Intensive Care Med (2016) 42:1970–1979DOI 10.1007/s00134-016-4338-z

ORIGINAL

Plasma cytokine levels predict response to corticosteroids in septic shockPeter Bentzer1,2,3,4*, Chris Fjell1,2, Keith R. Walley1,2, John Boyd1,2 and James A. Russell1,2

© 2016 Springer-Verlag Berlin Heidelberg and ESICM

Abstract

Purpose: To investigate if plasma cytokine concentrations predict a beneficial response to corticosteroid treatment in septic shock patients.

Methods: A cohort of septic shock patients in whom a panel of 39 cytokines had been measured at baseline (n = 363) was included. Patients who received corticosteroids were propensity score matched to non-corticosteroid-treated patients. An optimal threshold to identify responders to corticosteroid treatment for each cytokine was defined as the concentration above which the odds ratio for 28-day survival between corticosteroid- and non-corti-costeroid-treated patients was highest.

Results: Propensity score matching partitioned 165 patients into 61 sets; each set contained matched corticoster-oid- and non-corticosteroid-treated patients. For 13 plasma cytokines threshold concentrations were found where the odds ratio for survival between corticosteroid- and non-corticosteroid-treated patients was significant (P < 0.05). CD40 ligand was associated with the highest odds ratio and identified 21 % of the patients in the propensity score matched cohort as responders to corticosteroid treatment. Combinations of triplets of cytokines with a significant odds ratio, using the thresholds identified above, were tested to find a higher proportion of responders. IL3, IL6, and CCL4 identified 50 % of the patients in the propensity score matched cohort as responders to corticosteroid treat-ment. The odds ratio for 28-day survival was 19 (95 % CI 3.5–140, P = 0.02) with a concentration above threshold for a least one of these cytokines.

Conclusion: Plasma concentration of selected cytokines is a potential predictive biomarker to identify septic shock patients that may benefit from treatment with corticosteroids.

Keywords: Corticosteroids, Predictive biomarker, Cytokines, Propensity score matching, Septic shock

IntroductionTreatment with systemic corticosteroids has been used for many years to modulate the acute inflammatory response of sepsis with conflicting results [1–6]. Because septic shock is a heterogeneous condition both with regard to pathogens and host response, attempts have

been made to identify subgroups of patients with septic shock that may benefit from corticosteroid treatment. Although sepsis is shown to increase plasma concentra-tions of cortisol, this response has been suggested to be inadequate in some patients [7, 8]. Based on this notion the corticotropin stimulation test was proposed as a bio-marker to identify the subgroup of patients with relative adrenal insufficiency that may benefit from hydrocor-tisone replacement therapy [2]. However, a subsequent trial could not validate these results and the use of the corticotropin test to identify patients who may benefit from corticosteroid replacement is now discouraged in the Surviving Sepsis Campaign guidelines [6, 9].

*Correspondence: peter.bentzer@med.lu.se 4 Department of Anesthesiology and Intensive Care, Helsingborg Hospital, 251 87 Helsingborg, SwedenFull author information is available at the end of the article

Take-home message: Plasma cytokine concentrations may be used to identify patients with septic shock who benefit from corticosteroid treatment.

1971

Thus, there is a need to define whether there is a sub-group of patients who do respond to corticosteroids. Accordingly, we took an alternative approach by assess-ing plasma cytokines for the following reasons. Increased plasma pro- and anti-inflammatory cytokines are asso-ciated with poor outcome following sepsis [10–12] and plasma cytokine levels are good predictors of outcome independent of the clinical status suggesting that the plasma cytokine levels provide information not read-ily available clinically [11]. Based on these findings our hypothesis was that plasma cytokine levels early in sep-tic shock could be used as predictive biomarkers to identify responders to corticosteroid treatment. We evaluated patients from the VASST randomized con-trolled trial (RCT) of vasopressin versus norepinephrine [13] in whom we had measured a panel of cytokines, chemokines, and growth factors [14]. Corticosteroid treatment was not randomized, blinded, or protocol-ized in VASST. Therefore, we propensity score matched patients who received corticosteroids to those who did not receive corticosteroids. We then analyzed the effect of corticosteroid treatment and levels of 39 cytokines on 28-day mortality.

MethodsSubjectsVASST was a multicenter RCT of vasopressin versus noradrenaline in septic shock [13]. The ethics boards of participating institutions approved the study. Patients, relatives, or surrogate decision makers gave written con-sent for the participation. Septic shock was defined by the presence of two or more of the systemic inflamma-tory response syndrome (SIRS) criteria, proven or sus-pected infection, new dysfunction of at least one organ, hypotension despite fluid resuscitation and vasopressor support of at least 5 μg/min of norepinephrine (or equiv-alent) for 6 h. Important exclusion criteria were unstable coronary syndromes, acute mesenteric ischemia, and severe chronic heart disease. Treatment with corticos-teroids was not protocolized and both duration and dose were left to the discretion of the treating physicians.

