LACTATE, PROCALCITONIN, AND AMINO-TERMINAL PRO-B-TYPE NATRIURETIC PEPTIDE VERSUS CYTOKINE MEASUREMENTS AND CLINICAL SEVERITY SCORES FOR PROGNOSTICATION IN SEPTIC SHOCK

Abstract

The biomarkers lactate, procalcitonin, and amino-terminal pro-B-type natriuretic peptide (NT-proBNP) are often promoted as being useful for prognostication in septic shock. This study aimed to compare the prognostic utility of these biomarkers with each other and with cytokine measurements and clinical severity scores, and to assess how these biomarkers may be combined to improve their prognostic utility. Seventy-two patients with septic shock were studied. The biomarkers were measured on the first 3 days of stay in the intensive care unit together with serum IL-1β, IL-6, IL-10, and TNF-α levels. Although elevated baseline lactate levels predicted 28-day mortality, elevated procalcitonin and NT-proBNP levels were only predictive from days 2 and 3, respectively. The prognostic utility of baseline lactate levels was poorer than that of baseline cytokine levels, the Acute Physiology and Chronic Health Evaluation II score, and the Sequential Organ Failure Assessment score. However, a rising trend in lactate and procalcitonin levels between days 1 and 2 had superior prognostic utility compared with absolute levels. Indeed, using multivariate analysis, the presence of a concurrent increase in both lactate and procalcitonin levels between days 1 and 2 superseded all cytokine measurements and clinical severity scores as the sole independent predictor of 28-day mortality. In conclusion, elevated baseline lactate levels offer superior prognostic accuracy to baseline procalcitonin levels, which in turn are superior to NT-proBNP levels. To improve their prognostic utility beyond those of cytokine measurements and clinical severity scores, serial lactate and procalcitonin measurements may be combined.

ABBREVIATIONS - APACHE-Acute Physiology and Chronic Health Evaluation; CI-confidence interval; ICU-intensive care unit; NPV-negative predictive value; NS-not significant; NT-proBNP-amino-terminal pro-B-type natriuretic peptide; OR-odds ratio; PPV-positive predictive value; ROC-receiver operating characteristics; r_s-Spearman ρ; SOFA-Sequential Organ Failure Assessment

INTRODUCTION

Severe sepsis remains a major cause of death today. In the presence of septic shock, mortality increases to approximately 60% (¹). Biomarkers are increasingly used to facilitate the early diagnosis and management of sepsis (²). Importantly, biomarkers also assist the clinician in prognostication. For a biomarker to be considered truly useful for prognostication, it should fulfill certain requirements. It should be easily measured and widely available. It should also provide prognostic information early in the course of the illness and, thus, give the clinician the opportunity to intervene quickly in an attempt to improve survival. However, it should be as accurate as-or better still, more accurate than-clinical severity scores for prognostication.

Three biomarkers that have recently featured prominently in both the clinical and the research settings are lactate, procalcitonin, and amino-terminal pro-B-type natriuretic peptide (NT-proBNP). Each of these biomarkers serves a distinct clinical purpose in the management of severe sepsis, for example, high lactate levels due to exaggerated aerobic or anaerobic glycolysis are used as an indication for early goal-directed therapy (³); high procalcitonin levels suggest bacterial infections, which trigger a release of this hormone via an increase in gene expression (⁴); and high NT-proBNP levels imply septic myocardial dysfunction with myocyte stretch and cytokine release (⁵). Concurrently, these same biomarkers have also been promoted as being useful for prognostication. They are widely available and contrast with cytokines, which are often only measured in research settings due to the high costs, logistical issues, and limited clinical use involved. However, several questions concerning how best to use them for prognostication remain unanswered.

Thus, the aim of our study was to answer these questions. First, is lactate, procalcitonin, or NT-proBNP more useful for prognostication in septic shock? Second, how do they compare with cytokines and clinical severity scores? Third, because there has been a recent drive toward the concurrent use of multiple biomarkers for the diagnosis of sepsis (^{6, 7}), can measurements of these biomarkers also be combined to improve their prognostic utility?

