Pediatric brain trauma outcome prediction using paired serum levels of inflammatory mediators and brain-specific proteins

Introduction TRAUMATIC BRAIN INJURY (TBI) is the most common cause of death anddisability in the pediatric population (Suominen et al., 1998; Walker and Cass, 1987). Accurate outcome prediction during the acute management of brain trauma remains difficult. Biochemical mediators are produced and released after brain trauma (Bell et al., 1997; Ergun et al., 1998; Fridriksson et al., 2000; Kalabalikis et al., 1999; Kossmann et al., 1995; McClain et al., 1991; McKeating et al., 1998; Perssonet al., 1987; Pleines et al., 1998; Raabe et al., 1999a,b; Skogseid et al., 1992; Takahashi et al., 1989; Whalen et al., 1998; Woertgen et al., 1997; Yamazaki et al., 1995) to regulate the complex cascade of secondary brain insult and repair processes, or when there is neuronal or glial damage. Many of these mediators are measurable in blood or cerebrospinal fluid (CSF), making them potentially useful for outcome prediction (Bell et al., 1997; Ergun et al., 1998; Fridriksson etal., 2000; Kalabalikis et al., 1999; Kossmann et al., 1995; McClain et al., 1991; McKeating et al. 1998; Persson et al., 1987; Pleines etal., 1998; Raabe et al., 1999a,b; Skogseid et al., 1992; Takahashi et al., 1989; Whalen et al., 1998; Woertgen et al., 1997; Yamazaki et al., 1995).

The biochemical properties of potential brain trauma biomarkers previously described in the literature are diverse and include brain-specific proteins such as S100b protein (S100b) and neuron-specific enolase (NSE), which represent the degree of astrocytic and neuronal damage (Ergun et al., 1998; Marangos and Schmechel, 1987; Rothoerl et al., 1998; Skogseid et al., 1992; Woertgen et al., 1997; Yamazaki et al., 1995); pro- and anti-inflammatory mediators such as interleukin-6 (IL-6) (Arand et al., 2001; Chiaretti et al., 2005; Minambres et al.,2003) and interleukin-10 (IL-10) (Csuka et al., 1999); chemotaxins such as interleukin-8 (IL-8) (Kossmann et al., 1997; Kushi et al., 2003; Whalen et al., 2000); adhesion molecules such as L-selectin (McKeating et al., 1998); and vasoconstrictors such as endothelin (Yang et al., 2002). Most of these studies have only investigated the relationship between individual biomarker concentration and outcome without comparing the outcome predictive values of unrelated classes of biomarkers. It is, therefore, unclear from the existing literature whether inflammatory mediators or brain-specific (neuronal or glial) proteinshave higher outcome predictive values. Translation of brain trauma biomarker research into clinically useful prognostic tools has not been successful. This may be because potentially useful prognostic thresholds have only been described in a few studies for a very limited number of selected biomarkers to predict unfavorable outcome (Ross et al., 1996). In addition, no previous report in the literature has described outcome prediction using combinations of serum biomarker concentrations from different mediator families, which may offer more accurate outcome prediction than may be achieved with individual marker levels. This study aims to determine the serum prognostic thresholds and predictive values of eight biomarkers from different mediator familiesfor unfavorable outcome after childhood isolated brain trauma. We additionally aim to determine which paired combinations of biomarkers are better at predicting unfavorable outcome after isolated pediatric brain trauma than individual marker levels. Methods Patient enrollment Twenty-eight children (21 boys and seven girls) consecutively admitted to neuro-intensive care after isolated accidental brain trauma were enrolled into a single-center prospective observational study. The local ethics and hospital management committees approved the study.Parental consent was obtained for participation in the study. Clinical and demographic details (including age, sex, injury severity, and initial brain computed tomography [CT] findings) were collected prospectively. Primary brain injury severity was assessed using the post-resuscitation/pre-intubation modified pediatric Glasgow Coma Scale (GCS) (Reilly, et al. 1988), and was dichotomized into "severe injury," where GCS was 8 or less, and "non-severe injury," where GCS was more than 8, to assess their relationships with biomarker levels.Initial brain CT findings were dichotomized into "diffuse" and "focal" injuries. Global outcome was assessed at 6 months post-injury using