Mortality prediction using a novel combination of biomarkers in the first day of sepsis in intensive care units

Early identification of infection severity and organ dysfunction is crucial in improving outcomes of patients with sepsis. We aimed to develop a new combination of blood-based biomarkers that can early predict 28-day mortality in patients with sepsis or septic shock. We enrolled 66 patients with sepsis or septic shock and compared 14 blood-based biomarkers in the first 24 h after ICU admission. The serum levels of interleukin-6 (IL-6) (median 217.6 vs. 4809.0 pg/ml, P = 0.001), lactate (median 2.4 vs. 6.3 mmol/L, P = 0.014), N-terminal prohormone of brain natriuretic peptide (NT-proBNP) (median 1596.5 vs. 32,905.3 ng/ml, P < 0.001), prothrombin time (PT) (median 15.6 vs. 20.1 s, P = 0.030), activated partial thrombin time (APTT) (median 45.1 vs. 59.0 s, P = 0.026), and international normalized ratio (INR) (median 1.3 vs. 1.8, P < 0.001) were significantly lower in the survivor group. IL-6, NT-proBNP, and INR provided the best individual performance in predicting 28-day mortality of patients with sepsis or septic shock. Furthermore, the combination of these three biomarkers achieved better predictive performance (AUC 0.890, P < 0.001) than conventional scoring systems. In summary, the combination of IL-6, NT-proBNP, and INR may serve as a potential predictor of 28-day mortality in critically ill patients with sepsis or septic shock.

Biomarker measurements. Blood samples were collected during first 24 h at ICU admission, and stored frozen until biomarker assays were performed (per manufacturer's instructions) in the Laboratory Department of Shenzhen People's Hospital. The serum levels of 14 biomarkers were assayed: immunoturbidimetry method (Roche) for high-sensitivity C-reactive protein (hs-CRP), enzyme-linked fluorescent assay (VIDAS) for PCT, troponin I (Tn-I) and D-dimer, electro-chemiluminescence immunoassay (Roche) for IL-6 and NT-proBNP, colorimetric assay (Roche) for serum lactate, enzymatic colorimetric method (Roche) for serum creatinine (Cr), colorimetric diazo method (Roche) for total bilirubin, impedance and single light-scatter method for platelet (PLT) counting, coagulation tests (Sysmex) for prothrombin time (PT), activated partial thrombin time (APTT) and fibrinogen (Fib), and international normalized ratio (INR) was calculated from the following formula: INR = patient PT/mean normal PT) ^International Sensitivity Index (ISI). If the levels of biomarkers were measured multiple times in the first 24 h, the average level of each biomarker was used in our study.
Data collection. We collected the following data: baseline demographic information, comorbidities, use of vasopressors, use of mechanical ventilator (MV), use of dialysis, length of hospital and ICU stay; mortality in 28 days; clinical indicators including mean arterial blood pressure (MAP), Glasgow Coma Scale (GCS) 10 , SOFA 11 , and APACHEII scores 12 . These scoring systems were calculated based on the data obtained during www.nature.com/scientificreports/ the first 24 h after ICU admission. Furthermore, the baseline pulse-induced contour cardiac output (PICCO) parameters of cardiac preload including the intrathoracic blood volume (ITBV) and global end diastolic volume (GEDV) were also collected. The follow-up phone calls were conducted with all subjects discharged from the hospital to collect the information for assessing 28-day mortality.
Statistical evaluation. Statistical Package for Social Sciences (SPSS), version 19.0 (SPSS Inc, Chicago, IL) was used for statistical evaluations, and then the normal distribution of the data was confirmed by Kolmogorov-Smirnov test. Normally and non-normally distributed continuous variables were summarized as the mean (standard deviation, SD) and as the median (interquartile range, IQR), respectively. Data was analyzed by T test, post hoc LSD test or Mann-Whitney U test when applicable. The chi-squared test was for categorical variables. The univariate and multivariate logistic regression analyses were used to analyze the independent effects of various parameters on 28-day mortality. In order to identify the correlation between blood biomarkers and APACHEII or SOFA, variables were analysed using the Pearson correlation coefficients. The Receiver operating characteristic (ROC) curves were performed to predict the association of biomarkers with 28-day mortality in patients with sepsis 13 . A P value of less than 0.05 (P < 0.05) was considered as a statistically significant difference.