Plasma cytokine measurementBaseline plasma samples were collected within 2 h of the start of infusion of study drug in a subset of centers that included patients in the VASST RCT and were available in 363 patients. Following exclusion of 69 patients on chronic corticosteroid treatment, 294 patients were avail-able for propensity score matching. Median time from arrival at the hospital to study inclusion was 12 h. Plasma was stored at −80 °C and had been exposed to two freeze thaw cycles. A panel of 39 cytokines was measured as described previously [14]. The cytokines were (current

and historical naming) CCL11 (eotaxin), CCL2 (MCP1), CCL22 (MDC), CCL3 (MIP1a), CCL4 (MIP1B), CCL7 (MCP3), CD40LG (CD40 ligand), CSF2 (GMCSF), CSF3 (GCSF), CX3CL1 (fractalkine), CXCL1 (GRO), CXCL10 (IP10), EGF (EGF), FGF2 (FGF2), FLT3LG (Flt3L), IFNA2 (IFNa2), IFNG (IFNG), IL10 (IL10), IL12B (IL12B), IL12P70 (IL12P70), IL13 (IL13), IL15 (IL15), IL17A (IL17), IL1A (IL1a), IL1B (IL1B), IL1RN (IL1RA), IL2 (IL2), IL2RA (IL2RA), IL3 (IL3), IL4 (IL4), IL5 (IL5), IL6 (IL6), IL7 (IL7), IL8 (IL8), IL9 (IL9), LTA (TNF-β), TGFA (TGFa), TNF (TNF-α), and VEGFA (VEGF). For sam-ples in which cytokines were detected but were reported as below or above calibration limits, the lower or upper limit for the respective cytokine was used [14]. Cytokine concentrations were below detection limit in 20 % of the measurements and above detection limit in 0.4 % of the measurements.

Propensity score matchingCorticosteroid- and non-corticosteroid-treated patients were propensity score matched to adjust for differences in baseline variables associated with outcome. The propen-sity score was calculated using linear logistic regression (using the optmatch package in R [15]) with the follow-ing baseline characteristics: age, gender, APACHE  II, diabetes, liver disease, ischemic heart disease, conges-tive heart failure, respiratory disease including chronic obstructive pulmonary disease, chronic renal failure (on dialysis), dose of norepinephrine at baseline, immune suppression (chemotherapy, radiation, leukemia, lym-phoma, HIV), source of infection, pathogen, vasopressin treatment, and surgical history. To retain the maximal sample size while simultaneously balancing baseline vari-ables, no restraints were set on the number of treated or untreated patients within each set. This means that one corticosteroid-treated patient could be matched with one or several untreated patients and vice versa, provided that all were within the same caliper width of the propen-sity score. The number of sets is dependent on (1) caliper width and (2) whether there are corticosteroid-treated and non-corticosteroid-treated patients within the cho-sen caliper width that can be matched. This approach has previously been suggested to reduce bias compared to using fixed numbers of controls or treated patients in each stratum [15, 16]. The caliper width for each matched strata was set at 0.2 of the standard deviation of the logit of propensity score [17]. The weighted standardized dif-ference was used as a balance diagnostic as it is not con-founded by sample size [18]. A standardized difference of ≤10  % is suggested to indicate negligible differences in the mean or prevalence of covariates between groups and caliper width was adjusted downward to 0.0125 to mini-mize the number of covariates above this value [19]. The

1972

propensity score matching was performed using R pack-age “optmatch” version 0.9-3.

Calculation of optimal thresholds for benefit of corticosteroid treatmentThe optimal threshold for each cytokine to predict effect of corticosteroid treatment was identified by an iterative calculation using logistic regression, which identified the cytokine concentration above which corticosteroid treat-ment had the highest odds ratio of 28-day survival.

OutcomesPrimary endpoint was 28-day survival. Statistical differ-ences in survival were calculated with R survival pack-age; differences in categorical survival outcomes were assessed with logistic regression stratified on propensity-matched sets using glm method and parameter fam-ily =  binomial; survival curve differences were assessed with Cox proportional hazard models using the coxph method without propensity matching. Secondary out-comes were days alive and free of mechanical ventilation (DAF vent), days alive and free of vasopressor therapy (DAF pressor), and days alive and free of renal replace-ment therapy (DAF renal) during the first 28  days of hospital stay. Any patient that died during the 28-day observation period was assigned 0 days alive and free of any organ support.

ResultsPatient demographics and propensity score matchingA consort diagram on patient flow is shown in Fig.  1. There were significant differences between the treat-ment groups in APACHE  II score, immunosuppression, and several physiological and laboratory values prior to matching (Table 1). Following propensity score matching, 61 sets of matched patients were created each contain-ing at least one corticosteroid-treated and one non-cor-ticosteroid-treated patient. Sets ranged in size from 2 to 8 patients with a total of 165 patients (100 on corti-costeroids and 65 not on corticosteroids). After match-ing, there was only a significant difference in arterial pH. The median ratio of non-treated to treated subjects was 1:1 (IQR 1:1–1:2) (range 2:1–1:7). Matching reduced weighted standardized differences between treatment groups in baseline variables to ≤10  % for most vari-ables and only four variables (body temperature, arterial pH, new neurologic failure, new hematologic failure) had standardized differences above 10 % after matching (Table 1 and Supplement Fig. 1). Hydrocortisone was the corticosteroid used in 95 % of the patients. The median daily dose of hydrocortisone equivalents in the corti-costeroid-treated group was 250 (IQR 100–305) mg and median length of treatment was 7 (IQR 4–14) days. There

were no differences in 28-day survival between corticos-teroid- and non-corticosteroid-treated groups before or after propensity score matching (Table  1). Demograph-ics of patients that were excluded during the propensity score matching are presented in Supplement Table 1.