the modified Glasgow Outcome Score (GOS) (Adelson et al., 1997), and dichotomized into "favorable recovery," when GOS 4 and 5 were achieved, and "unfavorable outcome," when GOS was 1-3. Sample collection and measurement All patients had an arterial blood sample collected at precisely 24 h (day 1) post-injury. The blood samples were collected in pyrogen-free plastic tubes and centrifuged at 1200xg for 10 min. The serum was removed immediately and stored in pyrogen-free plastic tubes at -70[degrees]C until analyses. Commercially available enzyme-linked immunosorbent assays (ELISAs)were used to quantify the serum concentrations of eight different biomarkers (Table 1). Samples were analyzed in duplicate and averaged to provide the final marker concentrations. Outcome predictive values and prognostic thresholds of individual biomarkers Mann-Whitney U-tests were employed to detect statistical difference (p<0.05) in mediator concentrations between the dichotomized injuryseverity and outcome groups. For each biomarker with significantly different levels between the outcome dichotomy, sensitivity and specificity of incremental values for prediction of unfavorable outcome at 6 months after brain injury were calculated. Receiver operator characteristic (ROC) curves were then plotted and the optimal cutoff value (prognostic threshold) for each mediator, which was defined as the point closest to the left upper corner of the ROC curve, was identified. Areas under the ROC curves (AUC) were measured to compare the outcome predictive values between individual biomarkers. Paired biomarker levels and multivariate ROC curves The algorithm of paired biomarkers consisted of a Boolean expression of the mediator concentrations related by an algebraic operator asthe components of the expression (i.e., biomarker A>a predefined "screening" threshold AND biomarker B>incremental thresholds predicting unfavorable outcome). One of the paired biomarkers was called the "screening marker" and had a single cut-point ("screening threshold") assigned. The other biomarker ("varying marker") had incremental thresholds, and corresponding sensitivities and specificities were calculated to plot the multivariate ROC (MultiROC) curve. Area under each MultiROC curve (AUCm) was measured to compare the outcome prediction performance between the different paired combinations. AUCm of the pairswere then compared with the AUC of the individual biomarkers to determine whether they or individual biomarkers had higher outcome predictive values. The "screening threshold" of each biomarker was determined by plotting individual mediator concentrations against the outcome dichotomyand was defined as the highest biomarker level that identified most patients with favorable recovery without including any patient with unfavorable outcome. In order to investigate the effect of different "screening threshold" levels within each combination and whether thesemight better differentiate outcome, the original "screening threshold" within each pair was reduced first by 10%, then 20%, and MultiROC curves were plotted again with the new "screening threshold" test levels, and the AUCm were determined. The analyses were repeated when the original "screening thresholds" were increased by 10% and 20%. The optimal threshold for the "varying marker" of each combinationwas determined, which was defined as the point closest to the left upper corner of the MultiROC curve. Biomarkers that individually did not predict outcome were excluded from all MultiROC analyses. Results Demographics and outcome Table 2 summarizes the demographic details. The median GCS was 7 (range, 3-13). Median age of the patients was 8.59 years (range, 0.33-14.17 years). At 6 months post-injury, 24 patients made a favorable recovery, whereas four children had an unfavorable outcome. Biomarker concentrations and demographic dichotomies Of the eight biomarkers assessed, only the soluble intracellular adhesion molecule (SICAM) and IL-6 levels were significantly higher inpatients with severe injury than those with non-severe injury (p=0.01 for SICAM, and p=0.02 for IL-6). The median NSE serum levels in patients with diffuse brain injury (identified on the initial brain CT scan) were two times higher than those with focal injury (p=0.01). None of the other biomarker levels differentiated diffuse and focal injuries. Individual neurochemical mediator concentrations and outcome Patients with unfavorable outcome had significantly higher day 1 serum concentrations of