Predicting sepsis and its trajectory

Biomarkers, automated systems, and rapid diagnostics.

Although half of all hospital deaths in the U.S. may be linked to sepsis, diagnosing the disease in its early stages remains tricky, and predicting which patients will get severely ill is even trickier.

“A substantial portion of those with sepsis who go on to die in the hospital are not admitted to the ICU immediately because they don't appear to be the sickest patients. It's worrisome that some patients have an underrecognized severity of sepsis that could potentially be preventable,” said Vincent Liu, MD, MS, a research scientist in Kaiser Permanente Northern California's Division of Research in Oakland and an intensivist at KP Santa Clara.

Photo by Thinkstock
Photo by Thinkstock

Challenges include that patients often don't show any outward signs of the condition and that their symptoms of sepsis overlap with symptoms of other comorbid conditions. Another problem is the lack of definitive tests. “There is no one test you can run to predict who will get sepsis. Determining risk requires clinical judgment based on multiple forms of evaluation,” said Matthew Sims, MD, PhD, FACP, director of infectious diseases research at Beaumont Hospital in Royal Oak, Mich.

Researchers are making progress on this front, however, developing electronic tools and evaluating biomarkers to better predict whether a patient will develop sepsis and if so, progress to severe sepsis or septic shock.

Predicting progression

Scoring systems that try to predict how sepsis will progress, including the Acute Physiology and Chronic Health Evalution (APACHE) and Sequential Organ Failure Assessment (SOFA), make good guides but do not stand alone, said Dr. Sims.

“Scoring systems like APACHE and SOFA take aspects of a patients' status into consideration and give a number correlated to the rate of survival, but they still require clinician evaluation of the whole patient,” Dr. Sims said.

Research over the last 20 years has also focused on serum lactate, procalcitonin, and C-reactive protein as potential biomarkers of sepsis, with serum lactate currently the “go-to.”

“We know that inadequate evidence of improvement in early therapy is a predictor, [particularly] that if the serum lactate doesn't go down the patient has an increased risk of mortality,” said Dr. Liu. Based on evidence of its prognostic significance, lactate measurement is now a component of the Surviving Sepsis bundle.

Research on procalcitonin has yielded mixed results, but a meta-analysis in the June 15 PloS One that evaluated 23 studies concluded that an elevated procalcitonin level in adults with sepsis was associated with a higher risk of death—although the authors stopped short of defining an optimal cutoff point for risk assessment.

The effectiveness of C-reactive protein as a predictor is uncertain, as well, with a study in the April 2012 Emergency Medicine Journal finding that it did not predict 28-day mortality in ED patients with sepsis as well as the Mortality in Emergency Department Sepsis score. Yet a study in the November 2012 Multidisciplinary Respiratory Medicine noted that C-reactive protein appeared to be as valuable a predictor of mortality as the SOFA score, especially when third-day values were higher than 100 mg/dL.

A study published in the November Critical Care Medicine identified heparin-binding protein as a potential new biomarker. Of 674 emergency department patients with sepsis, 487 did not have organ dysfunction at presentation, but 141 (29%) went on to develop organ dysfunction within 72 hours. Of those 141 patients, 78% had an elevated plasma heparin-binding protein level, defined as higher than 30 ng/mL. The protein was elevated a median of 10.5 hours prior to development of organ dysfunction.

“This looks potentially very interesting and helpful, but it's an early study in only a few hundred patients,” said Dr. Sims.

Another study identified bandemia as a potential predictor of sepsis progression. Among 1,316 patients who had sepsis within 4 hours of emergency department arrival, those who had bandemia of at least 10% were more than twice as likely to progress to septic shock between 4 and 48 hours later, according to results published in the May Critical Care Medicine.

“Patients who have more bands are more likely to be infected, but not every place checks bands,” said Dr. Sims. “It tends to be a manual test, and there's some data [in the literature] to suggest that people are not great at doing them.”

