Biomarkers and clinical scoring systems help physicians to detect sepsis in an early stage in the ED. In this systematic review we investigated the combinations of biomarkers and clinical scoring systems to predict 1-month mortality in patients with sepsis. We found 18 different studies in which 33 combinations of biomarkers and clinical scoring system were investigated. The combination of PCT, lactate, IL-6 and SAPS-2 resulted in the highest AUC on 1-month mortality.13 Despite the high AUC found in this study, this specific combination has not been adopted in the latest guidelines for surviving sepsis.2 The SAPS-2 score is a clinical scoring system, using four vital parameters, seven laboratory tests and four other patient characteristics and was originally developed for patients in the ICU or general wards to predict in-hospital mortality. Combining this clinical scoring system with another three biomarkers results in a total of 18 variables used to predict 30-day mortality in this study. This study enrolled 131 subjects, of which 19 died. Therefore, the high AUC found in this study may possibly be due to overfitting by using too many predictors in the multivariate logistic regression analysis.32, 33
Lactate is a product of anaerobic glycolysis and is often elevated in patients with sepsis. It has been adopted as criterion for septic shock in the Sepsis-3 definitions.7 We found two studies using lactate in combination with other biomarkers or clinical scoring system, both with a high predictive value on 1-month mortality.13, 17 Unlike many novel biomarkers, lactate is widely available as a standard measurement during the workup in the ED. Therefore, lactate is an important biomarker in assessing the severity of sepsis in the ED. IL-6 is an inflammatory cytokine and plays an important role in the early phase of sepsis.34 However, the prognostic values of IL-6 are controversial due to the short window in which IL-6 rises and falls during inflammation and infection.35
The SAPS-2 score was also used in combination with suPAR by Kofoed et al.14, resulting in an AUC of 0.930, which was the second highest AUC we found in our study. These findings suggest that the SAPS-2 is a clinical scoring system with a high diagnostic accuracy on 30-day mortality, although it has not been validated for assessing severity of disease in the ED. However, the limitation of an overfitted prediction model in a relatively small cohort is also present in the study of Kofoed et al., with only 161 patients enrolled of which 9 patients died. suPAR is a biomarker which has been investigated as general disease severity biomarker, mostly in the ED. A large study showed that suPAR is an accurate predictor of mortality, but does not influence disposition or clinical outcome when it was used in the ED.36 In a meta-analysis, suPAR showed similar results as PCT in diagnosing sepsis.37
We found four studies using PCT combined with another biomarker or clinical scoring system. PCT has been studied as biomarker for bacterial infections and disease severity in infectious diseases. PCT is the precursor of calcitonin and physiologically produced by thyroid cells. In bacterial infections it is also synthesized outside of the thyroid, and rises rapidly in systemic infections. It is often referred to as the biomarker with most potential of replacing or substituting CRP.38 However, PCT has yet to establish a role in routine care in the ED.39 Combining PCT with other biomarkers or clinical scoring systems, we found an increase in predictive value on 30-day mortality. From all available sepsis biomarkers, PCT is probably the most well-known among physicians in the ED. Combining PCT with clinical scoring systems, as done in many studies, might therefore be the key in being adopted as part of regular care.