Plasma cytokine levels predict benefit of corticosteroid treatmentA total of 13 plasma cytokines predicted a beneficial effect of corticosteroid treatment (28-day survival) as defined by an odds ratio between corticosteroid- and non-corticosteroid-treated patients having a P value less than 0.05. Plasma concentrations with the largest effect size and range of plasma concentrations with a P value less than 0.05 are presented in Table 2. The percentage of patients identified as responders to corticosteroid treat-ment ranged from 21  % for CD40LG to 67  % for IL15. CD40LG was associated with the highest odds ratio of 28-day survival. Potential thresholds and associated odds ratios for CD40LG are shown in Fig. 2.

To investigate whether combinations of plasma cytokine levels could identify a higher proportion of patients as responders to corticosteroids, all combina-tions of pairs or triplets of cytokines with a significant odds ratio were tested using the optimal thresholds iden-tified above. The optimal combination of cytokines was sought by finding the pair or triplet with the highest sig-nificant odds ratio. The cytokine pair with the highest odds ratio was IL3 and IL6, and 71 patients were above threshold for at least one of these cytokines and the odds ratio between corticosteroid- and non-corticosteroid-treated patients was 47 (95 % CI 5.8–653, P = 0.035).

The triplet with the highest odds ratio was IL3, IL6, and CCL4, and 82 patients were above the threshold for a least one of these cytokines and the odds ratio was 19 (95  % CI 3.5–140, P =  0.02). Figure  3 shows a Kaplan–Meyer curve for the propensity score matched corticos-teroid-treated and non-corticosteroid-treated patients above any of the thresholds for the corticosteroid pre-dictive cytokine triplet IL-3, IL6, and CCL4 and for cor-ticosteroid- and non-corticosteroid-treated patients below all of the thresholds. Cox proportional hazard analysis supports that these thresholds predict a ben-eficial response to corticosteroid treatment in patients above any of the thresholds (P =  0.01 for effect of cor-ticosteroids above triplet). Time to death in patients above triplet threshold that died within the first 28 days in the present study was 10 (IQR 3–23, n =  13) and 6 (IQR 5–10, n =  14) in corticosteroid-treated and non-corticosteroid-treated patients, respectively (P = 0.37). In patients with cytokine triplet below all of the thresholds corticosteroids are harmful [P  <  0.01, OR survival 0.02 (95 % CI 0.001–0.13)]. APACHE II score and preexisting

1973

!

Assessed for eligibility in VASST study (n= 6229 )

Excluded (n= 5427)Met exclusion criteria (n= 3758)Other reasons (n= 1669)

Included in VASST analysis (n= 779)

Randomized (n= 802)

Did not receive infusion (n= 21)Lost to follow up (n = 1)Withdrew consent (n = 2)

Patients with plasma samples available at baseline (n = 363)

Included in propensity match

(n = 294)

Plasma not available for analysis (n = 416)

Patients on chronic steroids (n = 69)

Non matched patients (n = 129)

Included in analysis (n = 165)

Fig. 1 Consort scheme of patients included in the study

1974

Table 1 Patient demographics before and after propensity matching

No steroids, (n = 86)

Steroids, (n = 208)

Standardized difference

P Propensity-matched sets Weighted standardized difference

P

No steroids(61 sets)

Steroids (61 sets)

Age 64 (49–73) 62 (50–72) −0.01 0.73 66 (52–74) 62 (52–70) −0.05 0.74

Male sex % (SD) 40 (49) 40 (49) 0.01 1 36 (48) 36 (49) 0.00 1

APACHE II score (median, IQR)

24 (19–30) 26 (22–32) 0.34 0.008* 24 (20–30) 25 (21–28) −0.09 0.53

Pre-existing conditions % or % (standard deviation)

Ischemic heart disease

21 14 0.18 0.16 24 (42) 19 (36) −0.03 0.8

Congestive heart failure

9 6 0.12 0.31 22 (30) 12 (31) −0.06 0.74

Chronic obstructive pulmonary disease

19 12 0.19 0.13 22 (41) 22 (39) 0.01 0.96

Chronic renal failure

8 11 −0.08 0.67 10 (30) 10 (28) −0.07 0.61

Diabetes 23 25 −0.04 0.75 22 (41) 23 (38) 0.02 0.9

Immunosup-pression

2 9 −0.27 0.047* 3 (18) 3 (14) 0.01 0.95

Malignancy 17 15 0.07 0.48 16 (37) 12 (29) −0.10 1

Recent surgical history no. (%)

29 22 −0.17 0.18 28 (44) 28 (41) −0.01 0.94

New organ failures % or % (standard deviation)

Respiratory 6 4 0.10 0.53 6 (23) 6 (24) −0.03 0.53

Renal 27 32 −0.11 0.49 60 (49) 57 (45) 0.00 1

Hematologic and coagula-tion

15 21 0.15 0.26 14 (34) 19 (35) −0.13 0.43

Neurologic 22 23 −0.01 1 25 (43) 14 (30) 0.24 0.16

Source of infection % or % (standard deviation)

