Rueda et al. Ann. Intensive Care (2019) 9:119 https://doi.org/10.1186/s13613-019-0593-9 RESEARCH Open Access Growth differentiation factor 15 and early prognosis after out‑of‑hospital cardiac arrest Ferran Rueda1,2,3, Germán Cediel1,2, Cosme García‑García1,2, Júlia Aranyó1,2, Marta González‑Lopera1,2, M. Cruz Aranda Nevado1,2, Judith Serra Gregori1,2, Teresa Oliveras1,2, Carlos Labata1,2, Marc Ferrer1,2, Nabil El Ouaddi1,2 and Antoni Bayés‑Genís1,2* Abstract Background: Growth differentiation factor 15 (GDF-15) is an inflammatory cytokine released in response to tissue injury. It has prognostic value in cardiovascular diseases and other acute and chronic conditions. Here, we explored the value of GDF-15 as an early predictor of neurologic outcome after an out-of-hospital cardiac arrest (OHCA). Methods: Prospective registry study of patients in coma after an OHCA, admitted in the intensive cardiac care unit from a single university center. Serum levels of GDF-15 were measured on admission. Neurologic status was evalu‑ ated according to the cerebral performance category (CPC) scale. The relationship between GDF-15 levels and poor neurologic outcome at 6 months was analyzed. Results: Among 62 patients included, 32 (51.6%) presented poor outcome (CPC 3–5). Patients with CPC 3–5 exhibited significantly higher GDF-15 levels (median, 17.1 [IQR, 11.1–20.4] ng/mL) compared to those with CPC 1–2 (7.6 [IQR, 4.1–13.1] ng/mL; p = 0.004). Multivariable logistic regression analyses showed that age (OR, 1.09; 95% CI 1.01–1.17; p = 0.020), home setting arrest (OR, 8.07; 95% CI 1.61–40.42; p = 0.011), no bystander cardiopulmonary resuscitation (OR, 7.91; 95% CI 1.84–34.01; p = 0.005), and GDF-15 levels (OR, 3.74; 95% CI 1.32–10.60; p = 0.013) were independent predictors of poor outcome. The addition of GDF-15 in a dichotomous manner (≥ 10.8 vs. < 10.8 ng/mL) to the resulting clinical model improved discrimination; it increased the area under the curve from 0.867 to 0.917, and the associated continuous net reclassification improvement was 0.90 (95% CI 0.48–1.44), which allowed reclassifica‑ tion of 37.1% of patients. Conclusions: After an OHCA, increased GDF-15 levels were an independent, early predictor of poor neurologic out‑ come. Furthermore, when added to the most common clinical factors, GDF-15 improved discrimination and allowed patient reclassification. Keywords: Out-of-hospital cardiac arrest, Prognostication, Neurologic outcome, Growth differentiation factor 15, Biomarkers Background leads to two-thirds of the mortality and persistent neu- The prognosis remains poor for comatose survivors of an rological disability in survivors [1–3]. Thus, early neuro- out-of-hospital cardiac arrest (OHCA) that are admitted logic prognostication could be useful for avoiding futile to intensive care units, despite current aggressive care; treatments and providing information to relatives. survival rates are around 50%. Outcome is mostly related Unfortunately, among the available predictors of poor to the degree of hypoxic-ischemic brain injury, which outcome, including clinical examination, electrophysi- ological tests, neuroinjury biomarkers, and neuroim- aging, no single factor can predict poor outcome with *Correspondence: abayesgenis@gmail.com certainty, particularly within the first 72 h after resuscita- 1 Heart Institute, Germans Trias i Pujol University Hospital, Carretera de tion. Moreover, some predictors can be affected by seda- Canyet s/n, Badalona, 08916 Barcelona, Spain Full list of author information is available at the end of the article tives and therapeutic hypothermia. Therefore, accurate © The Author(s) 2019. This article is distributed under the terms of the Creative Commons Attribution 4.0 International License (http://creat​iveco​mmons​.org/licen​ses/by/4.0/), which permits unrestricted use, distribution, and reproduction in any medium, provided you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made. Rueda et al. Ann. Intensive Care (2019) 9:119 Page 2 of 10 prognostication requires a multimodal approach that The protocol was approved by the institutional Ethics lasts several days in the majority of cases [4]. Committee; all patients or their representatives provided Although biomarkers represent a potential tool, they written informed consent. are still underused, especially because most current neu- roinjury biomarkers lack consistent thresholds for iden- Neurologic assessment and withdrawal of care tifying patients with no chance of survival. Moreover, Sedation was discontinued at normothermia [24 h after further evidence is needed about the optimal time points return of spontaneous circulation (ROSC) in patients for measuring them. not treated with hypothermia]. In patients remaining In the last decade, growth differentiation factor 15 comatose, a multimodal neurological evaluation was (GDF-15), an inflammatory cytokine released in response undertaken by a consultant neurologist a minimum of to tissue injury, has emerged as a biomarker with prog- 72 h after normothermia (72 h after ROSC in patients nostic value in cardiovascular diseases and other acute not treated with hypothermia), according to current rec- and chronic conditions. Moreover, GDF-15 provides ommendations [4]. This evaluation was based on clini- information on the severity of disease [5]. In addition, we cal findings together with electroencephalogram (EEG) recently reported that admission levels of GDF-15 could and somatosensory evoked potentials (SSEP; only avail- serve as a robust, independent predictor of mortality in