Background
Sepsis is a potentially life-threatening condition which is estimated to affect over 30 million people worldwide each year [
1]. Recent data suggest that sepsis contributes to between one third and half of all hospital deaths in the USA and accounts for a significant part of the overall costs of healthcare [
2‐
4]. The incidence of severe sepsis and septic shock in developed countries has been reported as increasing steadily over recent decades; for example, during 2004–2009, the incidence of severe sepsis increased from 300 to 1031 cases per 100,000 population in the USA [
5]. Conversely, several reports from different geographical areas have shown a sequential decrease in sepsis mortality over time. In a large study of European ICU patients over time, the odds of ICU mortality in sepsis patients were lower in 2012 versus 2002 [
6]. Similar to this, a study from Germany based on retrospectively collected administrative data for 2007–2013 found a declining mortality trend, although the incidence was increasing [
7]. A similar trend was also reported for sepsis patients treated in intensive care units in Australia and New Zealand during 2000–2013 [
8], though the authors also identified a substantial variation in mortality risk according to the classification of the diagnosis of sepsis, implying a potential risk that reported temporal trends in sepsis mortality based on administrative data may be biased by misclassification. Results might have been influenced by differences due to increased awareness of the disease or changes in diagnosis coding practices over time, as well as incorrect coding due to retrospectively derived data [
9]. Extracting clinical data from electronic databases seems to be the most unbiased way to study trends in hospital mortality [
10‐
14]. With this approach, previous assumptions of declining in-hospital sepsis mortality in the USA were shown to be partly confounded by increased disease awareness and sepsis diagnosis coding [
10]. However, the accessibility to large sets of clinical data from electronic health records is often limited in many countries outside the USA and only hospital mortality can be studied.
In order to study trends of mortality in patients with severe sepsis with reduced risk of classification bias, Stevenson et al. assessed the trends of sepsis mortality based on data from randomised controlled trials (RCTs) [
15]. They found a decrease in sepsis mortality between 1991 and 2009, consistent with administrative data and supported the use of administrative data to monitor mortality trends in patients with severe sepsis. As this was the first meta-analysis performed with this research question, we wanted to assess if the results were reproducible in a prospectively designed meta-analysis of high-quality sepsis trials, published during a different time interval.
Accordingly, we hypothesised that a trend in the mortality of patients enrolled in the usual care arm of interventional high-quality sepsis studies published between 2002 and 2016 should be concordant with the previously reported trends from RCT data where quality assessments of the included studies had not been performed.
The primary aim of this study was to investigate whether the reported 28-day mortality of patients with severe sepsis and septic shock, included as controls in randomised controlled trials, has declined over time. The secondary aim was to assess how many of the included trials could show any beneficial effect of the studied intervention with regard to 28-day mortality.
According to international guidelines, there is currently no approved sepsis-specific drug or intervention [
16]. In addition to investigating any apparent temporal trends in mortality, we therefore also sought to explore how many studied interventions have shown efficacy in terms of mortality.
The study was registered in the International Prospective Register of Systematic Reviews (
https://www.crd.york.ac.uk/PROSPERO/) and given number CRD42018091100. The Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) Statement and the Meta-Analysis of Observational Studies in Epidemiology (MOOSE) Statement were referred to for methodological guidance [
17].
Discussion
In this study, we performed a meta-regression analysis of data derived from RCTs published between 2002 and 2016 (inclusion start date from 1991 to 2013) after a systematic literature search of studies including severe sepsis and septic shock. Our primary aim was to study the mortality trend of patients enrolled in the control arm. We found an annual decrease of 0.42% in the 28-day mortality among all patients enrolled as controls. Interestingly, the mortality trend was still a significantly decreasing one after adjustments for age and was found to be more pronounced in studies enrolling patients with a mean age ≥ 65 years. Severity scores were not adjusted for in our primary meta-regression analysis as different severity scores were used between the studies. However, when performing subgroup analyses with adjustments for APACHE II, SAPS II and SOFA scores on study inclusion, the mortality trend was no longer significantly decreasing over time. Consequently, the observed sepsis mortality decline in the control arm of RCTs could in part be explained by the inclusion of patients with a lower predicated mortality over time.