S-100b, NSE, L-selectin, IL-6, and IL-8 than those with favorable recovery (Fig. 1). The outliners in the S100b andIL-8 figures did not have any clinical, physiological, or other apparent differences to explain their outlying neurochemical levels. The outliners in both of these figures (S100b and IL-8) were different patients. Day 1 L-selectin level had the highest predictive value for unfavorable outcome at 6 months after brain trauma (AUC=0.92). Table 3summarizes the area under the ROC curves for individual biomarkers, their prognostic thresholds (optimal cutpoint), and the correspondingsensitivities and specificities. Paired biomarker levels and outcome prediction Because IL-10, endothelin, and SICAM levels did not differentiate between the outcome dichotomies, they were excluded from MultiROC analyses. There were, therefore, five possible screening markers. "Screening threshold" for each "screening marker" is shown in Figure 1. Each "screening marker" was paired in turn with the remaining four biomarkers to form 20 different combinations. Table 4 summarizes the top 10 combinations for unfavorable outcome prediction after childhood brain trauma, their AUCm, and optimal threshold for the "varying marker"within each of these combinations. Two pairs achieved 100% sensitivity and 96% specificity for unfavorable outcome prediction (Table 4). Comparison of the predictability for unfavorable outcome after brain trauma When using individual biomarker level to predict unfavorable outcome, only the day 1 L-selectin level had an AUC>0.90 (Table 3). Fifteen of the 20 combinations examined had AUC>0.90 for prediction of unfavorable outcome after brain trauma. Combinations using S100b as the "screening marker" had higher predictive values for unfavorable outcome than those using other mediators as the "screening marker." Outcomepredictive values were higher in combinations that used L-selectin as the "varying marker" than those using it as the "screening marker."Increasing or decreasing the "screening threshold" of each combination did not improve prediction for unfavorable outcome (Table 5). A day 1 L-Selectin level of 1200 ng/ml was 75% sensitive and 88% specific for predicting unfavorable outcome after brain trauma, but the specificity increased to 96% when this threshold was coupled with either S100b or NSE as the "screening marker" (Fig. 2). Discussion This pediatric brain trauma study demonstrated that the day 1 serum concentrations of inflammatory mediators had higher prognostic values than brain-specific proteins, but the best outcome predictive value was achieved with combinations of two biomarker levels from different mediator families. [FIGURE 1 OMITTED] Serum or CSF levels of brain-specific proteins and inflammatory mediators such as S100b, NSE, IL-6, and IL-8 measured at various time points after brain trauma have individually been described in the literature as potentially useful predictors of unfavorable outcome (Arandet al., 2001; Bandyopadhyay et al., 2005; Chiaretti et al., 2005; Jackson et al., 2000; Kushi et al., 2003; Minambres et al., 2003; Ross et al., 1996; Rothoerl et al., 1998; Spinella et al., 2003). It has not, however, been possible to ascertain from the existing literature which class and specific biomarker offers the most accurate outcome prediction after brain trauma. This may be because the majority of previous studies have concentrated on evaluating a single mediator or limited numbers of mediators from the same biomarker family. Additionally, reports in the literature often use different specimen types (CSF, arterial or venous blood), sample collection time-points, and data analyses, making comparison between studies difficult. Buttram et al. (2007) recently used multiplex bead array to quantify the CSF levels of 21 different cytokines at four various time points after pediatric brain trauma but failed to demonstrate any associations with outcome. Their cohort was comprised of 36 infants and children who had diverse mechanisms of injury, including isolated TBI, inflicted brain trauma, and polytrauma with widely variable time rangesfor each sample collection point (Buttram et al., 2007). Our study, in contrast, restricted evaluations to serum biomarker levels measured at exactly 24 h and only in isolated accidental brain trauma. This may explain the difference in outcome prediction of biomarkers between both studies, and the high predictive values for unfavorable outcome observed in our study. A recent study has demonstrated that the time courses of