Lung 42 42 0.00 1 38 (48) 38 (44) 0.02 0.91

Abdomen 28 26 −0.03 0.89 28 (45) 29 (42) 0.04 0.83

Other 30 32 0.03 0.89 34 (47) 33 (42) 0.05 0.73

Pathogen

Gram (+) 57 48 −0.19 0.16 54 (49) 60 (45) −0.08 0.58

Gram (−) 37 28 −0.20 0.13 37 (48) 36 (44) −0.01 0.93

Physiological and laboratory variables at baseline [median, IQR or % (standard deviation)]

MAP (mmHg) 73 (68–77) 72 (66–77) −0.15 0.27 72 (69–77) 74 (69–79) −0.03 0.6

Arterial pH 7.36 (7.3–7.4) 7.31 (7.24–7.37) 0.10 <0.001* 7.36 (7.32–7.4) 7.33 (7.29–7.37) 0.20 0.005

Lactate (mmol/L)

1.8 (1.3–3.9) 2.2 (1.4–4.4) 0.20 0.3 1.85 (1.3–4.1) 1.9 (1.5–3.4) −0.04 1

Norepinephrine (μg/min)

8 (5–14) 16 (8–26) 0.67 <0.001* 10 (7–14) 12 (8–18) 0.07 0.65

Vasopressin (yes/no)

51 % 51 % 0.01 0.97 55 (50) 58 (45) 0.05 0.78

PaO2 (mmHg) 92 (79–110) 90 (77–107) 0.05 0.63 92 (80–108) 92 (80–108) 0.02 0.94

FiO2 (%) 0.45 (0.4–0.55) 0.5 (0.4–0.61) 0.30 0.036* 0.4 (0.4–0.5) 0.5 (0.4–0.6) 0.04 0.63

WBC (109 cells/L) 16 (9–23) 13 (7–21) −0.04 0.052 15 (10–21) 14 (10–18) −0.13 0.39

Temperature (°C)

37.9 (37–38.6) 37.4 (37–38.2) −0.19 0.025* 37.8 (37.1–38.5) 37.5 (37.0–37.9) −0.38 0.03

Outcomes [% or median (IQR) or % (standard deviation)]

Survival 28-day 77 70 0.15 0.32 73 (44) 75 (37) −0.68 0.63

1975

conditions were similar in patients above and below the threshold, whereas physiological and laboratory vari-ables were generally more abnormal in patients above the novel corticosteroid predictive cytokine triplet threshold (Supplement Table 2). We also tested our triplet cytokine threshold in the whole cohort of patients that were not on chronic corticosteroid treatment (n = 294, see Fig. 1) and found that it predicted response to corticosteroids in this cohort, too (P =  0.03 for effects of corticoster-oids above threshold). To evaluate the importance of the inclusion of cytokine data above and below the calibra-tion limits (see “Methods”) we redid the propensity score matching including only patients with all three triplet cytokines within calibration limits. Following propen-sity matching, 117 patients in 38 matched sets remained in the analysis. Using the triple cytokine thresholds reported in Table 2 on this subset of patients, we found

that the beneficial effect of corticosteroids was no longer significant but directionally the same [OR 3.5 (95  % CI 0.5–26.7, P = 0.20)].

Is the beneficial effect of steroids only seen in vasopressin-treated patients?A synergistic interaction between vasopressin and cor-ticosteroids in septic shock has been suggested [20]. We made an attempt to test the effect of corticosteroids in patients above the cytokine triplet threshold in patients not receiving vasopressin. To maintain statistical power we redid the propensity score matching without caliper restrictions. Following propensity score matching a total of 30 sets containing a total of 120 patients were left for analysis. In this group of patients not receiving vasopres-sin no effect of corticosteroids on 28-day survival could be detected [OR 0.4 (95 % CI 0.02–7, P = 0.68)]. A similar

Table 2 Plasma concentrations of cytokines predicting a beneficial effect of steroid treatment on 28-day survival in pro-pensity-matched patients

Only cytokines with an odds ratio with a P value below 0.05 are reported. Threshold is defined as concentrations ≥ concentration with highest odds ratio (OR)

Cytokine OR for survival with  treatment (95 % CI)

Concentration at  highest OR (pM)

P value Range of conc. with effect on OR (P < 0.05) (pM)

Patients above threshold, n (%)

CD40LG 89 (4–4930) 22 0.010 16–31 35 (21)

IL3 38 (3–1600) 0.09 0.017 0.08–0.10 47 (28)

IL6 18 (2–251) 16 0.018 6–17 49 (30)

CCL4 15 (2–190) 4 0.019 4–6 50 (30)

CSF3 7 (2–38) 4 0.017 3–4 104 (63)

IL12P70 6 (1–37) 0.09 0.027 0.09–0.10 71 (43)

TGFA 6 (1–39) 0.06 0.030 0.05–0.06 81 (49)

LTA 6 (1–38) 0.03 0.033 0.03 78 (47)

IL15 5 (1–23) 0.06 0.021 0.01–0.08 110 (67)

EGF 5 (1–23) 0.07 0.038 0.07 88 (53)

IL7 4 (1–22) 0.25 0.044 0.2–0.3 94 (57)

IL9 4 (1–18) 0.01 0.036 0.01 107 (65)

IL4 4 (1–16) 0.07 0.039 0.07–0.08 109 (66)

Median values for the propensity-matched cohort were calculated by calculating the median of the matched sets of patients weighted for number of patients in each set. Groups or sets were compared using Fisher or Wilcoxon tests as appropriate.