ROC curve analysis of blood-based biomarkers for predicting 28-day mortality.
The ability of selected biomarkers, SOFA, and APACHE II scores to predict the 28-day mortality in patients with sepsis according to ROC curve analysis was shown in Table 4 and Fig. 2 www.nature.com/scientificreports/ of IL-6, NT-proBNP, INR, lactate and PT in all patients was positively associated with APACHEII (r = 0.492, 0.535, 0.471, 0.371, 0.471, respectively; P all < 0.05) and SOFA scores (r = 0.495, 0.572, 0.288, 0.347, 0.355, respectively; P all < 0.05). The APTT levels were positively associated with APACHEII scores (r = 0.298, P = 0.015). However, we didn't find an association between APTT levels and SOFA scores (P = 0.222). More details were shown in Table 5 and Fig. 3.

Discussion
We evaluated the inflammatory and organ-specific biomarkers to predict 28-day mortality for critically ill patients with sepsis, and found that a novel combination of INR, NT-proBNP and IL-6 within first 24 h of ICU admission may rapidly predict mortality risk in patients with sepsis or septic shock. This combination of three markers is solely based on cardiac dysfunction, coagulation disorders, and inflammatory responses, which are readily available during the patient's course of ICU stay. Therefore, it may help improve risk stratification, early assessment and intervention of sepsis patients in ICU. Our findings also suggest that the optimal combination of INR, NT-proBNP and IL-6 has a significantly higher AUC value than that of conventional scoring systems such as SOFA and APACHEII scores, and provides need to confirm its validity for screening sepsis-related mortality. Sepsis is also associated with coagulopathy ranging from hypercoagulability to acute disseminated intravascular coagulation (DIC), which may be likely play a key role in multiple organ dysfunction syndrome (MODS) 14 .
In particular, hemostasis-related parameter has been shown to be a predictor of poor outcomes in ICU patients. Thrombocytopenia occurs in up to 40% of patients on admission, and an INR ≥ 1.5 is present in approximately two-thirds of patients. We found that the INR value in survivors was significantly lower than in non-survivors, and strongly associated with greater risk of serious illness. Furthermore, multivariate analysis demonstrated that INR level was an independent factor for critically ill patients with sepsis, and high INR (> 1.47) was significantly associated with increased mortality risk. In fact, because coagulopathy (INR > 1.5) is defined as one of the critical characteristics in patients with sepsis according to sepsis 3.0 criteria, this definition would partly support our results 15 . However, some authors still questioned the use of INR as a management tool because INR value can be falsely increased, and not necessarily associated with risk of bleeding events 16 . Our study also showed that the Table 1. Demographic characteristics at ICU admission. IQR range interquartile, BMI body mass index, MAP mean arterial pressure, SOFA sequential organ failure assessment, GCS Glasgow coma scale, APACHE II acute physiology and chronic health score II, PICCO Pulse-induced contour cardiac output, GEDI Global end-diastolic blood volume, ITBV Intrathoracic blood volume, ICU intensive care unit. *Indicates a significant value, P < 0.05.

