Introduction

Sepsis has become a major health issue worldwide owing to its high mortality rate [1]. It occurs when an inadequate host response to infection causes life-threatening organ dysfunction [2]. Conquering sepsis was ranked as one of the highest-priority tasks by the World Health Organization (WHO) during the 70th World Health Assembly in 2017, and governments worldwide have been urged to invest greater efforts in achieving this goal [3]. However, it remains a challenge for clinicians to identify high-risk patients with sepsis. Early mortality prediction is of great importance for timely and intensive management, which is a high priority for improving the outcomes of sepsis [4,5,6]. Traditional prediction models rely mostly on scoring systems that require many variables that are difficult to obtain [7, 8]. Otherwise, the majority of existing models have shortcomings, including tedious processes and poor operability [9]. There is still a lack of effective, simple, and convenient models for predicting the prognosis of sepsis [10]. However, several biomarkers have been shown to play an important role in predicting patient outcomes, we develop a 28-day sepsis mortality prediction model, according to the transparent reporting of a multivariable prediction model for individual prognosis or diagnosis (TRIPOD) [11].

Cytokine storms have been one of the hottest research topics in sepsis [12]. Among these, major pro-inflammatory cytokines, interleukin-6 (IL-6) reflects sepsis at the acute stage [13]. Several studies have assessed the diagnostic and prognostic value of IL-6 in patients [13,14,15,16]. However, its prognostic value for predicting sepsis outcomes remains controversial. Some studies concluded that IL-6 levels had a high value in predicting 28-day sepsis mortality [17, 18]; on the other hand, IL-6 was not associated with survival in sepsis [15]. In this study, IL-6 levels were measured consecutively for 72 h from admission, and IL-6 clearance was calculated on D3 (48–72 h) to assess the prognostic value of IL-6 in sepsis.

Lactate (LAC) is an important biomarker of cellular metabolism and energy production [19]. Elevated LAC levels indicate tissue hypoxia. It has been demonstrated as a valuable prognostic marker for hypoperfusion, especially in sepsis, and as a predictor of 28-day sepsis mortality [16]. Some studies have reported that LAC levels and clearance are both predictors of mortality in sepsis [20,21,22]. Nevertheless, studies on the prognostic value of LAC levels and clearance for sepsis have been limited since the release of the Sepsis-3 criteria [20].

PCT is a protein produced by the thyroid gland that is undetectable in healthy individuals [23]. Since PCT was first described as an infection marker in 1993, it has been reported as a potential marker for assessing the presence of infection and clearance of inflammation and has been used as a guide for antibiotic therapy and prediction of mortality [24]. In the past two decades, it has become the most popular and widely studied sepsis biomarker [14]. Moreover, some studies have suggested that PCT clearance may serve as an outcome predictor in sepsis [25, 26]. However, the performance characteristics of PCT are still insufficient to make it the gold standard for sepsis survival [24]. Thus, our study aimed to evaluate the combined predictive performance of PCT (including PCT levels and clearance) as well as of IL-6 and LAC.

The present study was established to preliminarily develop a novel simple prediction model for the prognosis of 28-day mortality of sepsis by combining the predictive performance of the levels and clearance of IL-6, LAC, and PCT.

Materials and methods

Study design and setting

We conducted this retrospective study at the Emergency Department of Shenzhen People’s Hospital. There were 3105 beds in the hospital and 120 beds in the emergency department (ER). An average of approximately 200 patients visit the ER daily. A total of 367 patients diagnosed with Sepsis-3 were included in our study from September 2019 to December 2021. This study was approved by the Ethics Committee of Shenzhen People’s Hospital (no. KY-LL–2,020,157–02). Informed consent was secured from all the participants.

