Introduction
Sepsis is a life-threatening syndrome of organ dysfunction caused by a dysregulated host response to infection that affects 50 million patients worldwide annually [
1]. Despite efforts to improve recognition and treatment, sepsis remains an important cause of morbidity and mortality among patients admitted to the hospital with mortality rates ranging from 20 to 36% [
2‐
7]. Timely initiation of adequate treatment is essential to prevent clinical deterioration. Therefore, the key to reducing sepsis-related mortality could lie in improving early recognition and identification of those at risk of poor outcomes [
7,
8]. Currently, the most commonly used clinical assessment tools to facilitate early recognition of sepsis are the Systemic Inflammatory Response Syndrome (SIRS, Sepsis-2) criteria, (quick) Sequential Organ Failure Assessment ((q)SOFA, Sepsis-3), and the National Early Warning Score (NEWS) [
1]. Of these scores, the NEWS and SOFA appears to have the highest accuracy to predict sepsis-related mortality [
9‐
11]. Although the NEWS was not specifically developed for early sepsis recognition, it can accurately identify patients with more severe disease and thereby facilitate early recognition of sepsis [
9,
10].
Most research validating qSOFA and NEWS are performed among general emergency department (ED) and ICU patients—not explicitly assessing the scores in immunocompromised patients. Yet, immunocompromised patients with an infection are at increased risk for sepsis development. Early recognition of sepsis and risk prediction in immunocompromised patients might be more difficult due to their altered immune response. Data describing the prognostic accuracy of the qSOFA and NEWS in immunocompromised patients are scarce and are either contradicting or cannot be generalized to pharmacologically immunocompromised patients. Immunodeficiency due to co-morbidity (i.e., cancer, neutropenia, AIDS) is a risk for 28-day mortality among patients with sepsis admitted to the ICU [
2]. Unexpectedly, patients with bacteremia after solid organ transplantation have a lower 28-day and 90-day mortality as compared to age-, sex-, and hospital-matched bacteremia patients without history of organ transplantation [
12]. The performance of SIRS, qSOFA, and NEWS has been assessed in patients with a suspected infection after hematopoietic cell transplantation. In this population, the NEWS had a moderate sensitivity (78%) and specificity (70%), which outperformed qSOFA and SIRS [
13]. Together, immunosuppression due to co-morbidity affects outcome in patients with sepsis, with relevance to the predictive performance of commonly used sepsis scores.
The effect of immunosuppressive medication on clinical outcomes in patients with sepsis, and thereby, the predictive performance of sepsis scores remains unknown. We hypothesized that dampening the immune response by immunosuppressive drugs affects the prognostic accuracy of NEWS and qSOFA, which is likely to depend on drug class. For this study we used the NEWS in addition to the qSOFA, as advised by the Dutch sepsis guideline to support sepsis recognition at the ED [
14]. To this end, we compared the prognostic performance of NEWS and qSOFA to predict the in-hospital mortality in pharmacologically immunocompromised patients with sepsis at the ED, adjusted to the immunosuppressive drug class.
Discussion
Here, we evaluated the effect of immunosuppressive drugs on the performance of commonly used sepsis scores to predict clinically relevant outcomes of sepsis. Of all classes of immunosuppressives studied, only the use of high-dose glucocorticoids (> 15 mg/day) was associated with a profoundly increased risk of in-hospital mortality, which remained significant in multivariate analysis. When comparing the performance of qSOFA and NEWS in predicting in-hospital mortality and attempting to adjust for high-dose glucocorticoid use, we found that NEWS is better in predicting in-hospital mortality than the qSOFA among immunocompromised patients.
A common concern in patients on immunosuppressive agents is a blunted immune response leading to a lack of clinical symptoms that usually accompany sepsis [
13]. Subsequently, early sepsis recognition can be more challenging in immunocompromised patients. In our cohort we found a sensitivity around 25% and specificity ranging from 92.4 to 96.1% for qSOFA to predict in-hospital mortality. The highest specificity of the qSOFA was found in patients on high-dose glucocorticoids. The sensitivity of qSOFA was slightly lower in our immunocompromised groups, being 25.0% versus 25.7% in immunocompetent patients. The qSOFA is known to suffer from a low sensitivity for predicting both in-hospital and 28-day mortality, ranging from 15 to 48% in other studies, while the reported specificity is around 90% and higher [
13,
19]. The qSOFA has a sensitivity of 47.8% and specificity of 90.5% among sepsis patients who are immunocompromised secondary to hematopoietic cell transplantation [
9,
13,
19‐
21]. The sensitivity we found is within the wide range as reported by others. However, it appears lower as compared to the sensitivity of the qSOFA among hematopoietic cell transplantation receivers—on the other hand, the specificity of the qSOFA in our cohort of pharmacologically immunocompromised groups was slightly higher. Although qSOFA is considered more user-friendly, since it requires only three, readily available parameters, clinicians should be aware that identification of immunocompromised patients at high risk of poor outcome is hampered.