able from 2015). In addition, results of brain computed primary ventricular fibrillation, due to ST-segment eleva- tomography (CT) and determinations of neuron-specific tion myocardial infarction (STEMI) [6], the most com- enolase (NSE; routinely measured from 2015), were con- mon cause of OHCA. sidered. In patients treated with hypothermia, additional In view of this background, in the present study, we EEG was performed within the first 24 h. Brain death aimed to explore the value of GDF-15 on admission as was diagnosed according to Spanish legislation. GDF-15 early predictor of neurologic outcome in comatose sur- results were not available for the treating physicians and vivors of OHCA. did not influence this process. Full intensive care was provided until prognostic evalu- Methods ation was completed. After a statement of “poor neuro- Study population logical prognosis”, decision on level of care was discussed We prospectively enrolled patients older than 18 years with the patient’s family. Findings allowing withdrawal of who were admitted in a persistent coma after successful life-sustaining therapies comprised: (1) brain death due resuscitation from an OHCA, to the intensive cardiac to cerebral herniation; (2) persisting coma with a Glas- care unit from a single university center from April 2011 gow motor scale (GMS) of 1–2, together with bilateral to May 2016. The diagnosis of cardiac arrest was estab- absence of N20-peak on SSEP, absence of pupillary and lished when patients exhibited an absence of spontane- corneal reflexes, or presence of malignant EEG patterns ous respiration, no palpable pulse, and no responsiveness (absence of reactivity, burst-suppression or refractory to stimuli. Successful resuscitation was defined as recov- status epilepticus); (3) persisting coma (GMS 1–2) for ery of blood pressure and pulse for more than 1 h, with or more than 7 days after ROSC in absence of confounders, without continuous catecholamine infusion. Coma was especially if elevated NSE levels (> 60 ng/mL at 48–72 h) defined as a Glasgow Coma Scale score < 9. Patients were or generalized ischemic changes on CT were present; (4) excluded when the arrest arose from a non-cardiac origin ethical reasons. (i.e., traumatic, toxic, or neurologic cause). Pre-hospital data were recorded by emergency phy- GDF‑15 samples sicians, according to the Utstein style. Baseline demo- Blood samples were obtained on admission (baseline) graphics and clinical data were prospectively recorded. and processed for central laboratory estimations of GDF- Treatment to restore spontaneous circulation and post- 15. Serum was isolated by centrifugation and stored at resuscitation care were provided according to interna- − 80 °C until assayed. Patients without baseline levels of tional guidelines and recommendations [7, 8]. Patients GDF-15 available were excluded from the study. Addi- were treated with hypothermia at 33 °C according to phy- tional samples were drawn at other two pre-specified sician’s judgement and a local protocol, which excluded time points (12 h and 24 h after OHCA) for exploratory those with cardiopulmonary resuscitation (CPR) time analyses. longer than 45 min, non-shockable rhythms, refractory shock, refractory ventricular arrhythmias, severe coagu- lopathy or terminal disease. Rueda et al. Ann. Intensive Care (2019) 9:119 Page 3 of 10 GDF‑15 assay calibration, and net reclassification by adding baseline Serum GDF-15 concentrations were determined with GDF-15 to a clinical model, in a dichotomous manner, (ECLIA; ­Elecsys® GDF-15 assay, Roche Diagnostics, Pen- a fully automated electrochemiluminescence assay according to the best cutoff value derived from a ROC analyses. We performed ROC analyses and the Hosmer– zberg, Germany) on the Cobas Analytics e601 analyzer Lemeshow test to obtain the net improvement in risk cat- (Roche Diagnostics). The analytic performance of this egory reclassification (NRI). Differences were considered assay had been validated, and it correlated closely with a statistically significant at p < 0.05. All analyses were per- previously established immunoradiometric assay method formed with STATA V.13.0 (StataCorp, College Station, [9]. The Elecsys GDF-15 assay had a measuring range TX). of 0.4 to 20 ng/mL, as stated by the manufacturer. Sam- ples with values above the measuring range were diluted Results accordingly. The upper limit of the reference interval in Study cohort characteristics healthy older individuals has been proposed to be 1.2 ng/ A total of 62 patients were included in the study. The mL. flow chart depicting the population included is shown in Additional file 1: Figure S1. Comparison of baseline Clinical endpoint characteristics between included and excluded patients The primary endpoint of the study was poor neurologi- are detailed in Additional file 2: Table S1. The median cal outcome at 6 months, evaluated by the five-graded age of participants was 59 (52–71) years, and 17.7% were Cerebral Performance Category (CPC) scale [10]. A CPC women. Baseline characteristics of the study subjects, score of 1 (good cerebral performance) or 2 (moderate according to the presence or absence of the outcome, are cerebral disability) was considered a favorable outcome, detailed in Table 1. and a CPC score of 3 (severe cerebral disability, conscious In-hospital mortality was 50% (n = 31). The cause but dependent), 4 (coma) or 5 (death) was classified as a of death was brain injury in 83.9% of patients (n = 26; poor outcome. Follow-up was performed by the inves- median survival 9 days [IQR, 6–11]), post-cardiac arrest tigators, who were unaware of the GDF-15 results, with shock and subsequent multiple organ failure in 12.9% telephone interviews and by reviewing electronic patient (n = 4; median survival 3 days [IQR, 2.5–3]), and other in records. 