MATERIALS AND METHODS

Study population

The only inclusion criterion for the study was the presence of septic shock early within 24 h of admission to the medical intensive care unit (ICU) of the Singapore National University Hospital. Septic shock was defined according to the 2001 International Sepsis Definitions Conference (⁸), that is, sepsis with hypotension despite adequate volume resuscitation. The diagnosis of sepsis required the presence of the systemic inflammation in response to infection-patients had to have a known infection or a suspected infection, as evidenced by one or more of the following: white cells in a normally sterile body fluid, perforated viscus, radiographic evidence of pneumonia in association with the production of purulent sputum, and a syndrome associated with a high risk of infection (e.g., ascending cholangitis). We excluded patients presenting with acute coronary syndromes and acute heart failure with cardiogenic pulmonary edema, and patients for whom withdrawal of intensive life support was considered early upon admission.

From February 2004 to April 2005, we identified 77 consecutive patients with septic shock. Seventy-two patients were enrolled after obtaining informed consent from themselves or their next of kin. Informed consent was unavailable in the remaining 5 patients. The study protocol was approved by the hospital's institutional review board.

Measurements of biomarker and cytokine levels

For all 72 enrolled patients, we measured plasma lactate, serum procalcitonin, and serum NT-proBNP levels at baseline (day 1 of ICU admission) and on days 2 and 3. We measured serum IL-1β, IL-6, IL-10, and TNF-α levels on day 1 in all 72 patients. These proinflammatory and anti-inflammatory cytokines were chosen because of the key roles they play in the pathophysiology of sepsis (^{8, 9}). Among these 72 patients, we repeated cytokine measurements on days 2 and 3 in 31 consecutive patients in the latter half of the study period from January 2005 onward. Blood was collected from arterial catheters and dispatched on ice to the hospital laboratory for immediate processing according to the respective manufacturers' protocols: using the colorimetric method for lactate (VITROS LAC slides; Ortho-Clinical Diagnostics, New York, NY) and using chemiluminescence immunoassays for procalcitonin (LumiTest; Brahms Diagnostica, Berlin, Germany), NT-proBNP (Elecsys 2010; Roche Diagnostics, Mannheim, Germany), IL-1β, IL-6, IL-10, and TNF-α (all IMMULITE; Diagnostics Products Corporation, Los Angeles, Calif). The attending physicians and investigators were blinded to the cytokine levels throughout the study.

Other data collection

We recorded the following variables on ICU admission for each patient: age, sex, diagnoses, vital signs, the Acute Physiology and Chronic Health Evaluation (APACHE) II score, Sequential Organ Failure Assessment (SOFA) score, and microbiological culture results. When available, serial hemodynamic data using transpulmonary thermodilution (PiCCO; Pulsion Medical System, Munich, Germany) were recorded simultaneously during blood sampling for biomarker and cytokine levels (n = 27). The catheters for transpulmonary thermodilution were inserted at the discretion of the attending physicians when it was deemed that additional hemodynamic data would be useful. We followed the patients for as long as 28 days, with the primary outcome measure being 28-day all-cause mortality.

Clinical management

The attending physicians administered all therapies for the patients according to local guidelines without interference by the investigators (⁹). Fluid was administered until it was decided that patients were no longer fluid responsive, with a minimum central venous pressure of at least 8 cm H₂O. Vasopressors were used to keep the mean arterial pressure ≥ 65 mmHg. Hydrocortisone was given intravenously (50 mg/6 h) and discontinued when vasopressors were no longer required or if there was a rise of more than 250 nM in cortisol levels at 30 to 60 min after a 250-μg adrenocorticotrophic hormone stimulation test, that is, responders. Broad-spectrum antibiotics were administered early and de-escalated according to clinical response and culture results. Antibiotic therapy was continued for between 7 and 14 days. Lung-protective ventilatory support, tight glucose control (kept at 4.4 - 6.1 mM), deep vein thrombosis, and stress ulcer prophylaxis and renal replacement therapy, when indicated, also formed part of the management strategy.

Statistical analysis

We expressed the test data as frequencies for nominal variables and as mean ± SD or medians with range for continuous variables. We performed univariate analyses using the χ² test or Fisher exact test for nominal variables, the t test for means, and the Mann-Whitney U test for medians. Spearman ρ (r_s) was calculated to assess the correlation between biomarkers and transpulmonary thermodilution measurements. To assess the time course of the various biomarkers and cytokines, repeated-measure ANOVA and the Friedman test were performed accordingly. We compared the characteristics of survivors versus nonsurvivors using univariate analysis and constructed receiver operating characteristics (ROC) curves to determine how well the biomarkers and cytokines predicted 28-day mortality.

