A Response to the NEJM's Sepsis Review Article
There's no test to "diagnose sepsis." What does that mean for clinical studies and health policy?
An excellent review article titled “Sepsis and Septic Shock” was published 4 December 2024 in the New England Journal of Medicine. It’s a wonderful contribution to the literature and well worth reading.
But it feels like something important was missed, or misplaced. Here’s a complementary take from a community physician’s perspective.
Best Guesses
Table 1 (“Sepsis Definitions Over Time”) is illustrative, not to help us identify sepsis, but to survey the decades-long project, so far unsuccessful, to create a formalized definition that can account for sepsis’s notorious variability in presentation and adequately perform in clinical use.
For an engaging digression into that history, read Rafael Olivé Leite:
Although the vagaries of the U.S. health system require us to enter a diagnosis to generate payment by a health insurer, it’s worth emphasizing that sepsis is a syndrome, not a disease.
Case identification (to be distinguished from diagnosis) has historically been made using signs of physiologic stress or organ failure in the setting of a known or suspected infection. The current regime (“Sepsis-3”) discarded the SIRS criteria and endorsed the SOFA score as a measure of organ dysfunction. But there remains no accepted definition of sepsis per se—it is defined (to misuse the word) circularly or self-referentially by its manifestations.
This has important wide-ranging implications. Perhaps sensing this, the review authors completely avoid using the word “diagnosis” in the text. But it’s not until the end of the paper, under the heading “Areas of Controversy or Uncertainty and Future Research”, that they finally say why:
Sepsis is recognized as a syndrome of acute organ dysfunction due to a dysregulated host response to infection. However, we lack a precise definition of the dysregulated host response and a diagnostic test to confirm its presence. Moreover, we have limited ability to confirm or characterize infection in real time.
This is “burying the lede”. The fact that we can’t define sepsis (and never could) is critical because it injects uncertainty (you can debate how much) into every published clinical study (as compared to, say, trials on ischemic stroke, myocardial infarction, or cancer).
Without valid diagnostic criteria, sepsis studies are prone to imprecise or biased findings. In a review, logically inconsistent statements like these then become inevitable:
Up to one third of patients who have been treated for presumed bacterial sepsis had a noninfectious illness in hindsight. Even among patients with sepsis, the cause of the infection is not determined in up to one third of cases.
These are nonfalsifiable claims that illustrate the problem well.
If infection is necessary for case identification, but a cause of infection cannot be established a third of the time in patients “with sepsis” (as determined by some unspecified method since evidence of infection is absent), then the mere suspicion of infection, plus organ dysfunction, can identify a case; but,
If a large proportion of cases can have sepsis with merely the suspicion of infection (i.e., no cause), it cannot also be valid in an equally large proportion of cases to initially identify sepsis based on suspicion, and yet somehow rule it out later to diagnose a solitary noninfectious illness. (Because a patient can present with a noninfectious illness along with sepsis that is then effectively treated, leaving them with the noninfectious illness to which all the diagnostic weight is then erroneously assigned.)
This imprecision and the resultant prevalence of equivocal cases cast a proportion of case identifications into doubt, a proportion whose size is impossible to measure.
This threatens the validity of clinical studies on sepsis generally (at least when milder cases are included) to an unknown (and unknowable) degree.
Or maybe it’s better to say: of course it’s possible to do this, we do it every day, but it’s never more than a best guess.
The “up to a third” language the authors used seems to underestimate their citation’s reported finding: in the 2015 cohort of 2579 patients followed prospectively who were treated for sepsis at two tertiary ICUs in the Netherlands, post hoc adjudication concluded that 13% had no likelihood of infection, while another 30% had no definite evidence of infection (i.e., 43% of patients treated for sepsis lacked a clear cause of infection).
The review authors’ estimate of a ≥67% rate of identifying a cause of infection may likewise be falsely high: data from the randomized PHANTASi trial of 1133 patients suggested that cultures are positive less than half the time in patients treated for sepsis.
Against a syndrome with protean manifestations lacking a clear definition or diagnostic test, sepsis is a guess more often than we admit or realize (at least when evidence of infection is lacking or equivocal). Unsurprisingly, many clinicians feel discomfort at being increasingly pressured to “rule in” or “rule out” sepsis under conditions of high uncertainty and potentially high stakes.
As for the body of research literature, clinical researchers usually do not have access to superior diagnostic techniques. Sepsis research studies are therefore comprised of large numbers of clinicians’ best guesses in varying settings and circumstances. Guesses carry value—they’re what we have to work with—but they should come labeled as such, especially before they’re packaged for consumption by policymakers and hospital administrators.
The uneasy truth is, for the large swaths of patients in these trials who lacked positive cultures, and those who had competing diagnoses for their organ dysfunction, no one ever knew whether they had sepsis or not—before, during or after the trial. This is also true for patients in clinical practice, but on a scale of tens of millions.
From this perspective, statements like “among patients with sepsis [emphasis added], the cause of the infection is not determined in up to one-third of cases” are misleading or even nonsensical.
As anyone who regularly treats people with presumed sepsis knows well, a dense thicket of uncertainty commonly dominates and remains impenetrable throughout the disease course (especially when shock is absent, encephalopathy is the main symptom in a patient with preexisting dementia, there are multiple competing or contributing diagnoses, etc). The massive increase in milder “suspected sepsis” cases resulting from policy efforts at increased case recognition has amplified this phenomenon. (Septic shock cases, and the clinical research based on them, probably represent less of a muddle, but most sepsis trials have also enrolled patients without shock.)
Although it may be uncomfortable for subject matter experts, an acknowledgment of the implications of this uncertainty—clinically and for the research and policy agenda—belongs at the beginning of any serious discussion of sepsis. Such an admission would encourage us all to adopt an appropriately humble and skeptical perspective before making decisions based in the complex and compromised body of clinical research on sepsis.
It would also acknowledge a correlated and important truth: clinicians’ impressions—our best guesses—are still the best diagnostic test for sepsis. The only one, in fact. At least for now, a person “has sepsis” when a clinician believes and says they do.
More to come on this soon (including an actual sepsis update).
Please, share your thoughts below.
Thanks for articulating the contradiction that I have felt for the last decade (or more) as sepsis bundle metrics seem to have become more important than an accurate diagnosis, and actual patient outcomes.
I think the value of the evidence on sepsis is that we can still say that people treated it with this syndrome, whatever you call it, respond or don't respond to the interventions in the trials.