3.2% (n = 1; survival 91 days). Among patients dying from neurologic injury, 11.5% (n = 3) were in cerebral death Statistical analysis (median survival 3 days [IQR, 2.5–4.5]) and remaining Categorical variables are expressed as a number and 88.5% (n = 23) died after withdrawal of life-sustaining percentage; continuous variables are expressed as the therapies (median survival 9 days [IQR, 7–11.5]). In median and interquartile range (IQR). Comparisons of those patients with poor outcome, neuroprognostica- categorical variables were performed with the χ2 test. tion included two or more ancillary tests in 75% of them. Comparisons of continuous variables were performed Although EEG was the most widely used (81.2%), since with the Wilcoxon’s rank sum test. Univariable and mul- SSEP were available, 81.8% of these patients were stud- tivariable logistic regression models were performed ied with them. Tests performed and their main results with the backward stepwise procedure to determine are detailed in Additional file 3: Table S2 and Additional whether baseline GDF-15 (logarithm transformed) con- file 4: Table S3. stituted an independent predictor of poor neurological Poor neurological outcome at 6 months occurred in outcome. Variables that were not normally distributed 51.6% (n = 32) of patients. Baseline GDF-15 levels were were transformed to their natural logarithm. Odds ratios significantly higher in patients with CPC 3–5 compared (ORs) with 95% confidence intervals (CIs) are reported. to those with CPC 1–2 (median, 17.1 [IQR, 11.1–20.4] The following variables, obtained from the previously ng/mL vs. 7.6 [IQR, 4.1–13.1] ng/mL; p = 0.004) (Fig. 1). validated CAHP score [11], were incorporated into the In the 34 patients in which this data was available, the regression model: age, delay between collapse and CPR, blood samples corresponding to baseline levels were delay between CPR and ROSC, home setting arrest, no obtained a median of 138 [IQR, 115–200] min after bystander CPR, non-shockable rhythm, dose of epi- arrest. nephrine, and pH at admission. Receiver operating characteristic (ROC) curve analyses were performed to Predictors of poor neurological outcome evaluate whether baseline GDF-15 levels could predict Multivariable logistic regression analyses showed that poor neurological outcome at 6 months. We assessed age (OR, 1.09; 95% CI 1.01–1.17; p = 0.020), home set- whether there were any improvements in discrimination, ting arrest (OR, 8.07; 95% CI 1.61–40.42; p = 0.011), Rueda et al. Ann. Intensive Care (2019) 9:119 Page 4 of 10 Table 1 Baseline characteristics of the study cohort Characteristics All patients Favorable ­outcomea Unfavorable ­outcomea p value (n = 62) (n = 30) (n = 32) Demographics Age, years 59 (52–71) 56 (48–60) 67 (57–74) 0.002 Female sex 11 (17.7) 7 (23.3) 4 (12.5) 0.264 Clinical history Tobacco use 34 (54.8) 19 (63.3) 15 (46.9) 0.193 Arterial hypertension 34 (54.8) 12 (40.0) 22 (68.8) 0.023 Diabetes mellitus 15 (24.2) 2 (6.7) 13 (40.6) 0.002 Prior MI 10 (16.1) 3 (10.0) 7 (21.9) 0.204 CVD 6 (9.7) 2 (6.7) 4 (12.5) 0.438 Resuscitation variables Home setting arrest 23 (37.1) 5 (16.7) 18 (56.3) 0.001 Witnessed arrest 61 (98.4) 30 (100) 31 (96.9) 0.329 Bystander CPR 32 (51.6) 23 (76.7) 9 (28.1) < 0.001 Collapse-CPR duration, min 4 (1–7) 3 (1–5) 5 (2–11) 0.004 CPR-ROSC duration, min 22 (13–30) 19 (11–31) 23 (15–30) 0.280 Non-shockable rhythm 10 (16.1) 1 (3.3) 9 (28.1) 0.008 Number of defibrillations 4 (2–6) 3 (2–6) 4 (1–6) 0.419 Epinephrine 0.004 0 12 (19.4) 11 (36.7) 1 (3.1) 1–2 mg 19 (30.7) 7 (23.3) 12 (37.5) ≥ 3 mg 31 (50.0) 12 (40.0) 19 (59.4) Admission GCS 3 (3–5) 5 (3–7) 3 (3–3) < 0.001 Admission creatinine, (µmol/L) 114.9 (97.2–139.7) 109.6 (91.0–132.6) 128.6 (109.6–141.9) 0.033 Admission pH 7.22 (7.13–7.28) 7.25 (7.18–7.31) 7.19 (7.09–7.26) 0.065 Admission lactate, mmol/Lb 4.4 (2.6–6.3) 3.4 (2.2–6.6) 5.2 (2.8–6.4) 0.337 Admission GDF-15, ng/mL 12.4 (5.7–19.6) 7.6 (41.4–13.0) 17.1 (11.1–20.4) 0.004 ICCU treatment Mechanical ventilation 62 (100) 30 (100) 32 (100) – Therapeutic hypothermia 37 (59.7) 20 (66.7) 17 (53.1) 0.277 Coronary angiography 52 (83.9) 28 (93.3) 24 (75.0) 0.050 Cardiac arrest etiology STEMI 38 (61.3) 20 (66.7) 18 (56.3) 0.400 NSTEMI 13 (21.0) 4 (13.3) 9 (28.1) 0.153 Vasospastic angina 4 (6.45) 3 (10.0) 1 (3.1) 0.271 Chronic CAD 3 (4.8) 0 3 (9.4) 0.086 Cardiomyopathy 2 (3.2) 2 (6.7) 0 0.138 Acute myocarditis 1 (1.6) 1 (3.3) 0 0.298 Others 1 (1.6) 0 1 (3.1) 0.329 Data are presented as the number of patients (%) or the median (IQR) MI myocardial infarction, CVD cerebrovascular disease, CPR cardiopulmonary resuscitation, ROSC return of spontaneous circulation, GCS Glasgow Coma Scale, ICCU​ intensive cardiac care unit, STEMI ST-elevation myocardial infarction, NSTEMI non-ST-elevation myocardial infarction, CAD coronary artery disease a Outcome favorability was based on the Cerebral Performance Category (CPC) score: scores 1–2 = favorable; scores 3–5 = unfavorable b Estimation in 55 patients no bystander CPR (OR, 7.91; 95% CI 1.84–34.01; neurological outcomes were higher in individuals with p = 0.005), and baseline GDF-15 levels (OR, 3.74; 95% high baseline GDF-15 levels, compared to those with CI 1.32–10.60; p = 0.013) were independent predictors low baseline GDF-15 levels (Fig. 2). of the occurrence