DAF days alive and free (0 if patient died before day 28), WBC white blood cell count, PaO2 arterial partial pressure of oxygen, FiO2 fraction of inspired oxygen, SD standardized differences, IQR interquartile range.

* P < 0.05

Table 1 continued

No steroids, (n = 86)

Steroids, (n = 208)

Standardized difference

P Propensity-matched sets Weighted standardized difference

P

No steroids(61 sets)

Steroids (61 sets)

DAF ventilator 16 (2–22) 8 (0–21) −0.23 0.1 15 (0–22) 11 (1–17) −0.12 0.41

DAF renal sup-port

28 (8–28) 23 (0–28) −0.27 0.01 28 (0–28) 20 (12–28) −0.01 0.42

DAF pressor 22 (1.25–24) 21 (0–24) −0.24 0.18 21 (0–24) 14 (1–23) −0.55 0.37

1976

analysis of propensity score matched sets including only patients treated with vasopressin above triplet showed beneficial effect of corticosteroids on 28-day survival [OR 14 [95 % CI 2–162, P = 0.016)].

Effects of corticosteroids on secondary outcomes and cytokine levels at 24 hFor patients who exceeded the triplet threshold the dif-ferences in organ support requirements trended in the same direction as mortality but did not achieve statisti-cal significance. Median DAF of ventilator support was 5 (IQR 0–18)  days in non-corticosteroid-treated patients and 14 (IQR 0–23) in corticosteroid-treated patients (P = 0.19, Wilcoxon test). Median DAF of renal support therapy was 0 (IQR 0–7) and 2.5 (IQR 0–20) in non-cor-ticosteroid- and corticosteroid-treated patients, respec-tively (P  =  0.32). Median DAF of vasopressor support

was 20 (IQR 0–23) and 23 (IQR 0–25) in non-corticos-teroid- and corticosteroid-treated patients, respectively (P  =  0.15). Organ support data for the cytokines that could predict a beneficial effect of steroids are presented in Supplement Tables 3 and 4.

To test the hypothesis that corticosteroid treatment exerts its beneficial effect through modulation of the cytokine response we compared the change in cytokine concentrations from baseline to 24 h in the patients who were above the cytokine triplet threshold and treated with corticosteroids within 24  h to non-corticosteroid-treated patients above the cytokine triplet threshold. Plasma concentrations of cytokines generally decreased from baseline to 24  h irrespective of treatment, and CCL11, CXL10, IL2, IL1RN, and CCL22 decreased more in the corticosteroid-treated than in the non-corticoster-oid-treated patients (Supplement Table 5).

Fig. 2 Odds ratios (ORs, circles) for 28-day survival by corticosteroid treatment as a function of measured plasma concentrations of CD40LG. Right y-axis shows fraction of patients available (black squares) for calculation of ORs. The range of concentrations associated with a significant effect on the odds ratio (P < 0.05) is indicated by the blue vertical dotted lines. Note that the left x-axis is in two sections with different scaling on the two sec-tions

1977

DiscussionBy analyzing treatment effects according to plasma con-centrations of 39 cytokines, we found 13 cytokines that identified a subset of patients with increased 28-day sur-vival from corticosteroid treatment. We further found a cytokine triplet (IL3, IL6, and CCL4) that identified patients with a beneficial response to corticosteroid treatment. Finally, in these patients corticosteroid treat-ment induced a more pronounced decrease in some of the measured cytokines during the first 24 h.

The cytokine pair of IL3 and IL6 identified a higher number of corticosteroid-responsive patients than either of these cytokines alone, and by adding CCL4 another 11 responders could be identified. These results show that the sensitivity of plasma cytokine levels, as biomarkers of response to corticosteroid treatment, can be increased while maintaining a high treatment effect. IL-6 is known to be important for both innate and adaptive immunity and has both pro- and anti-inflammatory properties [21]. IL3 promotes myelopoiesis and was recently sug-gested to be essential for amplification of the inflamma-tory response in sepsis [22], and CCL4 is important for recruitment of regulatory T  cells in inflammation [23]. All three cytokines have been suggested to predict poor outcome in sepsis [12, 22, 24, 25].

Our definition of optimal threshold for respective cytokine as the concentration associated with the highest odds ratio could be questioned as being too conservative and that sensitivity could be increased by using the low-est concentration at which a significant beneficial effect of corticosteroids on the odds ratio could be detected (illustrated by the leftmost of the two blue vertical lines in Fig.  2). We chose the former definition because this definition is less influenced by sample size. It could also be argued that by choosing IL15 as a single cytokine an even higher number of responders could be identified albeit with a lower treatment effect (Table 2).

