepsis Campaign guidelines because patients with lactate concentrations greater than 4 mmol/L are more likely to be in a severe state of sepsis.
The SSC guidelines also suggest using lactate clearance—a 10 percent decrease in lactate in the first six hours—to guide resuscitation. “When they use lactate clearance to drive therapy decisions compared to standard of care, the patients whose lactate clearance was monitored did much better when it comes to mortality,” she said (Pan J, et al. Medicine [Baltimore]. 2019;98[8]:e14453).
Procalcitonin (PCT), a precursor to calcitonin, is “proteolytically cleaved prior to secretion from thyroidal C-cells in healthy patients, but in inflammatory conditions, PCT is expressed by nonthyroidal cells and secreted without cleavage of the pro-sequence. In sepsis, PCT expression correlates with disease severity and response to therapy.”
In the ADLM guidance document on the use of procalcitonin, several points were made about its utility (Chambliss AB, et al. J Appl Lab Med. 2023;8[3]:598–634). An extensive review of the literature found that its utility in critically ill patients is its use in conjunction with other signs and symptoms of sepsis to aid in the diagnosis, “but certainly not alone,” Dr. Woodworth, a coauthor, said. For prognosis, “we’re looking for the decline in procalcitonin over time” because it is associated with decreased mortality risk, while steady or rising procalcitonin concentrations are associated with higher mortality risk.
In adult sepsis patients, they found procalcitonin to be useful in monitoring cessation of antimicrobial therapy and to reduce its duration, which is where it’s used most commonly, she said. And the numerous confounding factors warrant the use of clinical decision support tools in conjunction with procalcitonin-based antimicrobial stewardship programs to guide procalcitonin use.
Performance varies for other proinflammatory biomarkers such as C-reactive protein, for proinflammatory cytokines such as IL-6 and TNF-α, and for anti-inflammatory cytokines such as IL-10, Dr. Woodworth said (Meisner M. Clin Chim Acta. 2002;323[1–2]:17–29). “The data is not so good for CRP,” which is not specific for sepsis, lacks sensitivity, and has a long half-life, making it “not the best choice” to monitor response to therapy over time. Cytokines are better biomarkers, particularly in the early phase of sepsis, because “they’re quite sensitive for initial infection and would be a good option for the future.” IL-6 is also an excellent predictor of mortality.
The hematological markers of inflammation include increased numbers of immature granulocytes and abnormally shaped monocytes, and monocyte distribution width has been studied as an early biomarker of sepsis. Dr. Woodworth cites one study whose authors evaluated MDW as a biomarker and compared it with C-reactive protein and procalcitonin in 549 adult patients who presented to the ED and who had blood collected for biomarkers and a CBC within four hours (Woo A, et al. PLoS One. 2021;16[4]:e0250101). The patients were categorized into three groups using sepsis-3: noninfection, infection, and sepsis. The AUC values were MDW (0.71), procalcitonin (0.76), CRP (0.75), and WBC (0.61).
MDW thus “reflected a diagnostic performance comparable to that of conventional diagnostic markers, implying that MDW is an alternative biomarker,” the authors write.
Said Dr. Woodworth, “We have a number of biomarkers that we know change with the pathobiology of the sepsis process, but nothing is standing out as a biomarker that we can use by itself in that early period.”
When Dr. Woodworth was at Vanderbilt University, she and colleagues did research to identify biomarkers in the early pathobiological sepsis process.
“We had a cool electronic alert that allowed us to find patients that had SIRS, and we could go back in time and retrieve the residual plasma and serum specimens sent to the laboratory. So we had the ability to create a biobank of SIRS patients who had well-classified diagnoses for sepsis,” she said, noting that the diagnosis was independently adjudicated by two ICU clinicians. The 12,000 specimens for about 1,500 patients were collected two days prior to and up to five days after the SIRS alert.
“Because many of our sepsis risk prediction scores—for example, SOFA—are based on clinical parameters and laboratory tests, we looked at a combination of clinical and common parameters”—heart rate, respiratory rate, WBC, and others—“with demographics and combined those with some of the most interesting biomarkers in the literature,” including pro- and anti-inflammatory cytokines and CRP. With logistic regression models she and colleagues evaluated the ability to predict early sepsis, up to two days prior to developing SIRS. The combination of biomarkers, demographics, and clinical values was found to have superior diagnostic strength compared with the use of individual biomarkers or clinical values alone.
“When we added hematological parameters in the same patient population, we found the combination of biomarkers, which included baseline characteristics, cytokines, and hematological parameters, significantly improved the ability to predict those patients who would develop SIRS very early in the process,” Dr. Woodworth said. They also compared hematological parameters alone with all biomarkers, “and again this showed statistically significant improved prediction for patients with sepsis.”
Dr. Woodworth created what she calls a sepsis biomarker scorecard (Fig. 1), noting that lactate, procalcitonin, and CRP lack the sensitivity needed to detect early sepsis, but “cytokines are sensitive and hematological parameters can be.” A combination of biomarkers significantly improved the ability to detect patients in the early sepsis phase. Specificity is lacking, but many biomarkers can be used for prognosis or to monitor therapy, “and only a couple right now are standard of care.”
“As more data comes out,” she said, “we will find they will be standard of care very soon.”
Complicating matters is that sepsis isn’t a single syndrome, she said. “It’s a conglomeration of multiple types of patients.”
A machine learning model may be one approach, she said. A meta-analysis of observational studies quantified the performance of a machine learning model to predict sepsis (Islam MM, et al. Comput Methods Programs Biomed. 2019;170:1–9). The authors of the study looked at other sepsis prediction methods: SIRS, modified early-warning system, SOFA, and qSOFA. For machine learning models, they found, the pooled area under receiver operating curve for predicting sepsis onset three to four hours before was 0.89 (95 percent CI, 0.86–0.92), sensitivity was 0.81 (95 percent CI, 0.80–0.81), and specificity was 0.72 (95 percent CI, 0.72–0.72). For SIRS it was 0.70, for modified early-warning system it was 0.50, and for SOFA it was 0.78. They concluded, “Our study findings suggest that the machine learning approach had a better performance than the existing sepsis scoring systems in predicting sepsis.”
“This machine learning algorithm may be something we want to pursue,” Dr. Woodworth said.
Amy Carpenter is CAP TODAY senior editor.