serum brain-specific protein levels differ significantlybetween different types of pediatric brain injury (Berger et al., 2006). Our study also evaluated biomarkers from different mediator families and demonstrated for the first time that some inflammatory markers(L-selectin and IL-8) had higher predictive values for unfavorable outcome than brain-specific (glial or neuronal) proteins. One possibleexplanation for our findings is that serum levels of proteins released from glia or neurons after brain injury may not accurately reflectthe extent of cellular damage. Serum inflammatory marker levels may,on the other hand, have better associations with the actual burden of neuro-inflammation, which in turn has a higher outcome predictive value. Alternatively, neuro-inflammation may have a greater influence on outcome after brain trauma than previously expected. [FIGURE 2 OMITTED] We demonstrated for the first time that the outcome predictive values vary between inflammatory mediators with L-selectin offering a higher predictive value than IL-6 and IL-8, while IL-10 and SICAM serumlevels had no association with outcome. Neuro-inflammation is a heterogeneous process with different pathways operating and subsiding at various time-points after brain trauma. To investigate this complex relationship between different biomarkers, their different time-courses and outcome prediction will require a larger cohort with sufficientnumbers of patients with poor outcome. This may be difficult to achieve in the light of the Western world-wide declining incidence of childhood brain trauma (Engberg and Teasdale, 1998; Ponsky et al. 2005; Thurman and Guerrero, 1999), and improvement in brain trauma management and outcome. Increasing numbers of biomarkers have been described as potentially useful prognosticators of brain trauma outcome, but none have actually been translated into clinically useful entities. This may be because the majority of articles in the literature have only reported on the varying concentrations of the different biomarkers in relation tooutcome, and only a few prior studies have proposed potentially useful prognostic thresholds, such as NSE (Ross et al., 1996). Our study is the first to describe serum prognostic thresholds for unfavorable outcome in five biomarkers simultaneously and includes some of the previously less well-investigated biomarkers such as L-selectin. Previous studies using individual biomarkers rarely yielded sensitivity and specificity of more than 85% (as reported here) for unfavorable outcome prediction (Ross et al., 1996). Brain trauma outcome is dependent upon many factors such as the severity of the primary braininjury (Bruce et al., 1978; Facco et al., 1986; Pal et al., 1989; Prasad et al., 2002) and the burden of secondary insults (Chambers et al., 2000, 2006; Chesnut et al., 1993; Jones et al., 2003; Miller and Becker, 1982), and to expect successful and accurate prognostication with a single serum biomarker is unrealistic. Berger et al. (2007) investigated the outcome predictive values of the serum levels of threebrain-specific proteins (S100b, NSE, and myelin basic protein) in 152 children with brain trauma. Using binary logistic regression to evaluate the simultaneous effects of these three biomarkers on outcome, they described a 77% correct classification rate and a positive predictive value of 75%. However, no biomarker prognostic threshold was described in their report (Berger et al., 2007). In two recent independent reports, Berger and colleagues described the usefulness of combined serum biomarker levels to respectively diagnose pediatric brain trauma (Berger et al., 2005) and inflicted brain injury among well infants with GCS of 15 (Berger et al., 2009). Neither of these studies investigated outcome predictive values using combinations of biomarkers(Berger et al., 2005, 2009). Our study is, therefore, the first to report brain trauma outcome prediction using biomarkers from differentmediator families, and to describe prognostic thresholds for severalpaired biomarker combinations. We used MultiROC curves for our analyses, because they retain all the simplicity of interpretations of the traditional ROC curve analysis but additionally allow comparisons between the performance of multivariate combinations without being restricted to the display of a single variable's performance and comparisons of individual tests (Shultz, 1995). Our chosen "screening thresholds" were based upon the highest concentration of each biomarker that identified