In addition to absolute levels, we compared the 28-day mortality of patients in which each biomarker level rose (or did not change) between days 1 and 2 versus patients in which each biomarker level fell or remained within normal reference ranges (lactate ≤ 2.0 mM, procalcitonin ≤ 0.50 ng/mL, NT-proBNP ≤ 125 pg/mL [according to the manufacturer's package insert]).

Those variables with P values less than 0.05 on univariate analysis were then entered into a multivariate logistic regression analysis to further identify the independent predictors of 28-day mortality and their adjusted odds ratios (OR) with 95% confidence intervals (CIs) (95% CI). A P value less than 0.05 was considered significant. All tests were two-tailed. We used the SPSS statistical software package 11.5 (SPSS, Inc., Chicago, Ill) for all statistical analyses.

RESULTS

Characteristics of the study population

The baseline characteristics of the patients are shown in Table 1. All 72 patients had septic shock and were deemed to have infection based on the available clinical findings. Twenty-three patients received a single antibiotic initially (a carbapenem, cephalosporin or β-lactam/β-lactamase inhibitor), whereas the other 49 patients received multiple antibiotics initially (a combination of carbapenems, cephalosporins, β-lactam/β-lactamase inhibitors, penicillins, aminoglycosides, quinolones, and antifungals). There was microbiological evidence of infection in 50 of these patients, all of whom received appropriate initial antibiotic therapy. All patients received hydrocortisone for possible relative adrenal insufficiency initially, although this was discontinued within 1 day in 25 patients who were deemed as responders to the adrenocorticotrophic hormone stimulation test. The 28-day mortality rate was 41.7%. Compared with survivors, nonsurvivors had higher baseline APACHE II and SOFA scores.

Table 1: Clinical characteristics of patients on admission to ICU

Prognostic use of absolute lactate, procalcitonin, and NT-proBNP levels

On day 1, only high lactate levels were predictive of 28-day mortality. High procalcitonin levels were only predictive from day 2, and high NT-proBNP levels were only predictive from day 3 (Fig. 1).

Fig. 1: Time course of lactate, procalcitonin, and NT-proBNP levels in survivors versus nonsurvivors. White plots depict survivors (n = 40), and gray plots depict nonsurvivors (n = 32). Box plots represent the 25th and 75th percentiles, with the internal horizontal lines showing the median and T bars the 10th and 90th percentiles. Circles represent outliers. *P < 0.01 and ^† P < 0.05 for comparisons between groups on the same day (Mann-Whitney U test). ^‡ P value refers to the difference in the overall time course (i.e., trends) of blood levels between survivors and nonsurvivors (repeated-measures ANOVA).

Prognostic use of trends in lactate, procalcitonin, and NT-proBNP levels

Biomarker levels showed an early general upward trend in nonsurvivors that was not seen in survivors (Fig. 1). Specifically, patients in which lactate or procalcitonin levels rose or did not change between days 1 and 2 had a significantly higher 28-day mortality rate than patients in which levels fell or remained within normal reference changes (Table 2). This finding did not achieve statistical significance for NT-proBNP.

Table 2: Rise in biomarker levels between days 1 and 2 as a predictor of 28-day mortality

Prognostic use of combining lactate and procalcitonin

Patients in which both biomarkers (lactate and procalcitonin) increased from day 1 to 2 had a high 28-day mortality rate of 86.7%. On the other hand, patients in which none or only one of these two biomarkers increased had a lower 28-day mortality rate of close to 20% (Fig. 2).

Fig. 2: Combining trends in lactate and procalcitonin levels between days 1 and 2 to predict 28-day mortality.

Univariate comparisons of biomarkers with cytokines and clinical severity scores

Absolute levels of IL-1β, IL-6, and IL-10 were predictive of 28-day mortality on day 1 (Fig. 3), which was earlier than procalcitonin and NT-proBNP levels. Unlike lactate, procalcitonin and NT-proBNP, however, the time course of cytokine concentrations did not seem to predict survival because their concentrations decreased over time in both survivors and nonsurvivors (Fig. 3).

Fig. 3: Time course of cytokine levels in survivors versus nonsurvivors. Nonparametric data subjected to logarithmic transformation (log₁₀) to facilitate depiction. White plots depict survivors (n = 20), and gray plots depict nonsurvivors (n = 11). Box plots represent the 25th and 75th percentiles, with the internal horizontal lines showing the median and T bars the 10th and 90th percentiles. Circles represent outliers. *P < 0.01 and ^† P < 0.05 for comparisons between groups on the same day (Mann-Whitney U test).