In the similar study from 2014 by Stevenson et al., also using data from sepsis patients included as controls of RCTs, a declining mortality trend by 3.0% annually was demonstrated [
15]. The results of the unadjusted analysis in the present study support their results of a declining mortality trend in the control arm of sepsis trials, although an important difference was found as they demonstrated that the declining sepsis mortality was not confounded by changes in disease severity at inclusion. This might be due to differences in the inclusion and exclusion criteria of the studies entered in the meta-regression analysis. In particular, we included trials during a later time span and included studies enrolling patients who were ≥ 16 years old and excluded studies with poor quality. Additionally, Stevenson et al. performed adjustments of the meta-regression from estimations of predicted mortality derived from different severity assessment scales while we performed separate subgroup analyses. This might include errors as SAPS II and APACHE II are shown to differ in the prediction of hospital mortality [
32].
Another important issue that should be discussed in relation to the study by Stevenson et al. is their final conclusion stating that a declining sepsis mortality over time in the control arm from RCTs supports the use of administrative data to monitor trends in sepsis mortality. In our opinion, this may be questioned as several studies comparing clinical data with administrative claims data have shown that the decreases in sepsis mortality over time shown by administrative data are overestimated [
10‐
13].
The so-called Will Rogers phenomenon explains reduced disease-related mortality as a result of increased disease awareness and a higher proportion of less severely ill patients [
33]. Indeed, the Will Rogers phenomenon could also be relevant in the current study due to changes in triage screening systems and to increased focus on early goal-directed treatment; as shown in a retrospective study on trends in usual care of septic shock patients between 2003 and 2013, time to initiation of antibiotics and fluid therapy in the emergency department was shown to be significantly shortened over time [
34]. Several retrospective studies have demonstrated a clear association between early administration of antibiotic treatment and reduced sepsis mortality [
35‐
37]. The results of these studies have led to the implementation of scoring systems for early identification of patients with suspected sepsis [
38] and evidence-based treatment bundles. In 2002, the Society of Critical Care Medicine and the European Society of Intensive Care Medicine implemented a new set of guidelines, the Surviving Sepsis Campaign, as an aid in the treatment of patients with severe sepsis and septic shock [
39]. Medical centres following these guidelines to a higher degree have been shown to have a lower mortality in septic patients [
40].
As a secondary aim of this study, we assessed the intervention effect on 28-day mortality. Overall, we found no efficacy regarding 28-day mortality. Four of the 44 included trials showed a beneficial effect of the studied intervention on mortality [
22‐
25]. However, two of these interventions, which used selenium and talactoferrin respectively, failed to demonstrate any benefit when replicated in larger studies with multicentre settings [
41,
42]. The other two treatment strategies with proven efficacy (esmolol and ulinastatin) [
24,
25] have not yet been implemented in the guidelines for standard care of sepsis, as their clinical effect needs further evaluation, although ulinastatin has been described as a promising novel treatment option [
43]. Our subgroup analysis of the interventions and their effect on mortality revealed a significant increase in relative risk in the group of interventions categorised as “modulation of coagulation and inflammation”. All these trials investigated the effect of activated recombinant protein C (Xigris®). This result supports the reported risks of this specific treatment which led to its discontinuation [
30].
As already established, sepsis is largely a heterogeneous syndrome. This diversity extends both to administrative aspects, leading to risks of classification errors, and to the pathobiology constituting the disease, creating a great variance in the presentation of the ailment based on properties of the pathogen and the host alike. The heterogeneity of patients included in RCTs investigating interventions for sepsis is probably an important reason for a large number of trials that have failed to prove efficacy. Although many post hoc analyses of large RCTs have been performed to control for this, including stratification to identify subgroups of patients who may benefit from the interventions that failed to prove efficacy, individualised treatments have not been implemented in clinical care [
44]. Precision medicine with treatments guided by specific geno- or phenotypes seems to be a promising future approach to find interventions with efficacy in certain subgroups [
45‐
47] but require molecular diagnostics with short turnaround to clinicians. Interestingly, four novel subtypes of sepsis based on routinely available clinical data, have recently been shown to predict for differences in treatment responses of previous sepsis trials [
48].