as many patients with favorable recovery without including any of those with unfavorable outcome. We did not know at the outset of the study whether these thresholds would offer the highest predictive values but have subsequently demonstrated that arbitrary increasing or decreasing the original threshold levels by 10% and 20% did not improve the outcome predictive values of these combinations. Using the MultiROC curves, we further refined our prognostic algorithm by defining the optimal threshold for the "varying biomarker" within each combination, and were able to propose for the first time pairs of biomarker levels that were highly predictive of unfavorable outcome. Combinations using a brain-specific protein (S100b or NSE) as the "screening marker" had higher predictive values for unfavorable outcome than those that used either of these proteins as the "varying marker." On the other hand, L-selectin offered a higher outcome predictive value when it was used as the "varying marker" rather than the "screening marker" within the prognostic algorithm. One possible explanation for these observations was that the "screening marker" thresholdswe chose for S100b or NSE were more sensitive in screening for patients with unfavorable outcome than that chosen for L-selectin. Alternatively, although elevated brain-specific proteins levels signaled glial and neuronal damage, their relatively short serum half-life meant that their serum levels at the time of sampling in our cohort might have been better employed as a screening tool for unfavorable outcome. The main limitation of our study is relatively small cohort size and the few patients with unfavorable outcome resulting in the need toassess outcome simply, rather than using more elegant measures such as detailed cognitive tests (Beers et al., 2007). Another limitation is that our findings may not be used to predict outcome of brain trauma patients with extra-cranial injuries who may have greatly altered serum cytokine profiles independent of the brain injury, because our cohort only includes patients with isolated TBIs. Despite these limitations, we have successfully defined prognostic thresholds for five individual biomarkers and combinations of paired biomarker levels withvery high outcome predictive values. Our proposed prognostic thresholds and their outcome predictive values were not validated because ofthe small sample size. Our findings may, however, be used as starting points for validation in future studies involving larger and independent cohorts before translation into potentially useful clinical tools for brain trauma outcome predictions. Conclusion Combining brain-specific protein and inflammatory mediator levels offer higher outcome predictive values than may be achieved with the individual biomarkers. Author Disclosure Statement The authors declare that there is no potential conflict of interest. References Adelson, P.D., Clyde, B., Kochanek, P.M., Wisniewski, S.R., Marion, D.W., and Yonas, H. (1997). Cerebrovascular response in infants andyoung children followin g severe traumatic brain injury: a preliminary report. Pediatr. Neurosurg. 26, 200-207. Arand, M., Melzner, H., Kinzl, L., Bruckner, U.B., and Gebhard, F.(2001). Early inflammatory mediator response following isolated traumatic brain injury and other major trauma in humans. Langenbecks Arch. Surg. 386, 241-248. Bandyopadhyay, S., Hennes, H., Gorelick, M.H., Wells, R.G., and Walsh-Kelly, C.M. (2005). Serum neuron-specific enolase as a predictor of short-term outcome in children with closed traumatic brain injury.Acad. Emerg. Med. 12, 732-738. Beers, S.R., Berger, R.P., and Adelson, P.D. (2007). Neurocognitive outcome and serum biomarkers in inflicted versus non-inflicted traumatic brain injury in young children. J. Neurotrauma 24, 97-105. Bell, M.J., Kochanek, P.M., Doughty, L.A., Carcillo, J.A., Adelson, P.D., Clark, R.S., Wisniewski, S.R., Whalen, M.J., and DeKosky, S.T. (1997). Interleukin-6 and interleukin-10 in cerebrospinal fluid after severe traumatic brain injury in children. J. Neurotrauma 14, 451-457. Berger, R.P., Adelson, P.D., Pierce, M.C., Dulani, T., Cassidy, L.D., and Kochanek, P.M. (2005). Serum neuron-specific enolase, S100B, and myelin basic protein concentrations after inflicted and noninflicted traumatic brain injury in children. J. Neurosurg. 103, 61-68. Berger, R.P., Adelson, P.D., Richichi, R., and Kochanek, P.M. (2006). Serum biomarkers after traumatic and hypoxemic brain injuries: insight into the biochemical response of the pediatric brain to inflicted brain injury. Dev. Neurosci. 