Response patterns of routinely measured inflammatory and coagulatory parameters in sepsis

Inclusion of patients

Patients with the diagnosis of sepsis or systemic infection, who were treated at the medical, post-surgical and trauma ICUs between November 2000 and July 2013 with suspected or proven infection were respectively screened for study inclusion.

Patients meeting the sepsis criteria according to the American College of Chest Physicians/Society of Critical Care Medicine (ACCP/SCCM) international sepsis definitions based on a proven or strongly suspected infection with at least two of the SIRS criteria were included in the study (Bone et al., 1992; Levy et al., 2003b). No exclusion criteria were defined. The study was approved by the institutional review board of the Medical University of Innsbruck (AN2013-0043).

Data collection

Demographic variables such as age, sex and the diagnosed underlying disease were recorded. The C-reactive protein (CRP) level during the septic episode was used to harmonize the progression of sepsis between the patients. We chose peak of C-reactive protein since this parameter reflects the inflammatory process and is widely used in clinical routine for the surveillance of inflammatory diseases. Many studies have described an interrelation between an elevated C-reactive protein level and sepsis (Schentag et al., 1984; Maury, 1989; Povoa et al., 1998; Presterl et al., 1997).

Inflammatory and coagulatory routine parameters such as fibrinogen, platelets, antithrombin, prothrombin time (PT), activated partial thromboplastin time (aPTT), bilirubin and leukocytes were collected on the day of the peak C-reactive protein. In-hospital mortality was chosen as the main outcome parameter.

Statistical analysis

A mathematician not involved in the study procedures and patient assessment (TH) was responsible for the statistical analyses using R, version 3.4.1. All statistical assessments were two-sided and a significance level of 5% was used.

Patients with only one missing parameter on the day of C-reactive protein peak were included in the analyses. Simple mean imputation was deemed sufficient to deal with missing data, i.e., not inducing significant bias. C-reactive protein, fibrinogen, platelets, antithrombin, aPTT, PT, bilirubin and leukocytes measured at the peak level of C-reactive protein entered model selection according to the Bayesian Information Criterion (BIC) for expectation maximization (EM) initialized by hierarchical clustering for parameterized Gaussian mixture models using the R package mclust (Fraley et al., 2012). Gaussian mixture models are used to identify subpopulations within an overall population. Such models do not require knowledge of the subpopulation to which a patient belongs, thus permitting the model to automatically learn the subpopulations. For more details we refer to Fraley et al. (2012).

We descriptively present the resulting clusters by depicting the constituting variables via boxplots. Patient characteristics and outcome variables are presented as n (%) for binary data and continuous data as medians (25th to 75th percentile).

Differences between clusters are assessed with Pearson’s Chi-squared test for binary variables and the Kruskal-Wallis test for continuous variables; pairwise differences with the Wilcoxon rank sum test and Fisher’s exact test. We provide Kaplan–Meier curves for survival during 28 days after peak C-reactive protein stratified by the clusters and assess differences in survival using the log-rank test. A Cox proportional hazards regression model for survival stratified by the clusters was fitted and corresponding hazard ratios are provided, with 95% confidence intervals.

Patient population

In our patient population, men (65.5%) were more prevalent than women. Average age was 65 years and the majority of patients suffered from cardiovascular disease followed by complications of the digestive tract, as for example bowel perforation. The patients were severely ill and at ICU admission had a SAPS 3 score of 72 points and a SOFA score of 10 points. Of the patients 81.5% needed catecholamine therapy and 97.6% required mechanical ventilation.

Cluster analysis

The resulting five clusters (C1-C5) consisted of 336, 364, 105, 148 and 78 patients, respectively. Those clusters constituted profiles of patients presenting a similar pattern of routinely measured parameters at the peak level of C-reactive protein in sepsis (Fig. 2). All parameters differ significantly between the clusters (p < 0.0001).

The distinct patterns of parameters in the various clusters also reflected the type and severity of already existing diseases as SAPS 3 score increases from Cluster 1 to Cluster 5 (see Table 1). Correspondingly, 28-day mortality rose towards Cluster 5 (see Fig. 3) and differed significantly between clusters (p < 0.0001). Except for C1 and C2 as well as C2 and C3, which did not significantly differ from each other (p = 0.2576 and p = 0.2329, respectively), all other pairwise differences in 28-day survival were significant (see Table 2).

Patients in Cluster 1 were characterized by a moderate increase in inflammation and coagulation parameters in comparison to the other clusters describing a milder course of sepsis. The central nervous system was the main underlying disease, while the rate of other comorbidities was low as was the rate of mortality.

Typical for Cluster 2 were decreased levels of antithrombin as well as a relatively low platelet count. A further characteristic of this group was the high rate of renal disease and cardiovascular problems at ICU admission. This group suffered from an increased bleeding rate.

Cluster 3 showed a highly inflammatory course of sepsis with extremely elevated fibrinogen levels and a distinct probable reactive thrombocytosis. These patients frequently showed complications of the intestinal tract. Patients in C3 also exhibited a higher thrombosis rate although the difference was not statistically relevant.