In our cohort, NEWS outperformed the qSOFA in predicting in-hospital mortality, both among the whole cohort as well as the studied subgroups. The AUROC to predict in-hospital mortality of the NEWS score was higher than the qSOFA in the whole cohort, as well as among immunocompromised patients and among those using high-dose glucocorticoids. The sensitivity of the NEWS ranged from 65.4% in the whole cohort to 80% in immunocompromised patients and 100% among high-dose glucocorticoid users. The specificity ranged from 73.0 to 74.0%. These findings are in line with previously reported sensitivity of 74.0–87.9% and specificity of 42.1%–90.2% to predict sepsis-related mortality in an immunocompetent population has been reported [
9,
10,
20]. When comparing our findings among pharmacologically immunocompromised patients to the accuracy of NEWS among other groups of immunocompromised patients due to hematopoietic cell transplantation, it appears that the sensitivity is higher (hematopoietic cell transplantation: 64.9%), while specificity is similar (hematopoietic cell transplantation: 71.2%) [
13]. Further, in kidney transplant recipients the NEWS and SOFA performed best in recipients compared to immunocompetent patients, compared to CRB-65, CURB-65, DS-CRB-65, qSOFA, and PSI (Pneumonia Severity Index) score [
22]. In our study, the predictive performance of NEWS was higher than qSOFA in the whole cohort, as well as in immunocompromised patients in general and specifically in patients on high-dose glucocorticoids, in line with observations among patients with a suspected infection after hematopoietic cell transplantation [
13].
The 28-day mortality rates in our study were 6.2% for the total cohort and 5.9% among patients who were pharmacologically immunocompromised. It should be noted that the studied population represents a group of patients with a severe infection at risk of developing sepsis and those who already meet sepsis criteria: a clinically relevant group of patients seen daily at the ED. More than half of the patients met Sepsis-3 criteria, while 5.1% was admitted to the ICU and 3.6% died in hospital. The ICU admission rate appears to be lower than in other studies, which could be explained by the relatively low number of patients with sepsis. Another explanation might be the support for more complex care at the general wards by ICU outreach teams in the Netherlands, which lowers the requirement for ICU admission. For the different groups of immunosuppressives, 28-day mortality rates ranged from 0 to 12%, with the highest mortality rate observed in patients on high-dose (> 15 mg/day) glucocorticoids. Although infections are a well-known risk associated with use of immunosuppressive drugs, the risk of deterioration and sepsis development varies between different classes of immunosuppressive drugs due to their different target cell(s) [
17,
23‐
26]. While some drugs have very specific targets, glucocorticoids have broad effects on the immune system by their anti-inflammatory properties [
27]. Previous studies among immunocompromised patients with an infection demonstrated similar or even better outcomes as compared to immunocompetent patients [
5,
12,
28‐
30], with 28-day mortality rates from 8 to 32% [
5,
12,
31]. Only a few studies took the specific immunosuppressive drug class and/or dose administered into account, rather than considering a general immunocompromised state when using any immunosuppressive drug or having specific co-morbidity (e.g., cancer, history of solid organ transplantation). The relevance of investigating specific immunosuppressive drug classes is supported by our findings, demonstrating an increased in-hospital and 28-day mortality risk among high-dose glucocorticoid users, while use of IMPDH inhibitors was associated with a lower 28-day mortality risk. In a study among rheumatic patients on either conventional/biological disease-modifying anti-rheumatic drugs (cs/b DMARDs) or glucocorticoids who developed sepsis, high-dose glucocorticoids were also associated with mortality, while bDMARD users had a lower risk on mortality [
32]. In contrast, however, a study using health care insurance data demonstrated lower in-hospital mortality from sepsis among glucocorticoid users (not stratified to the dose) and among patients taking non-steroid immunosuppressive drugs. Yet, patients taking immunosuppressive drugs were more frequently admitted to the ICU because of sepsis and had a longer length of stay in hospital [
29]. Unfortunately, the drug class was not further specified among the non-steroid immunosuppressive drug users.
Strengths and limitations
There were several limitations to our study. First, the diagnosis of sepsis was based upon the combination of a (suspected) infection and a qSOFA/SOFA score of ≥ two according to information present in the electronic patient chart. Patients in our dataset could well be using ≥ one immunosuppressive drug at the same time and interactions and effects of concomitantly used agents cannot be ruled out. We stratified patients into groups based on the immunosuppressive drug use to identify specific drugs that would affect the predictive performance of the qSOFA and NEWS. Consequently, a number of patients on TNF blocker and selective immunosuppressives are small, potentially underpowered, and therefore at risk for type II error. Although we do not have evidence to support that the inclusion time from 8:00 to 23:00 h may have caused selection bias, the risk of such bias should be taken into account when interpreting the evidence provided in the study. It should be noted that there is an inherent risk of overfitting the model that we developed in this single-center cohort. Although the results do allow to expand understanding of the potential risks of immunosuppressives among patients with an infection at the ED, the model should not be used to screen patients without external validation.