Sepsis research is scary
Sepsis consensus definitions are to blame for a generation's failure in sepsis research
Some expressions trigger anxiety when I am reading research in our beautiful and bewildering specialty. One is sepsis. Another is Bayesian. I never found both words combined. Now you think about combining both words with PERPETUAL. You can imagine how terrified I was at the prospect of a perpetual sepsis study with Bayesian analysis.
It caused me tachycardia, tachypnea, and probably fever, and I can't be sure about my leukocyte counts. After a while, I reassured myself “Thank God now we live under Sepsis-3 and we don't believe in SIRS criteria anymore, otherwise, I'd be kind of half-septic. Now we are smart enough to believe in qSOFA criteria, which are better criteria". But a chilling wave of thoughts seized me again “Well… I'm tachypneic, this is for sure. And if this reaction counts as altered mental status? I have just scored 2 qSOFA points” By this point, as a dutiful intensivist, I should have launched the sepsis protocol on myself. But I confess I didn't. I wasn't in the mood to discuss arterial or venous lactate. Anyway, I had secretly decided I wouldn't let them perforate any of my arteries in the case they presumed I had an infection.
Sepsis-3 introduced qSOFA in 2016 as a predictive tool to identify patients with worse prognoses, and the panelists matter-of-factly informed it was also a diagnostic tool. In 2024, we are still told to believe we can diagnose a disease by its prognosis, but now, the studies based on such diagnosis are perpetual. Dystopian.
Sepsis definitions are not disease definitions. They have always been triage tools.
Since the first consensus definition in 1989, sepsis definitions have been teleological. They exist for one reason: triaging patients for research. The first definition (Sepsis-1) is the inclusion criteria of a methylprednisolone RCT conducted by the group of the late intensivist and professor Roger C. Bone. Sepsis was defined as a complicated infection severe enough to meet the inclusion criteria for methylprednisolone. More than 30 years later we are still triageing patients. However, we mostly lost sight that sepsis is no more than a complicated infection.
A patient has an infection. The infection produces a host response. I like the Sepsis-3 idea of defining sepsis as a dysregulated host response. It is as far as one can go without defining what sepsis is. However, a useful definition should point to a pathobiological hallmark to differentiate sepsis from infection. Until someone finds the biological feature that decisively tells sepsis from infection, sepsis is infection. Hence, the only available treatment is antibiotics.
We need to advance towards a disease model for sepsis that prompts a straightforward strategy, like ST-elevation myocardial infarction. When we find such a model sepsis will be an independent nosological entity. Meanwhile, we have arbitrary triage tools for giving antibiotics now instead of giving them a bit later. The current triage tools are qSOFA and SOFA. They help you decide whether antibiotics are urgently needed but say nothing about the dysregulated immune response.
Some infection cases will be severe enough to be triaged in as a case of sepsis. Here is the fundamental point: if we triage out infected patients who are not bad enough to fulfill sepsis criteria, we lose most infected patients. If we only study patients with sepsis, we won't discover which feature of the dysregulated host response to infection produces sepsis. We just can't. The consequence is the decades-long story of failed treatments.
I have introduced you to a monumental selection bias. It is as pervasive and long-lasting as one can imagine. Medical scientists will never discover how sepsis differentiates from uncomplicated infection without comparing sepsis with uncomplicated infection. A new approach is needed, starting from the pre-clinical research. In the pre-clinical arena, thousands study isolated pathways using healthy animals or healthy volunteers as the control group. The consequences are (1) No one knows how an uncomplicated infection becomes sepsis and (2) Researchers treat each of the uncountable host response pathways as equally capable of provoking sepsis. The field became a marginal utility nightmare. I have been a pre-clinical sepsis researcher struggling to make sense of this mess. Now you understand why sepsis research gives me chills.
Let's come back to the sepsis perpetual study. The PALLETE study, conducted by a French group, is so ambitious I must applaud. It's a pleasure to see ambitious researchers at play. It helps me forget the less ambitious. Moreover, I am ready to forgive the same French group for compelling me to chase "relative adrenal insufficiency" two decades ago.1
The PALLETE study is a platform study for personalized sepsis treatment. To my knowledge, it is the boldest initiative in the history of sepsis research. They plan to divide the patients according to three domains, which one could call endotypes: (1) Hyper or hypo inflammation, (2) Corticosteroid response,2 and (3) Hypercoagulation or hypofibrinolytic state, and then treat the patients focusing on the perceived mechanism. The implications are straightforward. If the approach proves successful, you will order lab tests to determine your patient's endotype and treat him or her accordingly. I place my chips in anakinra, given previous findings.
However, we can't lose sight of the shortcomings of sepsis so-called “definitions”. The PALLETE study uses SOFA score > 1 for patient inclusion. It means they will count only the stories of the patients who fulfilled such criteria. If a patient has a 70 mmHg mean arterial blood pressure he is out. If another patient has 68 mmHg, she's in. However, although both patients may have the same endotype, only one is considered to have a “dysregulated” response. This selection bias is caused by using a triage tool instead of a diagnosis. Researchers should focus on observational studies relating endotypes to infection outcomes, and only then go for a clinical trial. Describe the disease first, Dude!
Some 900 words ago I stated I was scared by the terms sepsis and Bayesian. Here's why. I’m afraid the PALLETE researchers have declared they will use Bayesian analysis as a vaccine to a negative result in frequentist analysis, using the strategy I termed Bayesian Spin. They only need to manipulate the prior probability to offer you something like “We found no meaningful difference, but our Bayesian analysis indicates a high probability of a meaningful difference”. Swallow it with a bit of sodium chloride or lactated Ringer's.
Although biased by the flawed Sepsis-3 definitions, the PALLETE study will be a landmark whatever the results. It may open a new horizon by spotting a treatable endotype or exposing the uselessness of Sepsis-3 "definitions". However, the worst-case Black Mirror-ish scenario is a perpetual Bayesian Spin of negative results to keep the status quo of sepsis scientists forever. Scary?
Share and subscribe!
The ACTH stimulation test became such a fever among intensivists that in a certain week in the early 2000s, we liquidated the ACTH stocks in one of the largest metropolitan areas in Brazil.
Old habits die hard