Characteristics
Survivor group (n = 49) Non-survivor group (n = 17) Total (n = 66) P value www.nature.com/scientificreports/ INR value of 1.47 has a sensitivity of 81%, specificity of 76% and AUC of 0.748, which may limit its use alone at accurate assessment on the outcomes of sepsis patients.
Since NT-proBNP is mainly secreted by cardiomyocytes in response to increased ventricular volume or pressure load, the level of this biomarker is known to be highly sensitive to detect acute heart failure, and correlated with the severity of heart failure 17,18 . In addition, low-grade systematic inflammation and insulin resistance are other possible causes of elevated level of NT-proBNP in patients with sepsis 19 . Previous studies revealed the levels of NT-proBNP in sepsis patients without heart failure were associated with in-hospital mortality 20 . Therefore, it has been accepted that the increased levels of NT-proBNP in septic shock patients are mainly due to sepsis rather than cardiac insufficiency. In agreement with previous studies, we found that on the first day of ICU admission, the levels of NT-proBNP in non-survivor group were significantly higher than in survivor group, and the higher Table 2. Biomarkers in blood of the study groups. IQR range interquartile, hs-CRP high-sensitivity C-reactive protein, PCT procalcitonin, IL-6 Interleukin 6, NT-proBNP N-terminal prohormone of brain natriuretic peptide, Tn-I Troponin I, Cr creatinine, PLT Platelet, PT prothrombin time, APTT activated partial thrombin time, Fib fibrinogen, INR international normalized ratio. *Indicates a significant value, P < 0.05.  www.nature.com/scientificreports/ NT-proBNP levels were associated with poorer health status, including higher SOFA scores and higher APACHE II scores. Moreover, the volume challenge during resuscitation process might be a possible reason for functional heart failure, and had some influence on the increased levels of NT-proBNP. In this study, we also found that NT-proBNP also presented as an independent risk factor for sepsis mortality. At a cut-off value of 9086.59 ng/ ml, NT-proBNP had a sensitivity of 70% and a specificity of 80% with an AUC of 0.825. These findings suggested NT-proBNP may be a potential predictor of sepsis mortality, but needs to increase its sensitivity and specificity by combining with other parameters. Elevated levels of serum IL-6 have been found in both neonatal and adult sepsis 21,22 . Previous studies have shown that IL-6 may be induced by infections and tissue injuries, and associated with a high risk of death in septic shock 23 . However, IL-6 is a multifunctional cytokine that has pro-and anti-inflammatory properties, and can be affected by different conditions such as exercise, aging, certain cancers, frailty, and chronic inflammatory response 24 . These factors should be comprehensively considered for a better understanding of sepsis when IL-6 Table 4. Prognostic value of clinical parameters. AUC Area under the curve, CI confidence interval, IL-6 Interleukin 6, INR international normalized ratio, NT-proBNP N-terminal prohormone of brain natriuretic peptide, PT prothrombin time, APTT activated partial thrombin time, SOFA sequential organ failure assessment, APACHE II acute physiology and chronic health score II. *Indicates a significant value, P < 0.05.  www.nature.com/scientificreports/ levels are elevated. In this study, we observed that the IL-6 levels measured on day 1 were significantly increased in non-survivors when compared with survivors, and positively correlated with SOFA and APACHEII scores. A cut-off value for IL-6 on 2580.5 pg/ml was found to have a moderate sensitivity of 63% and a high specificity of 90% (AUC 0.785) for predicting mortality. The results suggested IL-6 may be used as a prognostic biomarker for 28-day mortality in patients with sepsis, while the predictive accuracy of IL-6 is lower than that of NT-proBNP.
Various studies have demonstrated that the APACHE II and SOFA scoring systems are good tools for predicting mortality in ICU patients 25,26 . Unlike APACHE II system, the SOFA was not initially designed to predict Table 5. Correlations between blood biomarkers and APACHE II or SOFA scores in all subjects. *Indicates a significant value, P < 0.05. IL-6 Interleukin 6, INR international normalized ratio, NT-proBNP N-terminal prohormone of brain natriuretic peptide, PT prothrombin time, APTT activated partial thrombin time, APACHE II acute physiology and chronic health score II, SOFA sequential organ failure assessment.  www.nature.com/scientificreports/ hospital mortality, while it has been broadly used in a range of applications such as organ dysfunction assessment, predicting hospital survival, and identifying non-ICU patients at risk of sepsis. Furthermore, the specificities of APACHE II and SOFA scores were found to be somewhat low in identifying the outcomes in different subgroups of ICU patients 27 . In the present study, we also revealed that in predicting 28-day mortality, APACHE II score of > 27.5, and SOFA score > 7.5 performed at sensitivities of 94%, and 71%, and specificities of 69% and 76%, respectively. Even though APACHE II score performed well in mortality risk stratification in the ICU patients with sepsis, it tended to a higher sensitivity and lower specificity, leading to an overestimation of mortality risk (and consequently, a higher burden on hospital costs). Moreover, the combination of three biomarkers, INR, NT-proBNP and IL-6 yielded an ROC value of 0.890, which provided higher diagnostic accuracy (sensitivity of 81% and specificity of 90%) than any single biomarker, APACHE II and SOFA scores. This result showed that the three-marker combination may have the best performance in predicting sepsis-related mortality. Therefore, instead of just screening APACHE II and SOFA scores during the initial 24 h in the ICU, we prefer to monitor the three-marker combination to predict the short outcome of sepsis patients. Once the patient meets the cut-off levels of the combination, we should closely observe the changes of multiple physiological systems, and frequently adjust medical therapies accordingly to reduce the risk of death. Furthermore, these three markers together may help to early stage the disease, predict prognosis and response of intervention in critically ill patients with sepsis. Our study has several limitations. First, it was a single-center study, and the small size of study population is the major limitation. Therefore, larger observations of multiple centers should be carried out to verify the results. Second, we did not include other clinical parameters such as arterial blood gas analysis and white blood cell count, because these variables have not shown a sufficient predictive power for the outcomes of sepsis patients. Future studies need to elucidate the potential application of other clinical biomarkers for prognosis in sepsis, including but not limited to the parameters associated with dysfunction of major organs.
Despite the above limitations, our findings indicate the combination of INR, NT-proBNP and IL-6 may represent a valuable tool for predicting mortality risk in critically ill patients with sepsis or septic shock. Furthermore, this simple combination shows a higher prognostic accuracy for 28-day mortality than the SOFA or APACHE II scores, and could have a potential application in guiding early management of sepsis patients in ICU. Future studies are necessary to explore whether the combination can improve patient managements and outcomes in ICU.

Data availability
Some or all data, or code generated or used during the study are available from the corresponding author by request.