Clusters 4 and 5 exhibited a state of decompensated coagulation and concomitant hyperbilirubinemia. Organ failure and the severity of dysfunction increased, resulting in elevated SOFA scores at the peak level of C-reactive protein. Liver failure frequently occurred in both clusters and was most pronounced in Cluster 5. 53.8% of the patients with a hepatic underlying disease suffered from pre-existing liver cirrhosis and of these patients, 71.4% developed acute-on-cirrhosis liver failure during the septic course (see File S1).

Sepsis severity was characterized by an increasing number of patients with multiple-organ failure towards Cluster 5. The disordered coagulation system resulted in an increased bleeding incidence in Clusters 2, 4 and 5. Especially in Cluster 5 80% of the patients with hepatic failure suffered from impaired liver synthesis and in most cases required vitamin K substitution. The thromboembolic events were most severe in the high inflammatory Cluster 3 and in Cluster 5, although the difference was not statistically significant. In Cluster 5 10.3% of the patients developed disseminated intravascular coagulation (DIC). The length of ICU stay did not significantly differ between clusters (p = 0.4085, Kruskal-Wallis rank sum test) and in median (IQR) lasted 14 (7–29) days.

Timeline analysis

Since the patient data were collected over a protracted period from 2000 to 2013, the effect of time was investigated. The mortality rate was significantly higher in patients treated before 2005 than in patients treated later.

All types of clusters were present during the entire study duration, but there was a shift in patient numbers in the individual parameter patterns (see Fig. 4). The number of patients with the highest survival rate (shown in green in Fig. 4) increased over the years. Even when looking at the survival rate in this cluster, the 28-day survival rate was seen to have increased.

After a peak in the period 2004–2007 the number of patients in the high inflammatory response group (shown in gold in Fig. 4) decreased. The number of patients with mild liver failure declined significantly over the entire period, while the number in the group with severe liver failure increased.

Coagulatory and inflammatory parameters associated with survival

A Cox proportional hazards regression model for survival stratified by clusters was fitted (Fig. 5) to evaluate the measured parameters with regard to their predictive value regarding the outcome. An increase in fibrinogen levels (100 mg/dl), number of platelets (50 G/l) and PT (10%) as well as a decrease in leukocyte count (5 G/l) was associated with improved 28-day survival.

Cluster analysis

The cluster analyses resulted in five different response patterns that interestingly could be attributed to various underlying diseases.

Cluster 1 grouped patients with a low rate of comorbidities and a high probability of survival (Bassetti et al., 2017; Banta et al., 2012). Their response pattern was characterized by a mild inflammatory response and sufficient coagulatory capacity.

Patients in Cluster 2 frequently exhibited renal complications already at ICU admission, which was probably leading to protein loss or the need for hemodialysis with increased consumption of antithrombin (Lanquetot et al., 2008) and platelets (Stefanidis et al., 1996; Mulder et al., 2003).

Patients in Cluster 3, who frequently showed complications of the digestive tract, exhibited considerable hyper-inflammation and pronounced thrombocytosis. Thrombocytosis indicates the presence of a severe underlying disease (Tchebiner et al., 2011) and is associated with soft tissue infection, severe bacterial infection (Tchebiner et al., 2011; Fouzas et al., 2010) and a prothrombotic tendency (Rottenstreich et al., 2018; Ho, Yip & Duff, 2012). This could be true in our Cluster 3 patients since this group had the highest number of underlying digestive complications associated with a higher thrombosis rate although the difference was not statistically relevant.

Coagulation parameters were lowest in the patients in Clusters 4 and 5, probably as a sign of decompensation and multiple-organ failure, especially failure of the liver with subsequently elevated SOFA scores, resulting in impaired liver synthesis in 80% of all patients with liver failure. About half (53.8%) of the patients suffered from pre-existing liver cirrhosis and 71.4% experienced “acute-on-cirrhosis liver failure.” Cirrhosis seems to be a risk factor for developing impaired liver synthesis during sepsis, thus resulting in a higher rate of bleeding complications. Elevated bilirubin levels as surrogate marker for hepatic failure have already been shown to be associated with high mortality (Dizier et al., 2015). The extensive high share of developed cardiac organ failure (76.9%) and renal failure (78%) in Cluster 5 also indicates decompensation.

Hepatic failure leads to impaired synthesis performance resulting in desolate coagulation and declining levels of acute phase proteins in our patients. Prolonged aPTT and decreased PT indicate a hypocoagulant stage; especially low PT values seem to reflect the impaired synthetic capacity of the liver (Agarwal et al., 2012). In acute liver failure decreased levels of most coagulation factors are known (Agarwal et al., 2012). The antithrombin level decreases more quickly than other coagulation factors (Tischendorf et al., 2016), thus giving the illusion of a “rebalanced hemostasis” (Habib et al., 2014; Lisman & Porte, 2010). However, our septic patients with liver dysfunction showed a significantly increased bleeding rate as well as a trend towards a larger number of thromboembolic complications. These two conflicting complications seem to affect especially patients in Cluster 5, as they have impaired liver synthesis and are developing DIC (10%). Extremely increased monitoring of coagulation is essential for survival in this patient population.