of the primary endpoint (CPC 3–5) In an exploratory analysis, we studied the dynamics of (Table 2). Moreover, the predictive margins of adverse GDF-15 within the first 24 h after admission in the 58 Rueda et al. Ann. Intensive Care (2019) 9:119 Page 5 of 10 patients in which these samples were available. We found a descending pattern, with a higher median GDF-15 concentration at baseline than at 12 or at 24 h. This pat- tern was observed in patients with and without adverse neurological outcomes. However, at all 3 specified time points, GDF-15 levels were higher in patients with poor neurological outcomes than in patients with favorable neurological outcomes. This pattern was also observed in patients with OHCA secondary to STEMI and in patients treated with therapeutic hypothermia (Additional file 5: Figure S2 and Additional file 6: Figure S3). Additionally, in another exploratory analysis, we analyzed associations between GDF-15 levels at 12 h and 24 h (one model for each specified time-point) and the clinical endpoint. We found associations that remained significant, even after a multivariable logistic regression analysis (OR, 3.90; CI 95% 1.15–13.26; p = 0.029 and OR, 3.65; CI 95%, 1.42– 9.44; p = 0.007, respectively) (Additional file 7: Table S4, Additional file 8: Table S5). Considering that levels of GDF-15 at 12 h and 24 h could have been influenced in TTM group, baseline characteristics of patients accord- ing to TTM treatment are detailed in Additional file 9: Table S6. Fig. 1 Violin plot showing the distribution of GDF-15 values in patients with and patients without adverse neurological outcome Table 2 Results of univariable and multivariable logistic regression analyses for identifying predictors of neurological outcome at 6 months Univariable logistic regression Multivariable logistic regression OR 95% CI p OR 95% CI p Age, years 1.07 1.02–1.13 0.004 1.09 1.01–1.17 0.020 Collapse-CPR ­durationa 2.17 1.24–3.79 0.006 CPR-ROSC ­durationa 1.74 0.65–4.68 0.269 Home setting arrest 6.43 1.96–21.07 0.002 8.07 1.61–40.42 0.011 No bystander CPR 8.40 2.67–26.37 < 0.001 7.91 1.84–34.01 0.005 Non-shockable rhythm 11.35 1.34–96.18 0.026 Epinephrine, mg 0 1 1–2 18.86 1.99–178.8 0.010 ≥3 17.42 1.99–152.7 0.010 Admission creatinine 4.87 1.14–20.88 0.033 Admission pH 0.034 0.001–1.754 0.093 Admission ­lactateb 1.54 0.65–3.64 0.321 Admission GDF-15a 2.88 1.40–5.92 0.004 3.74 1.32–10.60 0.013 Multivariate results are presented after backward elimination was completed CPR cardiopulmonary resuscitation, ROSC return of spontaneous circulation a Transformed on a natural logarithmic scale b Estimation in 55 patients Rueda et al. Ann. Intensive Care (2019) 9:119 Page 6 of 10 Fig. 2 Graph plot showing predictive margins (solid line), with 95% confidence intervals (shaded area), of adverse neurological outcome, according to GDF-15 levels. GDF-15 was transformed on a natural logarithmic scale Fig. 3 ROC curves showing the accuracy of GDF-15 for predicting Incremental prognostic value of GDF‑15 over clinical risk poor neurologic outcome at 6 months, when added in a dichotomous manner (≥ 10.8 ng/mL) to the short clinical model factors We found that 10.8 ng/ml was the optimal GDF-15 cut- off level for maximum classification efficiency (Table 3). which allowed the reclassification of 37.1% (CI 95% First, we evaluated the addition of GDF-15, in a dichot- 11.3–54.8) of patients. Similarly, the addition of GDF- omous manner (≥ 10.8 ng/mL vs. < 10.8 ng/mL), to a 15 to an extended clinical model (i.e., the short clinical short clinical model, which contained variables signifi- model, plus the collapse-to-CPR duration, non-shockable cantly associated with adverse outcome after a multi- rhythm, and epinephrine) improved discrimination. The variable regression analysis (i.e., age, home setting arrest, AUC increased from 0.895 (CI 95% 0.818–0.972) to 0.942 and no bystander CPR). We found that the addition of (CI 95% 0.886–0.997), and the associated continuous NRI GDF-15 improved discrimination. The area under the was 1.15 (CI 95% 0.32–1.73), which allowed the reclassifi- curve (AUC) increased from 0.867 (CI 95% 0.775–0.959) cation of 32.3% (CI 95% 4.8–46.8) of patients. Additional to 0.917 (CI 95% 0.849–0.984) (Fig. 3), and the asso- file 10: Table S7 summarizes the calibration, discrimina- ciated continuous NRI was 0.90 (CI 95% 0.48–1.44), tion, and reclassification metrics used. Discussion Table 3 Sensitivity and specificity for the prediction In this study, we found that the circulating GDF-15 level, of poor outcome for different baseline GDF-15 cutoff measured at admission after an OHCA, was an independ- values ent and early predictor of poor neurologic outcome at Cutoff (ng/mL) Sensitivity (%) Specificity (%) Correctly 6 months. In addition, when the GDF-15 level was added to classified (%) a clinical model that included the usual clinical predictors available shortly after admission, it improved discrimina- 1.6 100 3.3 53.2 tion, and one-third of patients could be reclassified. GDF- 4.5 90.6 30.0 61.3 15 levels showed a large, early increase after an OHCA, 7.0 84.4 46.7 66.1 followed by a sustained decline during the first 24 h. 10.8a 78.1 66.7 72.6 Prognostication after an OHCA remains challeng- 12.5 68.8 70.0 69.3 ing, particularly regarding brain damage, and accurate 15.0 62.5 76.7 69.4 evaluation usually needs several days [4, 8]. Unlike 17.5 46.9 76.7 61.3 other predictors, like examination or EEG, biomark- 20.9 25.0 83.3 53.2 ers can provide early, quantitative data, independent 27.1 15.6 96.7 54.8 of the