The most effective route for prognostication may be a combination of vital signs and lab values that point to the severity of systemic inflammation, said David F. Gaieski, MD, associate professor of emergency medicine, vice-chair of resuscitation services, and director of emergency critical care at Thomas Jefferson University in Philadelphia. “Heart rate, white blood cell count, temperature, systolic blood pressure, serum lactate, and in some cases urine output can all combine as a measure of severity of illness,” he said.

Using technology to predict risk

Technology will play an increasing role in predicting risk and progression of sepsis, noted Marin H. Kollef, MD, FACP, professor of pulmonary and critical care medicine and director of both the medical intensive care unit and respiratory care services at Barnes-Jewish Hospital in St. Louis.

“There are many opportunities to develop low-cost technology that can be applied in the real-time assessment of vitals, lab values, and other elements of care and then integrate data in a way that will identify who will likely develop sepsis or have an event,” Dr. Kollef said. “There is likely to be a way to develop informatics systems that use electronic medical records [EMRs] to integrate data that gets entered at the bedside and use algorithms to predict risk and events.”

Sepsis-specific systems show promise. A team at the University of Alabama published a paper in the April 10, 2014, PeerJ describing how an EMR-based automated sepsis identification system was able to detect cases of sepsis in the ED. The sepsis detection system triggered a “sepsis alert” if the EMR identified 2 or more of the criteria for systemic inflammatory response syndrome and at least 1 sign of shock. Of 795 alerts, 355 were true cases of sepsis as confirmed by a manual review of physician, nursing, and lab records.

General risk evaluation systems that incorporate scoring systems may play a role, too. Kaiser Permanente Northern California is currently testing such a system, which reviews the EMR of every patient in the general wards and facilities every 6 hours.

The system calculates a risk score for each patient based on his or her vital signs and laboratory values to alert clinicians about patients who might end up having unexpected transitions to the ICU or dying. Although the system is not specific to sepsis, alerting staff to the possible development or worsening of sepsis would be part of it.

“Early results suggest that this warning system will result in improved outcomes, and if that's the case, we will roll it out to a larger region,” said Dr. Liu.

The trick is to determine which algorithm and scoring system are the most accurate and efficient. “We're not exactly sure which predictors will give us the signals-to-noise ratio we need to act on a warning. Physicians, nurses, and other clinicians are all extremely busy, so it's important that the system has more signal than noise. If there's too much noise, that could result in worsened care because it could take time away from actually delivering care,” Dr. Liu said. “We need definite proof of an effective scoring system because the last thing anyone wants to do is invest in a scoring system that does not actually prevent any poor outcomes.”

Ideally, each hospital would not have to start from scratch developing a system. “Many hospitals don't have the computer science backup to develop their own algorithms and implement early warning systems. They just buy proprietary EMR systems that don't do it,” Dr. Kollef said. “The market will dictate, but I think that proprietary systems will eventually become available that allow individual hospitals to use their own data through EMR or more general algorithms to predict patient deterioration.”

How such a system fits into a hospital's overall technology could be a sticking point, said Dr. Gaieski. “In some hospitals, vital signs are still being obtained manually on most patients. Other hospitals use one system to chart and another to record vitals or orders, and the systems don't talk to each other very well.”

Rapid diagnostics are another future avenue to improve sepsis prediction, according to Dr. Sims. “We're highly dependent on culture technology. You take a blood or urine sample and wait for it to grow. It takes a day to see, a day to identify, and a day to determine antibiotic susceptibility. Meanwhile, the bacteria is also growing in your patient,” he said. “We need a way to get a blood sample and see all of that, quickly.”

Knowing the culprit bacteria could not only speed treatment, but assist clinicians in understanding the risk or severity of progression. “Some pathogens are worse than others. For example, community-acquired MRSAs are particularly bad if they get into the lung. We'd need rapid diagnostics that not only detect the bacteria, but pathogenicity factors,” said Dr. Sims.

He noted that some rapid diagnostics systems are available now, but as with integrating scoring or vitals into EMR, cost can be a hurdle. “First you need the platform to run the test, then you need to address reimbursement. If a test costs $100 to run and the hospital gets back $20, you have to know how it will save money in the long run [through better outcomes] and work in the EMR, so that you're not just adding a new test.”