The patients in these clusters are also characterized by lower numbers of platelets and leukocytes. This may be due to septic-induced immunosuppression (Schmoeckel et al., 2018; Venet & Monneret, 2018), in which patients are unable to make up for the increased consumption of platelets and leukocytes.

Coagulatory and inflammatory parameters associated with survival

Despite the various clinical characteristics and constellation of inflammatory and coagulatory parameters, regardless of the cluster, four significant parameters remained as independent predictors for survival in our patient population, namely fibrinogen, platelets, PT and leukocytes.

An increase in fibrinogen levels per 100 mg/dl was associated with an 11% improvement in survival. The same observation was also made in a previous study (Ersoy et al., 2007). Moreover, sepsis animal models have confirmed that high fibrinogen levels are associated with better outcome (Journal of Cellular and Molecular Medicine Jr, Kinasewitz & Lupu, 2012; Fiusa et al., 2015; Luo et al., 2011; Sun et al., 2009; Mullarky et al., 2005). In our septic patient population low fibrinogen levels were associated with poor outcome, even if within the normal range.

Fibrinogen is not only responsible for the fibrin network during coagulation, but also plays a role in inflammation and the innate immune system (Davalos & Akassoglou, 2012). Both formation of the fibrin network and successive fibrinolysis exhibit an antimicrobial function by first capturing bacteria with subsequent lysis and so preventing further dissemination (Pahlman et al., 2013; Herwald, 2013; Loof et al., 2011; Wang et al., 2010). This can predominantly be observed in bacteria exhibiting fibrinogen binding sites, which is most common in gram-positive bacteria, such as Staphylococcus aureus, and also in some gram-negative bacteria, such as Moraxella catarrhalis (Pahlman et al., 2013) or Yersinia enterocolitica (Luo et al., 2011).

On the other hand, low fibrinogen levels may be due to consumption via fibrin deposition and subsequent disseminated intravascular coagulation (DIC) as well as multiple-organ dysfunction including liver damage (Semeraro et al., 2015; Parker, 2013). During liver failure the synthesis of fibrinogen is impaired (Nesseler et al., 2012) although in our patients with liver failure the fibrinogen levels were still mainly within or slightly above the normal range, but associated with poor outcome.

Thrombocytopenia during sepsis was associated with higher mortality, whereas the survival rate increased by 14% per 50 G/L of thrombocytes. A previous study revealed that the independent risk factors for thrombocytopenia during septic shock might be several comorbidities such as immunosuppression and liver cirrhosis, but are also assessed with severity of disease and site of infection (Thiery-Antier et al., 2016). Platelets are not only consumed by coagulation, but also by their immunological actions (Gros et al., 2015; Kim & Jenne, 2016; Speth et al., 2013).

Another parameter associated with a 15% improvement in survival when rising was the prothrombin time (PT) of 10%, namely independently of the cluster to which the patient was assigned. In some infectious diseases prothrombin time seems to be an important predictor for severe complications (Chang et al., 2019) and mortality in septic shock (Ishizuka, Tago & Kubota, 2014). Although the leukocyte count was described as a very weak prognostic marker for the survival of sepsis (Suberviola et al., 2012), we observed better survival in patients with fewer leukocytes.

Overall, coagulatory parameters appear to be well-suited for stratification of sepsis patients, particularly fibrinogen and platelets. Among other things, when both are elevated, they have a diagnostic value and may indicate a multi-pathogenic infection. When patients suffer from liver dysfunction, this is well reflected in prolonged aPTT, reduced PT and elevated bilirubin levels. In sepsis, a high fibrinogen level is associated with better survival while thrombocytopenia, leukocytosis, and low PT levels are associated with increased mortality. However, C-reactive protein and antithrombin appear to be unspecific for the stratification of septic patients.

Timeline analysis

Patient data were collected over a protracted period from 2000 to 2013, and an effect of time on mortality is given. The mortality rate was significantly higher in patients treated before 2005 than in patients treated later. The decline in mortality rates among patients with sepsis was also observed in other studies (Martin et al., 2003; Angus et al., 2001; Kumar et al., 2011). In our study, all types of clusters were present during the entire study duration, but patient numbers shifted in the various parameter patterns.

The number of patients with the highest survival rate increased over the years, reflecting advanced therapy strategies. Interestingly, the number of patients with mild liver failure significantly decreased over the entire period, while the number in the group with severe liver failure increased. This is in accordance with other studies that showed that although the overall mortality from sepsis decreased, there was also a shift towards increased severity of sepsis and the number of organ failures (Kumar et al., 2011; Vincent et al., 2018).

This development of an increased survival rate is probably attributed to early and improved treatment of sepsis in order to not let sepsis become severe or the patients are able to survive until an extremely severe stage of disease when those patients are unresponsive to further therapy. This is supported by the fact that the cluster with severe liver failures shows the most distinct shift towards increasing mortality.