effects of sedatives. Consequently, they repre- a Optimal cutoff point for maximum efficiency (false negative cost = false sent a growing area of interest. Several markers of brain positive cost) Rueda et al. Ann. Intensive Care (2019) 9:119 Page 7 of 10 damage have been studied [12], in particular NSE, [5]. Thus, GDF-15 also represents a marker of biological which is the only included in current clinical guide- age and chronic disease burden. lines [4, 8]. However, the present role of biomarkers is In the present study, blood GDF-15 increased rapidly limited to cases in which more robust predictors, such to high levels after circulation was restored; this increase as physical examination or SSEP, provide inconclusive was followed by a decline during the first 24 h (Additional results. Despite providing a notable discrimination, file 5: Figure S2). Neuronal damage can be a source of cir- their main limitation is the difficulty in establishing a culating GDF-15. In support of this hypothesis, experi- consistent threshold with a zero false positive rate. In mental data have demonstrated that GDF-15 was locally addition, another weakness of most proposed neu- overexpressed in rodent models of brain injury [28, 29]. roinjury biomarkers is a delayed release to blood in Furthermore, other studies have described an association response to ischemia/reperfusion processes. The peaks between rising GDF-15 levels and functional outcome of blood levels, and as a result the optimal time point after an ischemic stroke [30, 31]. Interestingly, a small for measuring, are usually observed from 24 h after study noted a significant correlation between the levels of ROSC. In the case of NSE, blood levels peak at 48–96 h GDF-15 and S-100, as well as similar discriminative capa- after an arrest [13] and, accordingly, the highest prog- bilities [30]. However, the great releasing of GDF-15 after nostic accuracy has been reported at 48 h post-ROSC, OHCA seems to respond mainly to the very early global or when measured serially during the first 72 h [14]. inflammatory response related to post-cardiac arrest syn- Protein S-100 exhibits an earlier peak and could pro- drome (PCAS), rather than ongoing brain injury. vide a predictive value similar to NSE already at 24 h PCAS is a unique, complex combination of pathologi- from the arrest [15, 16]. Secretoneurin [17] and glial cal processes, which include brain injury, myocardial dys- fibrillary acidic protein [18] have also been evaluated, function, systemic ischemia/reperfusion responses, and but they showed a lower discrimination and no tempo- often, the unresolved disease process that caused the car- ral advantage respect NSE and S-100. In recent years, diac arrest [32]. Severity of PCAS is a major determinant two new biomarkers, tau protein and neurofilament of outcome [33, 34], largely dependent on the duration of light chain (NFL), have been proposed. Serum tau may the whole body ischemia and reperfusion injury, which offer better diagnostic accuracy than NSE for poor are also the main trigger of the inflammatory system. outcome, although the highest predictive values are Therefore, there is a close association between the sever- likewise observed at 48–72 h from ROSC [19]. By con- ity of PCAS, the magnitude of the inflammatory response, trast, NFL seems superior to other biomarkers (NSE, the severity of organ dysfunction and neurologic out- S-100 and tau) when assessed as early as 24 h after car- come, as inferred from small previous studies on inflam- diac arrest [20]. Nevertheless, despite these promising matory biomarkers in this clinical setting. Copeptin levels results, more data should be available before its routine on admission have been associated with death and subse- use in prognostication. quent organ failure in one study [35], and were predictive GDF-15 is a stress-responsive member of the transform- for neurological outcome in another [36], when measured ing growth factor-β (TGF-β) cytokine superfamily. GDF- within 48 h after ROSC. Circulating procalcitonin were 15 is weakly expressed in tissues, including the central related with the severity of PCAS and predicted neuro- nervous system [21], under normal conditions. Although logical outcome accurately already at 12 h from the arrest its pathobiology is not fully understood, it is strongly [33]. More recently, interleukin-6 emerged as a poten- induced by macrophages in response to inflammation and tial early biomarker. Its levels on admission, but not the tissue injury. Thus, circulating levels of GDF-15 have been high-sensitivity C-reactive protein or the S-100 ones, have identified as an inflammatory biomarker with prognostic been associated with extra-cerebral organ dysfunction value in several conditions, particularly in cardiovascular and were independent predictors for poor neurologic out- diseases. In the acute setting, increased levels are a robust come [37]. Nevertheless, it is uncertain whether IL-6 can predictor of organ dysfunction and death from acute myo- provide incremental value above traditional factors asso- cardial infarction [6, 22, 23] to cardiogenic shock [24]. ciated with a poor prognosis [38]. GDF-15 also serves as biomarker in other critical disease It is clear that GDF-15 shares with the previously dis- conditions, such as acute pulmonary embolism [25], acute cussed neuroinjury biomarkers the same limitations respiratory distress syndrome [26], or sepsis [27]. On the regarding the possibility of false positives. In addition, other hand, at lower cut-off values, GDF-15 levels can