The Accuracy of Infection Diagnoses Among Patients Meeting Sepsis-3 Criteria in the Emergency Department

To the editor—We read with interest the article by Hooper et al who studied the accuracy of infection diagnoses among patients who met Sepsis-3 criteria in the emergency department [1]. Sepsis is a heterogenous syndrome with multiple noninfectious mimics, and distinguishing infection from noninfection is difficult [2, 3]. Data on rates of infection among patients who present with presumed sepsis can guide clinicians who care for these patients as well as inform policy around early sepsis care [4]. Therefore, studies that accurately determine whether patients with apparent sepsis have infection are crucially important.

A key complicating factor in these studies is that there is no gold standard for infection diagnosis. Short of definitive autopsy findings, expert clinician post hoc adjudication is likely the most accurate method to determine whether a patient was infected. Physicians, even experts who conduct retrospective case reviews, disagree on whether a patient truly had an infection or not [5]. To attempt to mitigate this risk of misdiagnosis, other studies that retrospectively determined the likelihood of infection have used, for example, consensus by a multidisciplinary team (including senior critical care physicians and infectious disease experts) [6], 2-physician independent review (with a third reviewer adjudicating disagreements) [7], and panels comprised of 3 experts [8] as the gold standard.

Here, Hooper et al do not explicitly report how many investigators reviewed each case to determine the primary outcome (infection presence or absence) [1]. However, it seems that the initial review was conducted by just 1 reviewer per patient, and a 10% subset re-reviewed. While we do not doubt that all reviewers were well trained, we are concerned that for 90% of patients, there was only 1 reviewer, and that reviewer was not necessarily a clinician. Even expert clinicians may disagree regarding the probability of infection [5]. Therefore, relying on single-reviewer adjudication is insufficient, especially since the proportion of false-positive infection diagnoses was markedly lower than in similar studies [3, 6]. A sensitivity analysis of the primary outcome (infection present vs absent) considering potential disagreements among multiple reviewers would have been helpful. At the very least, the authors could provide further details on infection vs noninfection adjudication by the second reviewer (aside from just reporting the kappa statistic) in the 10% of cases that were re-reviewed.

In the future, better biomarkers and diagnostic algorithms will hopefully make infection diagnosis easier and more accurate in patients with presumed sepsis. For now, we have to rely on imperfect and subjective measures such as clinician determination. Since these data drive clinical and policy decision-making for nearly 2 million Americans with sepsis per year [4, 9, 10], we should attempt to make them as accurate as possible.

Notes

Author contributions. All authors contributed to writing the manuscript.

Financial support . This work was supported by the National Institute of Allergy and Infectious Diseases at the National Institutes of Health (K24AI121296 to C. A. A.).