28, 327-335. Berger, R.P., Beers, S.R., Richichi, R., Wiesman, D., and Adelson,P.D. (2007). Serum biomarker concentrations and outcome after pediatric traumatic brain injury. J. Neurotrauma 24, 1793-1801. Berger, R.P., Ta'asan, S., Rand, A., Lokshin, A., and Kochanek, P.(2009). Multiplex assessment of serum biomarker concentrations in well-appearing children with inflicted traumatic brain injury. Pediatr.Res. 65, 97-102. Bruce, D.A., Schut, L., Bruno, L.A., Wood, J.H., and Sutton, L.N. (1978). Outcome following severe head injuries in children. J. Neurosurg. 48, 679-688. Buttram, S.D., Wisniewski, S.R., Jackson, E.K., Adelson, P.D., Feldman, K., Bayir, H., Berger, R.P., Clark, R.S., and Kochanek, P.M. (2007). Multiplex assessment of cytokine and chemokine levels in cerebrospinal fluid following severe pediatric traumatic brain injury: effects of moderate hypothermia. J. Neurotrauma 24, 1707-1717. Chambers, I.R., Jones, P.A., Lo, T.Y., Forsyth, R.J., Fulton, B., Andrews, P.J., Mendelow, A.D., and Minns, R.A. (2006). Critical thresholds of intracranial pressure and cerebral perfusion pressure related to age in paediatric head injury. J. Neurol. Neurosurg. Psychiatry 77, 234-240. Chambers, I.R., Treadwell, L., and Mendelow, A.D. (2000). The cause and incidence of secondary insults in severely head-injured adults and children. Br. J. Neurosurg. 14, 424-431. Chesnut, R.M., Marshall, L.F., Klauber, M.R., Blunt, B.A., Baldwin, N., Eisenberg, H.M., Jane, J.A., Marmarou, A., and Foulkes, M.A. (1993). The role of secondary brain injury in determining outcome from severe head injury. J. Trauma 34, 216-222. Chiaretti, A., Genovese, O., Aloe, L., Antonelli, A., Piastra, M.,Polidori, G., and Di Rocco, C. (2005). Interleukin 1beta and interleukin 6 relationship with paediatric head trauma severity and outcome.Childs Nerv. Syst. 21, 185-194. Csuka, E., Morganti-Kossmann, M.C., Lenzlinger, P.M., Joller, H., Trentz, O., and Kossmann, T. (1999). IL-10 levels in cerebrospinal fluid and serum of patients with severe traumatic brain injury: relationship to IL-6, TNF-alpha, TGF-beta1 and blood-brain barrier function.J. Neuroimmunol. 101, 211-221. Engberg, A., and Teasdale, T.W. (1998). Traumatic brain injury in children in Denmark: a national 15-year study. Eur. J. Epidemiol. 14,165-173. Ergun, R., Bostanci, U., Akdemir, G., Beskonakli, E., Kaptanoglu, E., Gursoy, F., and Taskin, Y. (1998). Prognostic value of serum neuron-specific enolase levels after head injury. Neurol. Res. 20, 418-420. Facco, E., Zuccarello, M., Pittoni, G., Zanardi, L., Chiaranda, M., Davia, G., and Giron, G.P. (1986). Early outcome prediction in severe head injury: comparison between children and adults. Childs Nerv. Syst. 2, 67-71. Fridriksson, T., Kini, N., Walsh-Kelly, C., and Hennes, H. (2000).Serum neuron-specific enolase as a predictor of intracranial lesionsin children with head trauma: a pilot study. Acad. Emerg. Med. 7, 816-820. Jackson, R.G., Samra, G.S., Radcliffe, J., Clark, G.H., and Price,C.P. (2000). The early fall in levels of S-100 beta in traumatic brain injury. Clin. Chem. Lab. Med. 38, 1165-1167. Jones, P.A., Andrews, P.J., Easton, V.J., and Minns, R.A. (2003). Traumatic brain injury in childhood: intensive care time series data and outcome. Br. J. Neurosurg. 17, 29-39. Kalabalikis, P., Papazoglou, K., Gouriotis, D., Papadopoulos, N., Kardara, M., Papageorgiou, F., and Papadatos, J. (1999). Correlation between serum IL-6 and CRP levels and severity of head injury in children. Intensive Care Med. 25, 288-292. Kossmann, T., Hans, V.H., Imhof, H.G., Stocker, R., Grob, P., Trentz, O., and Morganti-Kossmann, C. (1995). Intrathecal and serum interleukin-6 and the acute-phase response in patients with severe traumatic brain injuries. Shock 4, 311-317. Kossmann, T., Stahel, P.F., Lenzlinger, P.M., Redl, H., Dubs, R.W., Trentz, O., Schlag, G., and Morganti-Kossmann, M.C. (1997). Interleukin-8 released into the cerebrospinal fluid after brain injury is associated with blood-brain barrier dysfunction and nerve growth factorproduction. J. Cereb. Blood Flow Metab. 17, 280-289. Kushi, H., Saito, T., Makino, K., and Hayashi, N. (2003). IL-8 is a key mediator of neuroinflammation in severe traumatic brain injuries. Acta Neurochir. Suppl. 86, 347-350. Marangos, P.J., and Schmechel, D.E. (1987). Neuron-specific enolase, a clinically useful marker for neurons and neuroendocrine cells. Annu. Rev. Neurosci. 