pre- GDF-15 is not specific for neuronal damage. This can dict long-term cardiovascular events, bleeding, cancer, and explain why we observed lower AUCs for poor outcome all-cause mortality, both in patients with chronic heart dis- prediction than the reported ones from neuroinjury bio- eases and in individuals that dwell in community settings markers (usually ≥ 0.90) when measured at their opti- mal time points. Even so, we should understand that Rueda et al. Ann. Intensive Care (2019) 9:119 Page 8 of 10 biomarkers are a means to explore the different path- Supplementary information ways involved in PCAS. They offer complementary and Supplementary information accompanies this paper at https​://doi. sequential information. Inflammatory biomarkers, such org/10.1186/s1361​3-019-0593-9. as GDF-15, could provide additional value to neuroinjury markers, particularly in the early acute phase [39]. Thus, Additional file 1: Figure S1. Flow chart of patients included in the study. to maximize the predictive value of biomarkers, instead Additional file 2: Table S1. Baseline characteristics of patients included of employing a single, best biomarker strategy, a multi- in the study in comparison with those excluded because measurement of GDF-15 on admission or other relevant data were missing. marker approach may be more advantageous [12]. This Additional file 3: Table S2. Distribution and results of clinical and ancil‑ may include acute-phase, myocardial dysfunction and lary examinations in all patients included in the study. neuroinjury biomarkers, summarizing the main compo- Additional file 4: Table S3. Distribution and results of clinical and ancil‑ nents of PCAS. Several scores based on clinical param- lary examinations in patients who died in-hospital from neurological eters rapidly available on admission have been developed causes. and validated [11, 40]. Further research might address Additional file 5: Figure S2. GDF-15 levels during the first 24 h of whether the addition of information from biomarkers admission in patients with and without adverse neurological outcomes. Comparisons between groups based on the Mann–Whitney test. could provide enough accuracy to guide decision-making Additional file 6: Figure S3. GDF-15 levels during the first 24 h of and information to relatives in the first hours after arrest. admission, according to therapeutic hypothermia treatment (TTM) and ST-elevation myocardial infarction (STEMI). Additional file 7: Table S4. Results from univariable and multivariable Limitations logistic regression analyses, including GDF-15 levels measured at 12 h. Several limitations of our study should be noted. First Additional file 8: Table S5. Results from univariable and multivariable and foremost, it was performed at a single center and logistic regression analyses, including GDF-15 levels measured at 24 h. involved a small number of patients, which limited the Additional file 9: Table S6. Baseline characteristics of patients who generalization of our observations. Included popula- received targeted temperature management in comparison with those who did not. tion was limited to OHCA of cardiac cause, and exter- nal validity is restricted to these patients. Second, other Additional file 10: Table S7. Performance of models for the neurological outcome at 6 months. biomarkers were not analyzed. NSE was only available in 27 (43.5%) patients. Analyses lacked statistical power Abbreviations to extract conclusions about its relationship with GDF- OHCA: out-of-hospital cardiac arrest; GDF-15: growth differentiation factor 15. However, the study was focused on variables with 15; STEMI: ST-segment elevation myocardial infarction; CPR: cardiopulmonary established predictive value short after admission, and resuscitation; ROSC: return of spontaneous circulation; EEG: electroencepha‑ logram; SSEP: somatosensory evoked potentials; CT: computed tomography; NSE is useful a minimum of 24 h after arrest. Third, in NSE: neuron-specific enolase; GMS: Glasgow motor scale; CPC: cerebral many cases, clinical outcome was determined by the performance category; IQR: interquartile range; OR: odds ratio; CI: confi‑ withdrawal of life support. Thus, the outcome could have dence interval; ROC: receiver operating characteristic; NRI: net reclassification improvement; AUC​: area under the curve; NFL: neurofilament light chain; been affected by a so-called “self-fulfilling prophecy”. PCAS: post-cardiac arrest syndrome. Nevertheless, treating physicians were not aware of GDF- 15 measurements, since analyses were performed after Acknowledgements Not applicable. treatment of all patients was completed. More studies are needed to validate our results externally and evaluate the Authors’ contributions clinical utility of GDF-15 in this group of patients. FR, JA, and GC: conceptualization and design of the study. JA, MGL, CAN, JSG, and FR: data collection. CAN and JSG: blood samples acquisition and process. GC: analysis of the data and statistics. FR, GC, and CG-G: preparation of original draft. All authors: review and editing. AB-G: review of the data analysis and Conclusions interpretation, and revision of the manuscript for the final version. All authors In comatose patients that survived an OHCA, high cir- read and approved the final manuscript. culating GDF-15 levels on admission were an independ- Funding ent and early predictor of severe neurologic disability at Roche Diagnostics supported this research by running the GDF-15 assays free 6 months. When added to the most common clinical fac- of charge. However, the