Two recent studies applying propensity score matching on large observational cohorts could not demonstrate a beneficial effect on mortality by corticosteroid treatment [26, 27]. Our results in the matched patients that corti-costeroids did not improve 28-day survival agree with these results and support the notion that only selected patients may benefit from corticosteroid treatment. One of the studies mentioned above reported a reduced 28-day mortality by corticosteroid treatment in patients with a high severity of illness defined by an APACHE II score >30 [26]. In the present study there was no differ-ence in APACHE  II between patients above and below the cytokine triplet threshold but some physiological

Fig. 3 Kaplan–Meyer curves for propensity score matched patients treated with corticosteroids (blue) or not treated with corticosteroids (red) above one or more of the threshold concentration for CCL4, IL3, and IL6, respectively (solid lines) or below all of the thresholds (dotted lines). Cor-ticosteroid treatment increased 28-day survival in patients with CCL4, IL3, or IL6 levels above the threshold (unadjusted Cox proportional hazard analysis including all 4 groups, P < 0.01) and decreased survival below the threshold (P < 0.05)

1978

parameters were more deranged in patients above the cytokine triplet threshold supporting that severity of ill-ness may predict response to corticosteroid treatment (Supplement Table 2). Based on these results it could be asked whether plasma cytokine concentrations merely reflect severity of illness or whether they add independ-ent predictive value to clinical parameters. Differences in physiological parameters in the present study were small with considerable overlap between treatment groups. This suggests that in a clinical setting it would be diffi-cult to identify corticosteroid-responsive patients using severity of illness and that cytokine levels add predictive value.

Our finding that hydrocortisone reduced plasma cytokine levels at 24  h align with the hypothesis that the benefi-cial effect of corticosteroids on mortality may be second-ary to modulation of an excessive inflammatory response [28, 29]. This is in agreement with our previous observa-tion that survivors of septic shock had a greater decrease in plasma cytokines than did non-survivors [14]. However, we acknowledge that a limitation of the present study is that our results do not allow us to make any firm conclusions with regard to mechanisms mediating the beneficial effect of corticosteroids. Also, by including data above and below the calibration limits it is possible that the true change in cytokine concentrations may have been underestimated. It should be noted that some cytokines tended to decrease less from baseline to day 1 in corticosteroid-treated patients than in non-corticosteroid-treated patients. Whether this trend reflects a true biological effect of corticosteroids, which did not reach statistical significance owing to the small sample size, or if it is a chance finding is uncertain.

It could be argued that the exclusion of 129 patients of which the majority were corticosteroid-treated (Supple-ment Table 1) could have biased our results. However, our result that the triplet of plasma cytokines predicted a benefi-cial response to corticosteroid treatment also when applied to the whole cohort of patients treated (Table  1) supports the robustness of the selected cytokine triplet as a predic-tor of response to corticosteroids. It should be noted that our results also suggest that corticosteroids may be harmful in patients below the triplet threshold. This result further supports the concept that use of corticosteroids should be reserved for specific subpopulations of sepsis patients. Our result that we could not demonstrate a beneficial effect of corticosteroids in patients not treated with vasopressin could reflect that a beneficial effect of corticosteroids in patients with high cytokines is dependent on co-administration of vasopressin. Alternatively, it could be related to the low power of this subgroup analysis and/or to poor matching.

In the main analysis we included all cytokine meas-urements, including those above and below the calibra-tions limits, because such an approach mimics a clinical

situation in which some patients will present with cytokine values outside the calibrations limits. The observation that corticosteroids lacked a significant effect in the subgroup analysis in which only patients with the triplet cytokines within the calibration limits were included may have sev-eral explanations. Firstly, only patients above the threshold are used to calculate the odds ratio. By excluding patients with the highest levels of cytokines (above calibration limits) it is possible that patients in which corticosteroids are most beneficial were excluded. Secondly, power was decreased owing to a decrease in number of matched sets.

The next step for this project is to investigate the derived threshold values of the different cytokines in a large validation cohort. If the data are confirmed, then we would propose that the cytokine triplet may be used in a prospective RCT for enrichment, i.e., to identify a subset of sepsis patients that are predicted to be responsive to corticosteroids and that should be used as inclusion crite-ria to investigate the efficacy of corticosteroid treatment.

Limitations and strengthsA limitation of our study was that the use of corticoster-oids was not randomized, blinded, or protocolized. In addition, we cannot exclude that in some cases it is pos-sible that corticosteroid treatment was initiated before baseline plasma samples were collected. The relatively small sample size also suggests that our results should be interpreted cautiously.

Strengths of our study were that we used a well-phe-notyped cohort of patients with septic shock, that we carefully matched corticosteroid-treated to non-corti-costeroid-treated patients, that the use of corticosteroids was according to individual physician choice (i.e., this is an efficiency/real-world trial), and that we assessed a large battery of 39 cytokines, chemokines, growth fac-tors, and other markers of inflammation.

ConclusionsPlasma cytokine levels can be used to identify patients that benefit from corticosteroid treatment. The cytokine triplet of IL3, IL6, and CCL4 identified a large propor-tion of the patients in whom corticosteroid treatment increased 28-day survival. We found biological plausibil-ity for this finding in that corticosteroid-treated patients had a more rapid decline in plasma cytokines compared to non-corticosteroid-treated patients above the cytokine triplet threshold. These findings align with the notion that corticosteroids may be beneficial in selected patients who have septic shock and elevated plasma cytokine levels through beneficial modulation of the immune response.

