Getting Sepsis Right

 Sepsis is relevant to almost everything that happens in the hospital. It’s common, infinitely possible, and deadly. Falling behind is unforgiveable, while keeping up necessitates antibiotics — a lot of antibiotics. Both the condition and its treatment have side effects, and the old and infirm are disproportionately affected. Thankfully, a slew of new sepsis studies that address the key questions have been published in mainstream medical journals.

Sepsis, like trauma, has a “golden hour” that drives outcomes. Time is morbidity and mortality (and money), and guidelines recommend sepsis screening on the hospital wards. Two studies using “early warning scores (EWSs)” were published this fall. The first provides external validation and comparison of six different EWSs (there are more than 30 in existence) that vary in complexity. Half the scores used models derived via artificial intelligence (AI), and the remainder did not. The primary outcome variable was clinical deterioration, and a model — called the eCART — constructed using machine learning (considered a form of AI by the authors) outperformed the others. The simple NEWS (National Early Warning Score) and NEWS2 were as good or better than the other two AI models studied. 

This cautionary tale shows that EWSs based on statistically sophisticated modeling can have subpar performance. The old maxim that “all models are flawed but some are useful” holds true in the era of AI. The study authors developed the eCART and hold patents for its use. Its superior performance in their analysis could be serendipity or a preordained outcome due to an obvious bias. It’s unfortunate this conflict of interest garnered no mention in the otherwise excellent accompanying editorial. 

A second study used an electronic health record (EHR) sepsis alert based on the quick Sequential Organ Failure Assessment (qSOFA) score and provided what others in the field have not: a prospective, randomized, and interventional trial. The alert reduced their primary outcome of in-hospital morality — that’s the headline. This finding is huge, but enthusiasm should be tempered by several realities. First, there were defined communication protocols along with training and education prior to implementation, so results weren’t related to the EHR alert alone. Second, it’s not entirely clear what drove the mortality difference and the intervention group had higher rates of Clostridioides difficile infection (expected) and kidney replacement therapies (not expected). A companion editorial provides a nice summary of the study’s other strengths and weaknesses. 

A third paper examined the next phase of sepsis management: antibiotic choice. Most often, empirical antibiotic coverage for suspected sepsis is achieved using vancomycin (V) with either cefepime (V-cefepime) or piperacillin-tazobactam (V-piperacillin-tazobactam. The 2023 ACORN study had me feeling smug and self-celebratory (like an obnoxious Coldplay song) regarding my excessive V-piperacillin-tazobactam use. This paper dents my confidence. 

In a retrospective, observational analysis, the authors found that empirical V- piperacillin-tazobactam was associated with a higher mortality than V-cefepime. V-piperacillin-tazobactam’s unnecessary anti-anaerobic activity was theorized to be responsible for the increase. Discrepancies with ACORN, where there was no mortality difference between V-cefepime and V-piperacillin-tazobactam, are attributed to time points; ACORN measured mortality at 2 weeks vs 90 days for the current study.

Two final studies examined antibiotic duration. The first study, in JAMA, assessed biomarkers— C-reactive protein (CRP) and procalcitonin — to guide antibiotic cessation in patients admitted to an intensive care unit (ICU) with suspected sepsis. The short version is that procalcitonin significantly decreased antibiotic duration by 0.88 day compared with standard of care. This was at the expense of a slight increase in 28-day mortality that fell below the preestablished noninferiority threshold. CRP didn’t provide any value. Procalcitonin-guided antibiotic cessation is already endorsed by the Surviving Sepsis guidelines, but conflicting and poor-quality data required a demure weak recommendation. This paper should strengthen it. 

The second study, in The New England Journal of Medicine (NEJM), compared an arbitrary 7-day vs a 14-day course of antibiotics in patients with bacteremia. They weren’t all critically ill; the median SOFA score for the population was 4, and a little more than half were in an ICU. There was no difference in 90-day mortality — an outcome that was fairly robust across the subgroups analyzed. Taken together, these two studies should result in reduced antibiotic exposure for sepsis and/or bacteremia treatment.

To summarize, while most of the world spent fall of 2024 preoccupied by movie sequels and election drama, the academic critical care community was publishing on sepsis. Early identification remains paramount; how best to achieve it beguiles at best. Edelson and colleagues' study of different early warning scales may comfort the AI luddites, but it provides little else. The randomized qSOFA alert trial is an important step forward but generates lots of questions. The V-cefepime versus V- piperacillin-tazobactam comparison was observational and retrospective but very well done, so much so that it’s pushed me away from V- piperacillin-tazobactam towards V-cefepime, and even further from Coldplay. The physicians in my ICU (including me) don’t use a reflexive 2-week antibiotic course for sepsis and/or bacteremia, but the new studies in JAMA and NEJM make me feel better about it. 

Aaron B. Holley, MD, is a professor of medicine at Uniformed Services University in Bethesda, Maryland, and a pulmonary/sleep and critical care medicine physician at MedStar Washington Hospital Center in Washington, DC. He covers a wide range of topics in pulmonary, critical care, and sleep medicine.

https://www.medscape.com/viewarticle/getting-sepsis-right-2025a10000e6