authors are solely responsible for the design and con‑ tors associated with a poor outcome, GDF-15 improved duct of the study, all analyses, and the drafting and editing of the manuscript. discrimination and allowed patient reclassification. Availability of data and materials Further studies are needed to assess whether the incor- The datasets used and/or analysed during the current study are available from poration of GDF-15 to a multimodal risk stratification the corresponding author on reasonable request. approach might provide early and accurate neurologic Ethics approval and consent to participate prognostication. The protocol was approved on 22 July 2011 by the Germans Trias i Pujol Hospital Ethics Committee for Investigation (reference EO-11-061). Patients or their representatives provided written informed consent. Rueda et al. Ann. Intensive Care (2019) 9:119 Page 9 of 10 Consent for publication spontaneous circulation and treated with hypothermia. Emerg Med J. Not applicable. 2016;33:690–5. 16. Stammet P, Dankiewicz J, Nielsen N, Fays F, Collignon O, Hassager C, et al. Competing interests Protein S100 as outcome predictor after out-of-hospital cardiac arrest AB-G received honoraria for lectures and for advisory board participation, from and targeted temperature management at 33 °C and 36 °C. Crit Care. Roche Diagnostics. The other authors report no competing interests. 2017;21:153. 17. Hasslacher J, Lehner GF, Harler U, Beer R, Ulmer H, Kirchmair R, et al. Author details Secretoneurin as a marker for hypoxic brain injury after cardiopulmonary 1 Heart Institute, Germans Trias i Pujol University Hospital, Carretera de Canyet resuscitation. Intensive Care Med. 2014;40:1518–27. s/n, Badalona, 08916 Barcelona, Spain. 2 Department of Medicine, CIBERCV, 18. Larsson IM, Wallin E, Kristofferzon ML, Niessner M, Zetterberg H, Ruberts‑ Autonomous University of Barcelona, Barcelona, Spain. 3 PhD Program in Inter‑ son S. Post-cardiac arrest serum levels of glial fibrillary acidic protein for nal Medicine, Autonomous University of Barcelona, Barcelona, Spain. predicting neurological outcome. Resuscitation. 2014;85:1654–61. 19. Mattsson N, Zetterberg H, Nielsen N, Blennow K, Dankiewicz J, Friberg H, Received: 17 May 2019 Accepted: 5 October 2019 et al. Serum tau and neurological outcome in cardiac arrest. Ann Neurol. 2017;5:665–75. 20. Moseby-Knappe M, Mattsson N, Nielsen N, Zetterberg H, Blennow K, Dankiewicz J, et al. Serum neurofilament light chain for prognosis of outcome after cardiac arrest. JAMA Neurol. 2019;76:64–71. References 21. Strelau J, Sullivan A, Böttner M, Lingor P, Falkenstein E, Suter-Crazzolara C, 1. Laver S, Farrow C, Turner D, Nolan J. Mode of death after admission et al. Growth/differentiation factor-15/macrophage inhibitory cytokine-1 to an intensive care unit following cardiac arrest. Intensive Care Med. is a novel trophic factor for midbrain dopaminergic neurons in vivo. J 2004;30:2126–8. Neurosci. 2000;20:8597–603. 2. Lemiale V, Dumas F, Mongardon N, Giovanetti O, Charpentier J, Chiche 22. Kempf T, Björklund E, Olofsson S, Lindahl B, Allhoff T, Peter T, et al. Growth- JD, et al. Intensive care unit mortality after cardiac arrest: the relative differentiation factor-15 improves risk stratification in ST-segment eleva‑ contribution of shock and brain injury in a large cohort. Intensive Care tion myocardial infarction. Eur Heart J. 2007;28:2858–65. Med. 2013;39:1972–80. 23. Khan SQ, Ng K, Dhillon O, Kelly D, Quinn P, Squire IB, et al. Growth dif‑ 3. Dragancea I, Rundgren M, Englund E, Friberg H, Cronberg T. The influence ferentiation factor-15 as a prognostic marker in patients with acute of induced hypothermia and delayed prognostication on the mode of myocardial infarction. Eur Heart J. 2009;30:1057–65. death after cardiac arrest. Resuscitation. 2013;84:337–42. 24. Fuernau G, Poenisch C, Eitel I, de Waha S, Desch S, Schuler G, et al. 4. Sandroni C, Cariou A, Cavallaro F, Cronberg T, Friberg H, Hoedemaekers C, Growth-differentiation factor 15 and osteoprotegerin in acute myocardial et al. Prognostication in comatose survivors of cardiac arrest: an advisory infarction complicated by cardiogenic shock: a biomarker substudy of statement from the European Resuscitation Council and the European the IABP-SHOCK II-trial. Eur J Heart Fail. 2014;16:880–7. Society of Intensive Care Medicine. Resuscitation. 2014;85:1779–89. 25. Lankeit M, Kempf T, Dellas C, Cuny M, Tapken H, Peter T, et al. Growth 5. Wollert KC, Kempf T, Wallentin L. Growth differentiation factor 15 as a differentiation factor-15 for prognostic assessment of patients with acute biomarker in cardiovascular disease. Clin Chem. 2017;63:140–51. pulmonary embolism. Am J Respir Crit Care Med. 2008;177:1018–25. 6. Garcia-Garcia C, Rueda F, Lupon J, Oliveras T, Labata C, Ferrer M, et al. 26. Clark BJ, Bull TM, Benson AB, Stream AR, Macht M, Gaydos J, et al. Growth Growth differentiation factor-15 is a predictive biomarker in primary differentiation factor-15 and prognosis in acute respiratory distress ventricular fibrillation: the RUTI-STEMI-PVF study. Eur Heart J Acute Car‑ syndrome: a retrospective cohort study. Crit Care. 2013;17:R92. diovasc Care. 2018. https​://doi.org/10.1177/20488​72618​79759​9. 27. Buendgens L, Yagmur E, Bruensing J, Herbers U, Baeck C, Trautwein C, 7. Soar J, Nolan JP, Böttiger BW, Perkins GD, Lott C, Carli P, et al. European et al. Growth differentiation factor-15 is a predictor of mortality in criti‑ Resuscitation Council Guidelines for Resuscitation 2015: section 3. Adult cally ill patients with sepsis. Dis Markers. 