Study population

We consecutively enrolled adult patients examined and assessed by a physician to have sepsis according to the Sepsis-3 guidelines at the ER. The following were excluded: (1) patients < 18 years of age, (2) patients resuscitated from cardiopulmonary arrest, (3) those with incomplete data requested during the first 48 h of hospitalization (i.e., one measurement), (4) patients on do-not-resuscitate status, (5) patients hospitalized for < 24 h, (6) patients with cancer, and (7) patients who underwent major surgeries in the previous 30 days. Patients with cancer were excluded, since tumors are proposed to use lactate as a fuel, which expands the metabolic functions in cancer [27]. Furthermore, PCT levels will rise in cancer patients. On the other hand, those who underwent major surgeries were excluded because studies have reported elevated IL-6 levels after surgery and returned to baseline levels 2 weeks post-operation [28].

Data collection

The following clinical data were obtained: age, sex, comorbidities, mechanical ventilation (MV) settings, mean arterial blood pressure (MAP), vasopressor use, type and source of infection, and days of hospital and ICU stay. The outcome was all-cause mortality at 28 days. The worst SOFA score was seen in each patient within 24 h of admission. We measured IL-6 and LAC levels using the Cobas 6000 analyzer (Roche Diagnostics System, Rotkreuz, Switzerland) and PCT levels using the VIDAS immunoassay system (bioMérieux, Marcy L’Etoile, France). All tests were completed during hospitalization in the emergency ward. Results collected within 24 and 48–72 h after admission were marked as D1 and D3, respectively (IL-6D1/D3, LACD1/D3, and PCTD1/D3). All data were retrospectively collected without any intervention. The IL-6, LAC, and PCT clearance values were calculated using the following formula (taking PCT as an example) [20, 29]:

$$\mathrm{PCTc}\;\mathrm D3=\frac{\mathrm{initial}\;\mathrm{PCT}-\mathrm{PCT}\;\mathrm D3}{\mathrm{initial}\;\mathrm{PCT}}$$

Statistical evaluation

If continuous variables had a normal distribution, they were presented as means ± standard deviations or as medians with interquartile ranges (IQRs). The chi-squared test or Fisher’s exact test was used to compare categorical variables, and the independent two-sample test or Mann–Whitney U test was used to compare continuous variables. Missing data were filled with multiple imputations using the MICE package in the R software (version 3.4.3). Then, the data were split into a training cohort and a validation cohort at a ratio of 7:3. Single factor linear regression was used to screen independent variables related to dependent variables (P < 0.05); multiple stepwise regression (forward–backward method, both) was performed on the selected independent variables to screen independent influencing factors related to dependent variables, and the AIC was used to determine the optimal model. ROC curves, calibration plots, and DCA were used to test the performance of the model. A nomogram was used to validate the model.

Results

Patient demographics

A total of 462 patients with sepsis were admitted to the ER. We enrolled 367 patients in this study (Fig. 1). After 28 days, there were 314 survivors and 53 mortalities, causing a mortality rate of 14.44%. Table 1 shows the patient demographics. The median (range) patient age was 73 (19–98) years, and 65.9% were male. The most common comorbidity was hypertension (47.4%). The causes of sepsis were pneumonia (68.9%), urinary tract infection (31.1%), intra-abdominal infection (22.6%), and hepatobiliary infection (14.4%). Microbial culture would be conducted twice for each site of the primary infection. Furthermore, 34.9% of the patients had two causes of infection; 9.0% patients had three causes of infection;, and 1.4% had four causes of infection. The pathogens of the infections were gram-positive bacteria (25.3%), gram-negative bacteria (8.2%), and fungi (12.5%).

Fig. 1
figure 1

Flow chart of the study population

Full size image
Table 1 Demographic characteristics
Full size table

Comparison of IL-6, LAC, and PCT levels and clearance between survivor and non-survivor groups

Table 2 provides a comparison of the levels and clearance of IL-6, LAC, and PCT between survivors and non-survivors. Aside from the IL-6cD3 and PCTD3 levels, all other variables were significantly different between the two groups.