The MEDS score was the most used clinical scoring system used in combination with biomarkers. The MEDS score is a risk prediction score specifically for patients with suspected sepsis in the ED.40 It consists of nine items which can easily be scored in the ED setting and results in a total score, categorized in 5 groups, which corresponds to a certain mortality risk. The AUC of the MEDS score combined with different biomarkers ranged from 0.731 to 0.891, indicating a moderate to good predictive value on 30-day mortality. Other clinical scoring systems we found in combination with biomarkers were the APACHE-2 and SAPS-2 score. These clinical scoring systems are mainly developed for use in the ICU and general wards. Despite being accurate predictors of disease severity, these clinical scoring systems may be less feasible for use in the ED, due to their complexity and large number of clinical parameters needed. In a prospective study comparing different clinical scoring systems individually in the ED, the MEDS score resulted in an AUC of 0.94 on 30-day mortality, which was higher than the SOFA or PIRO score.41 However, another study which also compares different clinical scoring systems concluded that the APACHE-2 score is superior to the MEDS and SOFA score.42
Three studies investigated biomarkers which are otherwise known as hormones and other functional circulating peptides, including IgE, cortisol, cell free DNA and nucleosomes.15, 22, 24 Zhang et al.22 studied IgE in combination with the MEDS score and found that adding IgE to the MEDS score resulted in a higher AUC than the MEDS score alone. This study emphasizes the multifactorial entity of sepsis, hypothesizing that IgE either plays a role in general immune activation during sepsis or is a marker of cytokine regulation/dysregulation. Another study by Zhang et al.24 investigated hormones and biomarkers from the hypothalamic–pituitary–adrenal axis and showed that cortisol and copeptin are associated with 30-day mortality and that combining these biomarkers with the MEDS score resulted in added value over using each biomarker individually. Cortisol has been identified as essential hormone in the immune response in sepsis and elevated levels of cortisol are associated with severity of sepsis.43 Extracellular cell free DNA and nucleosomes, basic units of DNA packaging, reflect cellular apoptosis and are therefore tested as predictors of severity of sepsis in the study of Duplessis et al.15 In this study the authors show that adding nucleosomes to the APACHE-2 score improved the AUC on predicting mortality. Adding cell free DNA to the APACHE-2 score did not result in a better predictive value. These studies emphasize that biomarkers originating from different pathways in sepsis can be used as predictor of disease severity.
Limitations
Our study has several limitations. We included articles that investigated the predictive value of biomarkers combined with clinical scoring systems on disease severity in sepsis. We used 1-month mortality as endpoint for severity of sepsis. However, there are many more biomarkers that have been investigated using other endpoints as marker of disease severity of sepsis. These endpoints to define severity of sepsis range from ICU admission to long term mortality. Despite the fact that these endpoints also are a surrogate marker of disease severity, these articles were not included because a comparison of these endpoints would not be possible. An example of a commonly used scoring system not included in this study is the National Early Warning System (NEWS). The NEWS score is widely adopted as early warning score to identify patients at risk of clinical deterioration both in the ED and the hospital wards, where it also has been shown to be superior to the qSOFA score.44 Adding biomarkers to the NEWS score could improve its predictive value even further. However, no studies on our specific endpoint with the NEWS score and biomarkers were found.
The definition of sepsis has changed over time, which is also reflected by the different inclusion criteria used by the studies we found. Most studies used the sepsis criteria as defined in 200121, and only a few studies used the latest Sepsis-3 criteria.7 Other studies used two or more SIRS criteria in combination with an infection, but there were also studies that included patients with only blood cultures taken. This might have resulted in a heterogeneous groups of patient populations in which the severity of sepsis in the ED differs at the moment of inclusion. Some studies included patients with already diagnosed sepsis, while other studies also included patients at risk or suspected of sepsis.
All but three studies did not use predefined cut-off values for the biomarkers. It is common practice not to do so as an AUC is usually constructed using biomarkers and scoring systems on a continuous scale. However, for translating these prediction models to clinical practice, using a predefined cutoff, categorizing the biomarker or clinical scoring system in a high or low risk category, is preferred.
The study population size of the included studies ranged from 114 to 1318. In these relatively small sample sizes, overfitting of prediction models often is a problem. When there are less than 10 fatal cases per predictor, the risk of overfitting of prediction models is high, resulting in an unrealistically high AUC.33 When also including clinical scoring systems, this problem is even bigger, since the clinical scoring system already consists of multiple predictors.
Conducting a meta-analysis to compare the outcomes of different studies would be preferred, but was not feasible. The variety in biomarkers and clinical scoring systems used was too large to compare one to another.