Introduction
Sepsis is a syndrome that arises when the body’s response to a severe infection injures its own tissues. In 1992 an international consensus panel proposed a clinical definition for sepsis, making use of the concept of a “systemic inflammatory response syndrome” (SIRS), involving alterations in body temperature, heart rate, respiration rate, and leukocyte counts [
1]. The panel defined sepsis as SIRS caused by suspected infection. It further introduced the terms severe sepsis to describe cases when sepsis is complicated by acute organ dysfunction and septic shock as severe sepsis complicated by hypotension refractory to fluid resuscitation. These definitions, generally referred to as the “Bone criteria”, have been used as inclusion criteria in many clinical sepsis trials, and until today have remained largely unchanged [
2].
Although the clinical suspicion of infection is a crucial factor in making a sepsis diagnosis, little is known about the accuracy of this diagnosis in the context of critically ill patients who present to the ICU with signs and symptoms of a “sepsis syndrome”. We hypothesized that in the clinical practice of an ICU the diagnosis of sepsis is not based on strict diagnostic criteria for infection and that as a consequence the occurrence of sepsis on the ICU might be overestimated. Quantification of this discordance is helpful for estimating incidence rates in epidemiological studies and the possible reduction of antibiotic use.
To address this hypothesis we assessed the concordance between the prospective clinical sepsis diagnosis made by bedside physicians and the post-hoc diagnosis of infection made by clinical researchers using strict criteria. In addition, we assessed the association of the likelihood of infection with outcome.
Discussion
We determined the accuracy of the infection diagnosis made by clinicians in the context of presumed sepsis upon admission to the ICU and found that up to 43 % of patients treated for sepsis were unlikely to have had an infection on post-hoc assessment. Although the accuracy of the infection diagnosis increased with increasing severity of disease, a considerable proportion of patients with severe sepsis and septic shock still had at most a possible infection. These results show that making an accurate infection diagnosis upon ICU admission in patients with suspected sepsis is difficult in many cases.
Our study is the first prospective comparison of sepsis diagnoses made by ICU physicians and post-hoc analyses of infection likelihoods based on strict diagnostic criteria, revealing that the true incidence of sepsis upon ICU admission is probably overestimated. Only few previous studies have specifically investigated the accuracy of infection diagnoses in patients with suspected sepsis in the ICU. A French study found that 49 % of patients were potentially unnecessarily treated for a new infection on the ICU [
16]. This finding was based on the level of microbiological evidence and not on well-defined diagnostic criteria, however, making it difficult to appreciate the true percentage of patients without infection in post-hoc analysis. Another study explored the correlation of clinical certainty at the start of antimicrobial therapy with the post-hoc presence of infection [
17]. The primary aim of this latter investigation focused on antimicrobial use, namely how often administration of antimicrobials for suspected infection could be justified by the presence of infection; a large proportion of patients treated with empirical antibiotics (58 of the 125; 46 %) actually had no infection according to the infectious diseases specialist in the post-hoc assessment [
17].
In crude analysis, the likelihood of infection in patients treated for suspected sepsis was not associated with mortality. Since several factors that impact on ICU mortality were unequally distributed between groups, we performed multivariable survival analysis and found that a lower likelihood of infection was associated with increased mortality. In other words, patients who were initially treated for sepsis but had, in retrospect, a noninfectious diagnosis had a higher mortality rate compared with patients with an infection. This observation is probably related to variations in underlying pathology, but may also partly be due to the diagnostic delay that resulted from an incorrect working diagnosis. Furthermore, these data suggest that infection does not result in worse outcome compared with other critical conditions. It is important to note that the increasing incidences of complications such as acute kidney injury and adult respiratory distress syndrome in cases with higher plausibilities of infection are markers of a “correct” sepsis diagnosis and should not be interpreted as causal factors of the lower adjusted mortality rates in noninfectious cases.
As the discrimination between infectious and noninfectious causes of critical illness in the ICU using clinical parameters only has proved challenging, multiple studies have been performed into the value of other markers, such as host biomarkers for the diagnosis of infection [
18,
19]. While some biomarkers, such as procalcitonin, may aid in limiting the duration of antibiotic therapy in ICU patients [
20], at present there are no biomarkers that provide sufficient diagnostic accuracy to withhold antibiotics as initial therapy in ICU patients with suspected infection [
18,
21,
22]. While biomarkers would be valuable for diagnosis in reducing antibiotic use in this patient population, our current study suggests that for stratification according to risk for an adverse outcome, the infection diagnosis itself is less important.
A limitation of this study concerns the inherently somewhat complex CDC and ISF infection definitions used for the post-hoc assessment for the presence of infection. We therefore determined the diagnostic agreement among the study team in a separate study, and concordance was found to be good [
3]. In contrast to this previous study, the current process of prospective surveillance involved discussions among observers, discussions with (senior) clinicians in multidisciplinary meetings attended by critical care physicians and infection specialists, and continuous checks of data integrity. All diagnoses were therefore made after consensus. As such, our post-hoc analyses represent an “ideal” situation with availability of all diagnostic data collected after the acute event. Consequently, our study should not be interpreted as an analysis of the adequacy of clinical action in the ICU, but rather as an attempt to assess the true incidence of infection in patients admitted with suspected sepsis. In this respect it is important to note that in large surveys the rapid administration of broad-spectrum antibiotics to patients with clinically diagnosed septic shock is associated with a time-dependent increase in survival [
23,
24], suggesting that the benefit of early antibiotic treatment in patients with infection is greater than the potential harm of unnecessary antimicrobial therapy in those without infection. Notably, relative to other sources of infection, only few pneumonia cases fulfilled the criteria for definite infection. This was most probably caused by a relatively strict definition for pneumonia [
3]. Furthermore, any systemic use of antibiotics before admission may have influenced culture results obtained in the ICU and therefore also the recorded likelihood of infection. However, a positive culture was not necessarily needed to diagnose a probable infection, including for the most commonly observed community-acquired infection in our study (pneumonia). For patients with hospital-acquired infections, this issue was deemed less problematic since blood (and other) cultures were typically collected before the start of antibiotics. Another limitation involves the fact that this study was performed in two centers in the Netherlands and may not reflect general ICU practice. Lastly, as is true for all observational studies, we cannot rule out the possibility that unobserved confounding might have occurred in the mortality analysis. However, the adjustment methods used were identical for all subgroups.
Conclusions
This first prospective analysis of the accuracy of the infection diagnosis in patients with suspected sepsis on ICU admission shows that the clinical diagnosis of sepsis corresponds poorly with the actual presence of infection, as defined by CDC/ISF diagnostic criteria. These results suggest that the true incidence of sepsis may have been overestimated in many studies. In fact, a substantial portion of patients being enrolled in clinical sepsis trials may in fact not have probable or definite infection, which may negatively impact the power of such trials to show benefit of certain sepsis treatments.
Competing interests
All authors declare that no conflicts of interest exist. The sponsors did not play a role in the design and conduct of the study; collection, management, analysis, and interpretation of the data; and preparation, review, or approval of the manuscript; and decision to submit the manuscript for publication.
Authors’ contributions
PMCKK, OLC, and TvdP substantially contributed to the conception and design of this study. PMCKK, OLC, LAvV, DSYO, JFF, MJS, MJB, and TvdP acquired the data. PMCKK had full access to all data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis. PMCKK, OLC, and TvdP were involved in the interpretation of data. PMCKK, OLC and TPvd drafted the manuscript and all authors revised it critically for important intellectual content. All authors gave final approval of this version to be submitted.