Based on ROC curve analyses, the accuracy of the APACHE II score and cytokines IL-10, IL-6, and IL-1β on day 1 for the prediction of 28-day mortality was moderate (area under the curve, >0.7), whereas the accuracy of the SOFA score and lactate levels on day 1 was low (area under the curve, <0.7) (Table 3).

Table 3: Univariate predictors of 28-day mortality on admission

Multivariate comparisons of biomarkers with cytokines and clinical severity scores

The baseline day 1 variables that were found to be significantly different between survivors and nonsurvivors on univariate analysis (APACHE II and SOFA scores, IL-1β, IL-6, IL-10, and lactate levels) were entered into a logistic regression model. Among these variables, only one variable remained independently associated with 28-day mortality: a high APACHE II score. An increase of one point for the APACHE II score was associated with an OR of dying of 1.15 (95% CI, 1.02 - 1.29; P = 0.02). The calibration of the model was adequate using the Hosmer and Lemeshow goodness-of-fit test (P = 0.31).

Given the univariate association between 28-day mortality and the concurrent increase in both lactate and procalcitonin levels between days 1 and 2 (Fig. 2), the latter was substituted for admission lactate levels in the logistic regression model, and the multivariate analysis was repeated. As a result, the APACHE II score was no longer independently associated with mortality. Instead, the occurrence of an increase in both lactate and procalcitonin levels between days 1 and 2 became the sole independent predictor of 28-day mortality (OR, 3.61; 95% CI, 1.43-9.09; P = 0.006). Again, the calibration of the model was adequate using the Hosmer and Lemeshow goodness-of-fit test (P = 0.73).

Other correlations and associations

Lactate levels were correlated with stroke volume index measured via transpulmonary thermodilution on days 2 (r_s = −0.64) and 3 (r_s = −0.49); NT-proBNP levels were also correlated with stroke volume index on days 2 (r_s = −0.66) and 3 (r_s = −0.47) (P all < 0.05). There was no correlation between stroke volume index and procalcitonin. Median lactate levels on day 2 were higher in patients with transpulmonary dilution catheters (2.7 mM; range, 0.9-12.0 mM) versus patients without (1.6 mM; range, 0.5-15.9 mM) (P = 0.02); there was no difference in procalcitonin and NT-proBNP levels.

Trends in lactate, procalcitonin, and NT-proBNP levels were not affected by whether patients had single or combination antibiotic therapy and whether patients were responders or nonresponders to the adrenocorticotrophic hormone stimulation test.

DISCUSSION

We found that in patients with septic shock, while elevated baseline lactate levels were predictive of 28-day mortality, elevated procalcitonin and NT-proBNP levels were only predictive from days 2 and 3, respectively. Even so, the prognostic utility of absolute baseline lactate levels was poorer than that of baseline cytokine levels and clinical severity scores. A rising trend in lactate and procalcitonin levels between days 1 and 2 had superior prognostic use compared with absolute lactate and procalcitonin levels. Indeed, the presence of a concurrent increase in both lactate and procalcitonin levels between days 1 and 2 superseded all cytokine measurements and clinical severity scores as the sole independent predictor of 28-day mortality.

The limitations of our study should be recognized. Not all-but 69%-of our patients had microbiologically proven infections. Nevertheless, this finding is consistent with the current epidemiology of septic shock (^{1, 8}). As laid out by the 2001 International Sepsis Definitions Conference, sepsis is often strongly suspected and not microbiologically proven (⁸). Our study is also limited by its relatively small sample size. As such, the wide variation in the levels of the biomarkers and cytokines measured may affect the reliability and validity of our findings. Given that we restricted our study population to patients with septic shock, caution should be exercised before extrapolating our results to all infected patients.

Several novel features of our study deserve highlighting. Although many investigators have evaluated the use of biomarkers in sepsis, to the best of our knowledge, no study before ours have compared the prognostic use of all the three commonly used biomarkers lactate, procalcitonin, and NT-proBNP. In addition, no data exist to compare the prognostic utility of all these biomarkers versus cytokines and clinical severity scores using a multivariate logistic regression model. Importantly, it is also not known how these biomarkers may be combined to optimize their prognostic value.