In order to address the difficulties caused by the heterogeneity of the disease, in 2016, the Society of Critical Care Medicine and the European Society of Intensive Care Medicine presented the third international consensus definition for sepsis and septic shock [
49]. This aims to minimise the risk of misclassification and ease the management of the condition by presenting a more restricted definition of sepsis. This restricted definition may however increase the false negatives, as shown by Fang et al. [
50]. It remains to be seen how this will affect heterogeneity in future studies, but hopefully this will ease the process of studying the disease, leading to a better understanding of the condition.
In recent decades, even though only a few specific interventions have shown some effect on mortality, progress has been made and embraced in the shaping of international guidelines. Several studies have raised a warning flag regarding excessive fluid therapy [
51,
52], one of which reported a correlation between the amount of administered fluids and death [
52], leading current guidelines to suggest a more restrictive approach [
16]. Moreover, in 2012, Perner et al. reported an increase in adverse events while treating patients with septic shock with hydroxyethyl starch (a synthetic colloid) [
53], with the result that it is no longer used in clinical management. Current guidelines advocate fluid resuscitation to be primarily constituted by crystalloids and, if needed, addition of albumin rather than any synthetic colloid [
16]. Furthermore, improvements in sepsis survival in ICUs may be related to the use of a protective-ventilation strategy in patients with sepsis-induced acute respiratory distress syndrome, which correlates with increased survival [
54].
Our study has several strengths. First, as this meta-analysis included over 13,000 patients enrolled in the control group in 44 trials from 55 countries and 6 continents, the reported results do not seem to be confined to a single part of the world. Moreover, all studies considered for inclusion were assessed for quality and excluded if they did not meet the pre-defined quality criteria. Lastly, as we used 28-day mortality ± 2 days as an inclusion criterion, the risk of missing important papers on the matter was reduced.
This study also has several important limitations. First, we only used one database to conduct our search and so may have left some studies out of our analysis. Second, the exclusion of trials published in languages other than English may have limited the findings of this meta-analysis to Western countries. Third, our choice to exclude trials assessed to be of poor quality may have further limited the number and diversity of the included trials. However, the limited number of trials excluded based on the quality assessment suggests that this is unlikely to have had any major impact on our results. Fourth, there might be differences related to the heterogeneity of the included sepsis trials as we included studies with severe sepsis and septic shock defined by either Sepsis-1 or Sepsis-2 criteria and from different parts of the world. Indeed, the degree and type of organ failure in sepsis are shown to be associated with the in-hospital mortality [
10]. Unfortunately, heterogeneity is a problem in sepsis and septic shock that seems to be difficult to adjust for. Availability of individual patient-level data can be helpful in this matter. In the meta-analysis by de Grooth and colleagues, a great heterogeneity was found in the control arms of sepsis RCTs, although they only included septic shock patients [
55]. This heterogeneity was only partly explained by differences in inclusion criteria and reported baseline characteristic. Interestingly, frequency, distributions and characteristics of the novel sepsis phenotypes identified by Seymour and colleagues were similar in studies with different definitions for sepsis [
48]. Due to inconsistency in severity assessment scales and lack of individual patient-level data, we could only adjust for illness severity in subgroup analyses. However, the APACHE II score, which was the most commonly reported score, is an independent risk factor for mortality that contains important components of organ dysfunction [
56]. The observation of inconsistency in severity assessment scales highlights the future importance of reporting disease severity by internationally adopted scores such as SOFA and SAPS III.
Finally, studies on trends in mortality from the usual care arm of sepsis trials may shed light on the history of sepsis research and partly mirror the trends in sepsis mortality, but are not suited for sepsis epidemiology surveillance due to the lack of clinical data and heterogeneity issues. The most reliable way to study trends in sepsis mortality seems to be based on clinical raw data derived from electronic healthcare data sets [
10]. In a recent paper by Rhee et al., an adjusted version of the SOFA score, the “eSOFA”, optimised for electronic health records, was shown to be of use as a tool in wide-scale surveillance of Sepsis-3 [
57]. Hopefully, the initiatives from Centers for Disease Control and Prevention prioritising sepsis as a global health priority will aid governments and healthcare providers to find similar solutions outside the USA in order to the measure the true effects of interventions aiming to reduce the global burden of sepsis.
Publisher’s Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.