10, 269-295. McClain, C., Cohen, D., Phillips, R., Ott, L., and Young, B. (1991). Increased plasma and ventricular fluid interleukin-6 levels in patients with head injury. J. Lab. Clin. Med. 118, 225-231. McKeating, E.G., Andrews, P.J., and Mascia, L. (1998). The relationship of soluble adhesion molecule concentrations in systemic and jugular venous serum to injury severity and outcome after traumatic brain injury. Anesth. Analg. 86, 759-765. Miller, J.D., and Becker, D.P. (1982). Secondary insults to the injured brain. J. R. Coll. Surg. Edinb. 27, 292-298. Minambres, E., Cemborain, A., Sanchez-Velasco, P., Gandarillas, M., Diaz-Reganon, G., Sanchez-Gonzalez, U., and Leyva-Cobian, F. (2003). Correlation between transcranial interleukin-6 gradient and outcomein patients with acute brain injury. Crit. Care Med. 31, 933-938. Pal, J., Brown, R., and Fleiszer, D. (1989). The value of the Glasgow Coma Scale and Injury Severity Score: predicting outcome in multiple trauma patients with head injury. J. Trauma 29, 746-748. Persson, L., Hardemark, H.G., Gustafsson, J., Rundstrom, G., Mendel-Hartvig, I., Esscher, T., and Pahlman, S. (1987). S-100 protein andneuron-specific enolase in cerebrospinal fluid and serum: markers ofcell damage in human central nervous system. Stroke 18, 911-918. Pleines, U.E., Stover, J.F., Kossmann, T., Trentz, O., and Morganti-Kossmann, M.C. (1998). Soluble ICAM-1 in CSF coincides with the extent of cerebral damage in patients with severe traumatic brain injury. J. Neurotrauma 15, 399-409. Ponsky, T.A., Eichelberger, M.R., Cardozo, E., Huang, Z.J., Pratsch, G.L., Thuma-Croom, S.E., and Newman, K.D. (2005). Analysis of headinjury admission trends in an urban American pediatric trauma center. J. Trauma 59, 1292-1297. Prasad, M.R., Ewing-Cobbs, L., Swank, P.R., and Kramer, L. (2002).Predictors of outcome following traumatic brain injury in young children. Pediatr. Neurosurg. 36, 64-74. Raabe, A., Grolms, C., and Seifert, V. (1999a). Serum markers of brain damage and outcome prediction in patients after severe head injury. Br. J. Neurosurg. 13, 56-59. Raabe, A., Grolms, C., Sorge, O., Zimmermann, M., and Seifert, V. (1999b). Serum S-100B protein in severe head injury. Neurosurgery 45,477-483. Reilly, P.L., Simpson, D.A., Sprod, R., and Thomas, L. (1988). Assessing the conscious level in infants and young children: a paediatric version of the Glasgow Coma Scale. Childs Nerv. Syst. 4, 30-33. Ross, S.A., Cunningham, R.T., Johnston, C.F., and Rowlands, B.J. (1996). Neuron-specific enolase as an aid to outcome prediction in head injury. Br. J. Neurosurg. 10, 471-476. Rothoerl, R.D., Woertgen, C., Holzschuh, M., Metz, C., and Brawanski, A. (1998). S-100 serum levels after minor and major head injury. J. Trauma 45, 765-767. Shultz, E.K. (1995). Multivariate receiver-operating characteristic curve analysis: prostate cancer screening as an example. Clin. Chem. 41, 1248-1255. Skogseid, I.M., Nordby, H.K., Urdal, P., Paus, E., and Lilleaas, F. (1992). Increased serum creatine kinase BB and neuron specific enolase following head injury indicates brain damage. Acta Neurochir. (Wien) 115, 106-111. Spinella, P.C., Dominguez, T., Drott, H.R., Huh, J., McCormick, L., Rajendra, A., Argon, J., McIntosh, T., and Helfaer, M. (2003). S-100beta protein-serum levels in healthy children and its association with outcome in pediatric traumatic brain injury. Crit. Care Med. 31, 939-945. Suominen, P., Kivioja, A., Ohman, J., Korpela, R., Rintala, R., and Olkkola, K.T. (1998). Severe and fatal childhood trauma. Injury 29,425-430. Takahashi, H., Sato, H., and Tsuji, Y. (1989). Biochemical markersin the acute stage of head injury-aldosterone and CKBB. Neurol. Med.Chir. (Tokyo) 29, 192-195. Thurman, D., and Guerrero, J. (1999). Trends in hospitalization associated with traumatic brain injury. JAMA 282, 954-957. Walker, P.J., and Cass, D.T. (1987). Paediatric trauma: urban epidemiology and an analysis of methods for assessing the severity of trauma in 598 injured children. Aust. N. Z. J. Surg. 57, 715-722. Whalen, M.J., Carlos, T.M., Kochanek, P.M., Wisniewski, S.R., Bell, M.J., Carcillo, J.A., Clark, R.S., DeKosky, S.T., and Adelson, P.D.(1998). Soluble adhesion molecules in CSF are increased in children with severe head injury. J. Neurotrauma 15, 777-787. Whalen, M.J., Carlos, T.M., Kochanek, P.M., Wisniewski, S.R., Bell, M.J., Clark, R.S., DeKosky, S.T., Marion, D.W., and Adelson, P.D. (2000). Interleukin-8 is increased in cerebrospinal fluid of children with severe head injury. Crit. Care Med. 28, 929-934. Woertgen, C., Rothoerl, R.D., Holzschuh, M., Metz, C., and Brawanski, A. (1997). Comparison of serial S-100 and NSE serum measurements after severe head injury. Acta Neurochir. (Wien) 139, 1161-1165. Yamazaki, Y., Yada, K., Morii, S., Kitahara, T., and Ohwada, T. (1995). Diagnostic significance of serum neuron-specific enolase and myelin basic protein assay in patients with acute head injury. Surg. Neurol. 