Table 2 Biomarkers in blood of the study groups
Full size table

Comparison of IL-6, LAC, and PCT levels, and clearance between the training and validation cohorts

Table 3 shows a comparison of the variables between the training and validation cohorts. There were no significant differences for all variables between the two groups.

Table 3 Biomarkers in blood of the training and validation cohort
Full size table

Logistic regression analysis for the training cohort

Table 4 shows the results of the logistic regression analysis. IL-6D3 (OR = 1.007; 95% CI: 1.003–1.010; P = 0.000), LACcD3 (OR = 0.993; 95% CI: 0.986–1.000; P = 0.042) and PCTD1 (OR = 0.957; 95% CI: 0.933–0.982; P = 0.001) are independent risk factors for 28-day sepsis mortality. Based on the AIC (164.335), PCTD1, PCTD3, IL-6D1, IL-6D3, and LACcD3 were used to develop the model.

Table 4 The logistic regression analysis of IL-6, LAC and PCT levels, and clearance for predicting 28-day mortality in 257 patients with sepsis or septic shock for training cohort
Full size table

Performance of the prediction model shown by ROC, calibration, and DCA curves

Fig. 2A, B, and C show the ROC, calibration, and DCA curves of the training cohort, while Fig. 2D, E, and F are those of the validation cohort. The ROC results revealed that the models in the training cohort (0.849, 95% CI: 0.0.787–0.911) and validation cohort (0.828, 95% CI: 0.0.727–0.929), both had good prediction accuracy. The calibration plot showed that the predicted values were consistent with the actual values. The DCA curve showed that there were positive benefits in the training and validation cohorts. The training cohort had positive benefits for all thresholds, whereas the validation cohort was positive below the threshold of 0.55. Figure 3 shows that the risk nomogram verified the predictive value of the model

Fig. 2
figure 2

Performance of the prediction model. A, B, and C show the ROC, calibration, and DCA curves of the training cohort, and D, E, and F show those of the validation cohort, respectively. The ROC results show that the models in the training cohort (0.849, 95% CI: 0.0.787–0.911) and validation cohort (0.828, 95% CI: 0.0.727–0.929), both have good prediction accuracy. The calibration plot shows that the predicted values are consistent with the actual values. The DCA curve shows that there are positive benefits in the training cohort and the validation cohort. The training cohort has positive benefits in all thresholds, while the validation cohort has positive benefits below the threshold of 0.55. ROC, receiver operating characteristic curves; DCA, decision curve analysis

Full size image
Fig. 3
figure 3

Nomogram to assess the risk of death in septic patients. To use the nomogram, first, a line is drawn from each indicator value to the points line to obtain the score. The points for all indicators are then added. Lastly, a line from the total points line to the lowest line of the nomogram is drawn to determine the risk of death

Full size image

Discussion

This retrospective single-center study demonstrated that combining the levels and clearance of IL-6, LAC, and PCT was valuable for predicting 28-day sepsis mortality. Specifically, the present study showed that combining IL-6D1, IL-6D3, PCTD1, PCTD3, and LACcD3 can provide a simple prediction model for 28-day sepsis mortality with good performance and good operability.

As a major pro-inflammatory cytokine, IL-6 levels increase in patients with infection, as well as with trauma, surgery, and neoplastic infarction [13, 30]. Several previous studies have demonstrated that IL-6 levels are correlated with sepsis mortality [17, 31,32,33]. Research suggests that a combination of initial and follow-up IL-6 levels could provide an additional prognostic value for sepsis mortality [16]. Another study reported that the prognostic value of IL-6 levels for 28-day sepsis mortality increased over time, up to 7 days [18]. In our study, IL-6D1 and IL-6D3 levels were significantly higher in non-survivors than in survivors. Although IL-6D3 was the only independent risk factor for 28-day sepsis mortality, both IL-6D1 and IL-6D3 were beneficial in the prediction model. However, IL-6 clearance was not a risk factor for sepsis in our study; thus, it was not used in developing the prediction model. There are very few studies on IL-6 clearance. A previous study found that continuous hemofiltration increases IL-6 plasma clearance but cannot change its plasma concentration effectively in systemic inflammatory response syndrome (SIRS) [34]. This suggests that controlling the production of IL-6, instead of IL-6 clearance, may be the key to treating infections.