Electronic supplementary materialThe online version of this article (doi:10.1007/s00134-016-4338-z) contains supplementary material, which is available to authorized users.

1979

Author details1 Centre for Heart Lung Innovation (HLI), University of British Columbia, Vancouver, Canada. 2 Division of Critical Care Medicine, St. Paul’s Hospital, University of British Columbia, 1081 Burrard Street, Vancouver, BC V6Z 1Y6, Canada. 3 Department of Anesthesia and Intensive Care, Helsingborg Hospital, Helsingborg and Lund University, Lund, Sweden. 4 Department of Anesthesiol-ogy and Intensive Care, Helsingborg Hospital, 251 87 Helsingborg, Sweden.

AcknowledgmentsThe VASST study was supported by Canadian Institutes of Health Research, Grant number: MCT 44152 Registration: ISRCTN94845869. PB was supported by grants from Region Skåne (ALF # 18401) and the Anna and Edwin Berger Foundation. JB is a recipient of a Providence Health Care Research Scholarship. JR was supported by Canadian Institutes of Health Research (CIHR) (SONRIS).

Compliance with ethical standards

Conflicts of interest PB, CF, KRW, and JB report no conflicts of interest. JR is a founder, director, and shareholder in Cyon Therapeutics Inc. (developing a sepsis therapy). JR has share options in Leading Biosciences Inc. JR reports receiving consulting fees from Cubist Pharmaceuticals (now owned by Merck; formerly was Trius Phar-maceuticals; developing antibiotics), Ferring Pharmaceuticals (manufactures vasopressin and is developing selepressin), Grifols (sells albumin), MedImmune (regarding sepsis), Leading Biosciences (developing a sepsis therapeutic), La Jolla Pharmaceuticals (developing angiotensin II), CytoVale Inc. (developing a sepsis diagnostic), and Asahi Kesai (developing recombinant thrombomodulin).

Received: 19 November 2015 Accepted: 19 March 2016 Published online: 12 April 2016

References 1. Angus DC, van der Poll T (2013) Severe sepsis and septic shock. N Engl J

Med 29(369):840–851 2. Annane D, Sébille V, Charpentier C, Bollaert PE, François B, Korach JM,

Capellier G, Cohen Y, Azoulay E, Troché G, Chaumet-Riffaud P, Bellissant E (2002) Effect of treatment with low doses of hydrocortisone and fludro-cortisone on mortality in patients with septic shock. JAMA 288:862–871

3. Annane D, Bellissant E, Bollaert PE, Briegel J, Confalonieri M, De Gaudio R, Keh D, Kupfer Y, Oppert M, Meduri GU (2009) Corticosteroids in the treatment of severe sepsis and septic shock in adults: a systematic review. JAMA 301:2362–2375

4. Bone RC, Fisher CJ Jr, Clemmer TP, Slotman GJ, Metz CA, Balk RA (1987) A controlled clinical trial of high-dose methylprednisolone in the treatment of severe sepsis and septic shock. N Engl J Med 317:653–658

5. Schumer W (1976) Steroids in the treatment of clinical septic shock. Ann Surg 184:333–341

6. Sprung CL, Annane D, Keh D, Moreno R, Singer M, Freivogel K, Weiss YG, Benbenishty J, Kalenka A, Forst H, Laterre PF, Reinhart K, Cuthbertson BH, Payen D, Briegel J, CORTICUS Study Group (2008) Hydrocortisone therapy for patients with septic shock. N Engl J Med 358:111–124

7. Annane D, Sébille V, Troché G, Raphaël JC, Gajdos P, Bellissant E (2000) A 3-level prognostic classification in septic shock based on cortisol levels and cortisol response to corticotropin. JAMA 283:1038–1045

8. Boonen E, Vervenne H, Meersseman P, Andrew R, Mortier L, Declercq PE, Vanwijngaerden YM, Spriet I, Wouters PJ, Vander Perre S, Langouche L, Vanhorebeek I, Walker BR, Van den Berghe G (2013) Reduced cortisol metabolism during critical illness. N Engl J Med 368:1477–1488

9. Dellinger RP, Levy MM, Rhodes A, Annane D, Gerlach H, Opal SM, Sevransky JE, Sprung CL, Douglas IS, Jaeschke R, Osborn TM, Nunnally ME, Townsend SR, Reinhart K, Kleinpell RM, Angus DC, Deutschman CS, Machado FR, Rubenfeld GD, Webb S, Beale RJ, Vincent JL, Moreno R, Surviving Sepsis Campaign Guidelines Committee including The Pediatric Subgroup (2013) Surviving Sepsis Campaign: international guidelines for management of severe sepsis and septic shock, 2012. Intensive Care Med 39:165–228

10. Bozza FA, Salluh JI, Japiassu AM, Soares M, Assis EF, Gomes RN, Bozza MT, Castro-Faria-Neto HC, Bozza PT (2007) Cytokine profiles as markers of disease severity in sepsis: a multiplex analysis. Crit Care 11:R49