43, 267-270. Yang, Y., Huang, W., and Lu, X. (2002). Early changes of endothelin, nitric oxide and arginine-vasopressin in patients with acute cerebral injury. Chin. J. Traumatol. 5, 259-262. Address correspondence to: Tsz-Yan M. Lo, MRCPCH 218 Queens Quay West, Suite 2504 Toronto, M5J 2Y6, Ontario, Canada E-mail: mils.lo@doctors.org.uk Tsz-Yan M. Lo, Patricia A. Jones, and Robert A. Minns Child Life and Health, University of Edinburgh, Edinburgh, United Kingdom. Table 1. Commercially Available ELISA Used in the Study Brain-specific proteins Glial protein S-100b (Nexus Dx[TM] S-100 Test Kit, Synx Pharma Inc.) Neuronal protein NSE (Nexus Dx[TM] NSE Test Kit, Synx Pharma Inc.) Inflammatory mediators Pro-inflammatory mediator IL-6 (IL-6 ELISA Kit, Diaclone Research) Chemotaxin IL-8 (IL-8 ELISA Kit, Diaclone Research) Anti-inflammatory mediator IL-10 (IL-10 ELISA Kit, Diaclone Research) Adhesion molecules L-selectin (human sL-Selectin Immunoassay, R&D Systems) SICAM (sICAM-1 ELISA Kit, Diaclone Research) Vasoconstrictor Endothelin (Endothelin 1-21 Test Kit, Biomedica) TABLE 2. DEMOGRAPHIC DETAILS AND OUTCOME OF THE COHORT Favorable Unfavorable outcome outcome Median age, years 7.92 10.5 (range, (range, 0.33-14.17) 2.33-13.42) Sex distribution Boys 19 2 Girls 5 2 Post-resuscitation GCS (primary brain injury severity) GCS 13-15 2 0 (non-severe) GCS 9-12 8 1 (non-severe) GCS 3-8 (severe) 14 3 Initial CT brain findings Diffuse injury 15 4 Focal injury 9 0 TABLE 3. OPTIMAL CUT-FF VALUE INDIVIDUAL BIOMAKERS AND TIBER SENSITIVITY AND SPECIFICITY FOR UNFAVORABLE OUTCOME PREDICTION Prediction of unfavorable outcome at 6 months post-injury Optimal cut-off value on ROC AUC Specificity Sensitivity L-selectin day 1200 ng/ml 0.92 88% 75% 1 level IL-8 day 1 level 30 pg/ml 0.88 92% 75% NSE day 1 level 25 ng/ml 0.83 83% 75% S100B day 1 level 0.05 ng/ml 0.83 79% 75% IL-6 day 1 level 60 pg/ml 0.83 71% 75% TABLE 4. TOP 10 PAIRED BIOMARKER LEVELS FOR UNFAVORABLE OUTCOME PREDICTION, THE OPTIMAL PROGNOSTIC THRESHOLD FOR THEIR "VARYING MARKER," AUCM, SPECIFICITY, AND SENSITIVITY Prediction of unfavorable outcome at 6 months post-injury Paired biomarkers Optimal cut-off ("screening marker"> value for the AUCm (area "screening threshold" "varying marker" under the AND "varying marker") on MutliROC MultiROC) Specificity S100b > 0.04 ng/ml 1000 ng/ml 0.98 96% AND L-selectin S100b > 0.04 ng/ml 40 pg/ml 0.98 96% AND IL-6 S100b > 0.04 ng/ml 10 ng/ml 0.97 92% AND NSE IL-6 > 30 pg/ml AND 1000 ng/ml 0.96 88% L-selectin NSE > 12 ng/ml AND 1200 ng/ml 0.95 96% L-selectin NSE > 12 ng/ml AND 0.05 ng/ml 0.94 92% S100b IL-6 > 30 pg/ml AND 30 pg/ml 0.93 96% IL-8 S100b > 0.04 ng/ml 30 pg/ml 0.93 92% AND IL-8 IL-8 > 20 pg/ml AND 1200 ng/ml 0.92 88% L-selectin L-selectin > 700 30 pg/ml 0.92 96% ng/ml AND IL-8 Prediction of unfavorable outcome at 6 months post-injury Paired biomarkers ("screening marker"> "screening threshold" AND "varying marker") Sensitivity S100b > 0.04 ng/ml 100% AND L-selectin S100b > 0.04 ng/ml 100% AND IL-6 S100b > 0.04 ng/ml 100% AND NSE IL-6 > 30 pg/ml AND 100% L-selectin NSE > 12 ng/ml AND 75% L-selectin NSE > 12 ng/ml AND 75% S100b IL-6 > 30 pg/ml AND 75% IL-8 S100b > 0.04 ng/ml 75% AND IL-8 IL-8 > 20 pg/ml AND 75% L-selectin L-selectin > 700 75% ng/ml AND IL-8 TABLE 5. AREA UNDER EACH MULTIVARIATE ROC CURVES (AUCM) USING ORIGINAL "SCREENING THRESHOLDS," AND THE ORIGINAL THRESHOLD LEVELS PLUS AND MINUS 20% AUCm AUCm AUCm (original (original (original screening screening Screening Varying screening threshold threshold marker marker threshold) plus 20%) minus 20%) S100b L-selectin 0.98 0.94 0.94 S100b IL-6 0.98 0.94 0.88 S100b NSE 0.97 0.96 0.89 IL-6 L-Selectin 0.96 0.95 0.95 NSE L-selectin 0.95 0.93 0.93 NSE S100b 0.94 0.93 0.84 IL-6 IL-8 0.93 0.91 0.89 S100b IL-8 0.93 0.90 0.89 IL-8 L-selectin 0.92 0.90 0.89 L-selectin IL-8 0.92 0.91 0.90 L-selectin S100b 0.92 0.88 0.87 IL-6 NSE 0.91 0.87 0.78 IL-8 IL-6 0.91 0.89 0.89 NSE IL-8 0.90 0.88 0.88 IL-6 S100b 0.90 0.89 0.89 IL-8 S100b 0.89 0.84 0.82 NSE IL-6 0.88 0.86 0.83 IL-8 NSE 0.88 0.84 0.80 L-selectin IL-6 0.88 0.86 0.86 L-selectin NSE 0.83 0.83 0.75