As a marker of tissue hypoxia, LAC has been widely used as a prognostic marker for sepsis. According to the Sepsis-3 criteria, LAC concentration is a critical marker for the diagnosis of septic shock. Patients with septic shock often have sustained or increased LAC levels, which may be due to either increased LAC production or impaired LAC clearance [35]. In addition, lactic acidosis is common in sepsis and causes a higher mortality rate in septic patients [27, 36]. As LAC levels indicate a balance between elimination and production, continuous monitoring may be more effective than a single measurement. During resuscitation, organ function often improves along with high LAC clearance, which is beneficial for decreasing the mortality risk. Our study showed that LACcD3 was a valuable predictor of 28-day mortality in septic patients. Although several previous studies have shown that increased LAC levels are a valuable prognostic marker of high mortality in sepsis [16, 17, 37, 38], another study showed that critically ill patients with sepsis had normal serum LAC levels [39]. Therefore, it is better to collect serial LAC measurements and combine LAC levels with other markers to evaluate sepsis prognosis more accurately.

PCT was first used as a biomarker of infection in 1993 [40]. Some studies have found that serum PCT concentrations are correlated with the severity of sepsis [24]. Currently, it is one of the most accepted biomarkers for the prognostic prediction of sepsis-related mortality. Previous research has indicated that the change range of PCT may be different from its baseline level; therefore, the concept of clearance was used [41]. In previous studies, serum PCT measured on days 1, 3, and 5 of ICU stay was not predictive of mortality [26], but PCT clearance significantly decreased in non-survivors, compared to survivors within 48 h. Conversely, in the present study, PCT clearance was not an independent risk factor for predicting 28-day sepsis mortality. One reason for this may be that we did not focus on finding the independent risk factors but looked for one of the best models in logistic regression analysis; thus, we used stepwise regression to construct the model. The step-by-step method combines the advantages of the forward and backward methods. After each new independent variable is introduced, the substituted independent variable must be recalculated to check whether it has the value of remaining in the equation. Based on this, the introduction and elimination of independent variables are alternately carried out until no new variable can be introduced or eliminated. Another reason may be the sample size, as a previous study only enrolled 48 patients, including eight non-survivors. Since our sample size is not sufficiently large, further research is needed to confirm these results.

This study had several advantages. First, the enrolled patients were diagnosed based on the latest Sepsis-3 criteria. Second, we serially measured the concentrations of three potential biomarkers of sepsis, including IL-6, LAC and PCT. Third, we calculated the clearance of IL-6, LAC, and PCT at different time points. Fourth, the prognostic values of both the levels and clearance of these markers were demonstrated. To the best of our knowledge, this study is the first to evaluate the combined predictive performance of IL-6, LAC, and PCT levels, and clearance. Owing to the small amount of data, the main purpose of the present study was to preliminarily evaluate the predictive value of continuous monitoring of these indicators for sepsis. In future studies, we will collect more cases to complete a more rigorous prediction model and simultaneously verify the model at the same time.

Study limitation

This study has some limitations. First, it was performed at a single center. Multicenter-based research with a larger quantity of data will still be necessary to confirm the results. Second, we excluded patients hospitalized for less than 24 h, and most of these patients died. For these critically ill patients, biomarkers with early diagnostic and prognostic values are urgently needed. Third, the results of this study may not adapt universally to other septic patients in ER because of our exclusion criteria. Such as cancer patients were excluded in our study, so further research is necessary to verify whether this result is applicable to tumor patients with sepsis.

Conclusions

The combination of IL-6D1, IL-6D3, PCTD1, PCTD3, and LACcD3 may be used to build a prediction model for 28-day sepsis mortality. These results still need to be verified in a larger sample size through multicenter-based research.