The prompt rise in lactate levels in nonsurvivors demonstrated in our study, as compared with procalcitonin and NT-proBNP, offers the clinician a greater opportunity to take the necessary actions early in the course of the illness in high-risk patients. This is exemplified by the use of early goal-directed therapy, with which septic patients with elevated lactate levels on presentation may be rescued (³). A review of the literature supports our results that baseline lactate measurements may indeed be more consistent than procalcitonin and NT-proBNP measurements as prognosticators. Multiple studies have shown that elevated baseline lactate levels are associated with higher mortality (^10-16). A recent study by Clec'h et al. (¹³) suggested that baseline lactate levels may have superior prognostic accuracy to procalcitonin levels. Although some studies have shown an association between baseline procalcitonin levels and mortality (^{13, 17-20}), others have not (^21-23). Presumably due to the multiple heterogeneous causes of elevated natriuretic peptide levels in sepsis (²⁴), a similarly confusing picture applies to NT-proBNP and its counterpart BNP, with studies finding no association between baseline levels and mortality (^24-29) existing alongside those that found an association (^{11, 30, 31}). Indeed, our results are consistent with previous data that absolute procalcitonin and natriuretic peptide levels beyond day 1 offer more accurate prognostic information than baseline levels (^{13, 20, 27, 31}).

Ideally, a biomarker that is used for prognostication should equal or outperform clinical severity scores in the prediction of outcomes. Although a handful of studies have demonstrated that lactate measurements are independent predictors of outcome using multivariate analysis of factors, including several clinical severity scores (SOFA score, organ system failure index, and Simplified Acute Physiology Score II) (^{10, 14, 15, 19, 32}), little data exist to compare the clinical use of lactate measurements versus the APACHE II score. In our study, elevated lactate levels had a lower sensitivity and specificity than the APACHE II and SOFA scores for predicting mortality. However, the finding of rising lactate or even procalcitonin levels between days 1 and 2 offered greater prognostic accuracy than baseline levels in our study, findings that are consistent with the results of prior studies (^{12, 16, 17, 22, 33}). In fact, a concurrent increase in both lactate and procalcitonin levels, and not clinical severity scores, was the sole independent predictor of 28-day mortality (OR, 3.61).

There has been a recent trend toward the combined use of several biomarkers and cytokines in sepsis. Although some investigators have used various permutations of biomarkers for the diagnosis of sepsis (^{6, 7}), others have combined various cytokine measurements (including IL-6, IL-10, and TNF-α) for prognostication (^34-36). Unfortunately, although some data suggest that baseline IL-6 measurements are more predictive of outcomes than baseline procalcitonin measurements (^{21, 33}), to date, the clinical use of cytokine measurements remains limited in general. This is not the case for lactate and procalcitonin. The former has been used to guide resuscitation (³), the latter, to aid in the diagnosis of infection and thereby guide antibiotic therapy on presentation (⁴). As such, they are frequently measured by clinicians. Thus, it is exciting to find that these two biomarkers, when used together and serially, may serve an additional role of prognostication that provides even greater accuracy than various cytokine measurements and all the components of the APACHE II and SOFA scores put together. This provides the clinician with a convenient means of risk stratification.

As in all models for prognostication, whether this method of risk stratification will actually benefit high-risk patients remains to be determined. Future prospective studies should evaluate the efficacy of clinical management algorithms in which physicians increase resuscitative efforts on day 2 of ICU admission if lactate levels rise and reevaluate the current antibiotic therapy if procalcitonin levels rise. The added costs of performing serial lactate and procalcitonin measurements will be justified if outcome benefits using such algorithms are demonstrated.

In conclusion, elevated baseline lactate levels offer superior prognostic accuracy to baseline procalcitonin levels, which in turn are superior to NT-proBNP levels in septic shock. To further improve their prognostic utility, lactate and procalcitonin measurements may be combined because a concurrent increase in both biomarker levels between days 1 and 2, but not cytokine measurements and clinical severity scores, independently predicts 28-day mortality.

ACKNOWLEDGMENTS

The authors thank the residents and nurses of the National University Hospital's medical ICU, Dr. Sharon Saw and the staff of the Department of Laboratory Medicine's Clinical Chemistry Division, and E-Ling Toh of the Clinical Trials Unit for all the assistance rendered throughout the study.