2017;2017:5271203. advanced life support. Resuscitation. 2015;95:100–47. 28. Schindowski K, von Bohlen und Halbach O, Strelau J, Ridder DA, Her‑ 8. Nolan JP, Soar J, Cariou A, Cronberg T, Moulaert VR, Deakin CD, et al. rmann O, Schober A, et al. Regulation of GDF-15, a distant TGF-β super‑ European Resuscitation Council and European Society of Intensive Care family member, in a mouse model of cerebral ischemia. Cell Tissue Res. Medicine Guidelines for Post-resuscitation Care 2015: section 5 of the 2011;343:399–409. European Resuscitation Council Guidelines for Resuscitation 2015. Resus‑ 29. Schober A, Böttner M, Strelau J, Kinscherf R, Bonaterra GA, Barth M, et al. citation. 2015;95:202–22. Expression of growth differentiation factor-15/macrophage inhibitory 9. Kempf T, Horn-Wichmann R, Brabant G, Peter T, Allhoff T, Klein G, et al. cytokine-1 (GDF-15/MIC-1) in the perinatal, adult, and injured rat brain. J Circulating concentrations of growth-differentiation factor 15 in appar‑ Comp Neurol. 2001;439:32–45. ently healthy elderly individuals and patients with chronic heart failure 30. Worthmann H, Kempf T, Widera C, Tryc AB, Goldbecker A, Ma YT, et al. as assessed by a new immunoradiometric sandwich assay. Clin Chem. Growth differentiation factor 15 plasma levels and outcome after 2007;53:284–91. ischemic stroke. Cerebrovasc Dis. 2011;32:72–8. 10. Jennett B, Bond M. Assessment of outcome after severe brain damage. 31. Gröschel K, Schnaudigel S, Edelmann F, Niehaus CF, Weber-Krüger M, Lancet. 1975;1:480–4. Haase B, et al. Growth-differentiation factor-15 and functional outcome 11. Maupain C, Bougouin W, Lamhaut L, Deye N, Diehl JL, Geri G, et al. The after acute ischemic stroke. J Neurol. 2012;259:1574–9. CAHP (Cardiac Arrest Hospital Prognosis) score: a tool for risk stratification 32. Neumar RW, Nolan JP, Adrie C, Aibiki M, Berg RA, Böttiger BW, et al. Post- after out-of-hospital cardiac arrest. Eur Heart J. 2016;37:3222–8. cardiac arrest syndrome: epidemiology, pathophysiology, treatment, and 12. Gul SS, Huesgen KW, Wang KK, Mark K, Tyndall JA. Prognostic utility of prognostication. A consensus statement from the International Liaison neuroinjury biomarkers in post out-of-hospital cardiac arrest (OHCA) Committee on Resuscitation (American Heart Association, Australian and patient management. Med Hypotheses. 2017;105:34–47. New Zealand Council on Resuscitation, European Resuscitation Council, 13. Rundgren M, Karlsson T, Nielsen N, Cronberg T, Johnsson P, Friberg H. Heart and Stroke Foundation of Canada, InterAmerican Heart Founda‑ Neuron specific enolase and S-100B as predictors of outcome after tion, Resuscitation Council of Asia, and the Resuscitation Council of cardiac arrest and induced hypothermia. Resuscitation. 2009;80:784–9. Southern Africa); the American Heart Association Emergency Cardio‑ 14. Stammet P, Collignon O, Hassager C, Wise MP, Hovdenes J, Åneman A, vascular Care Committee; the Council on Cardiovascular Surgery and et al. Neuron-specific enolase as a predictor of death or poor neurological Anesthesia; the Council on Cardiopulmonary, Perioperative, and Critical outcome after out-of-hospital cardiac arrest and targeted temperature Care; the Council on Clinical Cardiology; and the Stroke Council. Circula‑ management at 33 °C and 36 °C. J Am Coll Cardiol. 2015;65:2104–14. tion. 2008;118:2452–83. 15. Choi S, Park K, Ryu S, Kang T, Kim H, Cho S, et al. Use of S-100B, NSE, CRP 33. Annborn M, Dankiewicz J, Erlinge D, Hertel S, Rundgren M, Smith JG, and ESR to predict neurological outcomes in patients with return of et al. Procalcitonin after cardiac arrest—an indicator of severity of illness, ischemia–reperfusion injury and outcome. Resuscitation. 2013;84:782–7. Rueda et al. Ann. Intensive Care (2019) 9:119 Page 10 of 10 34. Roberts BW, Kilgannon JH, Chansky ME, Mittal N, Wooden J, Parrillo JE, 38. Bro-Jeppesen J, Kjaergaard J, Stammet P, Wise MP, Hovdenes J, Åneman et al. Multiple organ dysfunction after return of spontaneous circulation A, et al. Predictive value of interleukin-6 in post-cardiac arrest patients in postcardiac arrest syndrome. Crit Care Med. 2013;41:1492–501. treated with targeted temperature management at 33 °C or 36 °C. Resus‑ 35. Ristagno G, Latini R, Plebani M, Zaninotto M, Vaahersalo J, Masson S, citation. 2016;98:1–8. et al. Copeptin levels are associated with organ dysfunction and death 39. Isenschmid C, Kalt J, Gamp M, Tondorf T, Becker C, Tisljar K, et al. Routine in the intensive care unit after out-of-hospital cardiac arrest. Crit Care. blood markers from different biological pathways improve early risk 2015;19:132. stratification in cardiac arrest patients: results from the prospective, 36. Broessner G, Hasslacher J, Beer R, Lackner P, Lehner GF, Harler U, et al. observational COMMUNICATE study. Resuscitation. 2018;130:138–45. Outcome prediction and temperature dependency of MR-proANP 40. Aschauer S, Dorffner G, Sterz F, Erdogmus A, Laggner A. A prediction and Copeptin in comatose resuscitated patients. Resuscitation. tool for initial out-of-hospital cardiac arrest survivors. Resuscitation. 2015;89:75–80. 2014;85:1225–31. 37. Vaahersalo J, Skrifvars MB, Pulkki K, Stridsberg M, Røsjø H, Hovilehto S, et al. Admission interleukin-6 is associated with post resuscitation organ dysfunction and predicts long-term neurological outcome after out-of- Publisher’s Note hospital ventricular fibrillation. Resuscitation. 2014;85:1573–9. Springer Nature remains neutral with regard to jurisdictional claims in pub‑ lished maps and institutional affiliations.