11. Fjell CD, Thair S, Hsu JL, Walley KR, Russell JA, Boyd J (2013) Cytokines and signaling molecules predict clinical outcomes in sepsis. PLoS One. doi:10.1371/journal.pone.0079207

12. Pinsky MR, Vincent JL, Deviere J, Alegre M, Kahn RJ, Dupont E (1993) Serum cytokine levels in human septic shock: relation to multiple system organ failure and mortality. Chest 103:565–575

13. Russell JA, Walley KR, Singer J, Gordon AC, Hébert PC, Cooper DJ, Holmes CL, Mehta S, Granton JT, Storms MM, Cook DJ, Presneill JJ, Ayers D, VASST Investigators (2008) Vasopressin versus norepinephrine infusion in patients with septic shock. N Engl J Med 358:877–887

14. Russell JA, Fjell C, Hsu JL, Lee T, Boyd J, Thair S, Singer J, Patterson AJ, Wal-ley KR (2013) Vasopressin compared with norepinephrine augments the decline of plasma cytokine levels in septic shock. Am J Respir Crit Care Med 188:356–364

15. Hansen BB (2004) Full matching in an observational study of coaching for the SAT. J Am Stat Assoc 99:609–618

16. Cepeda MS, Boston R, Farrar JT, Strom BL (2003) Optimal matching with a variable number of controls vs. a fixed number of controls for a cohort study trade-offs. J Clin Epidemiol 56:230–237

17. Austin PC (2011) Optimal caliper widths for propensity-score matching when estimating differences in means and differences in proportions in observational studies. Pharm Stat 10:150–161

18. Austin PC (2008) Assessing balance in measured baseline covariates when using many-to-one matching on the propensity-score. Pharma-coepidemiol Drug Saf 17:1218–1225

19. Austin PC (2011) An introduction to propensity score methods for reduc-ing the effects of confounding in observational studies. Multivariate Behav Res 46:399–424

20. Russell JA, Walley KR, Gordon AC, Cooper DJ, Hébert PC, Singer J, Holmes CL, Mehta S, Granton JT, Storms MM, Cook DJ, Presneill JJ, Dieter Ayers for the Vasopressin and Septic Shock Trial Investigators (2009) Interaction of vasopressin infusion, corticosteroid treatment, and mortality of septic shock. Crit Care Med 37:811–818

21. Hunter CA, Jones SA (2015) IL-6 as a keystone cytokine in health and disease. Nat Immunol 16:448–457

22. Weber GF, Chousterman BG, He S, Fenn AM, Nairz M, Anzai A, Brenner T, Uhle F, Iwamoto Y, Robbins CS, Noiret L, Maier SL, Zönnchen T, Rahbari NN, Schölch S, Klotzsche-von Ameln A, Chavakis T, Weitz J, Hofer S, Weigand MA, Nahrendorf M, Weissleder R, Swirski FK (2015) Interleukin-3 amplifies acute inflammation and is a potential therapeutic target in sepsis. Science 347:1260–1265

23. Bystry RS, Aluvihare V, Welch KA, Kallikourdis M, Betz AG (2001) B cells and professional APCs recruit regulatory T cells via CCL4. Nat Immunol 2:1126–1132

24. Nowak JE, Wheeler DS, Harmon KK, Wong HR (2010) Admission chemokine (C-C motif ) ligand 4 levels predict survival in pediatric septic shock. Pediatr Crit Care Med 11:213–216

25. Reinhart K, Menges T, Gardlund B, Harm Zwaveling J, Smithes M, Vincent JL, Tellado JM, Salgado-Remigio A, Zimlichman R, Withington S, Tschai-kowsky K, Brase R, Damas P, Kupper H, Kempeni J, Eiselstein J, Kaul M (2001) Randomized, placebo-controlled trial of the anti-tumor necrosis factor antibody fragment afelimomab in hyperinflammatory response during severe sepsis: the RAMSES Study. Crit Care Med 29:765–769

26. Funk D, Doucette S, Pisipati A, Dodek P, Marshall JC, Kumar A, Cooperative Antimicrobial Therapy of Septic Shock Database Research Group (2014) Low-dose corticosteroid treatment in septic shock: a propensity-match-ing study. Crit Care Med 42:2333–2341

27. Póvoa P, Salluh JI, Martinez ML, Guillamat-Prats R, Gallup D, Al-Khalidi HR, Thompson BT, Ranieri VM, Artigas A (2015) Clinical impact of stress dose steroids in patients with septic shock: insights from the PROWESS-Shock trial. Crit Care 28(19):193

28. Keh D, Boehnke T, Weber-Cartens S, Schulz C, Ahlers O, Bercker S, Volk HD, Doecke WD, Falke KJ, Gerlach H (2003) Immunologic and hemodynamic effects of “low-dose” hydrocortisone in septic shock: a double-blind, randomized, placebo-controlled, crossover study. Am J Respir Crit Care Med 167:512–520

29. Oppert M, Schindler R, Husung C, Offermann K, Gräf KJ, Boenisch O, Barckow D, Frei U, Eckardt KU (2005) Low-dose hydrocortisone improves shock reversal and reduces cytokine levels in early hyperdynamic septic shock. Crit Care Med 33:2457–2464