Journal Logo

Feature Articles

Biomarkers for Estimating Risk of Hospital Mortality and Long-Term Quality-of-Life Morbidity After Surviving Pediatric Septic Shock: A Secondary Analysis of the Life After Pediatric Sepsis Evaluation Investigation*

Wong, Hector R. MD1; Reeder, Ron W. PhD2; Banks, Russell MS2; Berg, Robert A. MD3; Meert, Kathleen L. MD4; Hall, Mark W. MD5; McQuillen, Patrick S. MD6; Mourani, Peter M. MD7; Chima, Ranjit S. MD1; Sorenson, Samuel BS2; Varni, James W. PhD8; McGalliard, Julie BA9; Zimmerman, Jerry J. MD, PhD9; for the Eunice Kennedy Shriver National Institute of Child Health and Human Development (NICHD) Collaborative Pediatric Critical Care Research Network (CPCCRN) and the Life After Pediatric Sepsis Evaluation (LAPSE) Investigators

Author Information
Pediatric Critical Care Medicine: January 2021 - Volume 22 - Issue 1 - p 8-15
doi: 10.1097/PCC.0000000000002572

Abstract

Pediatric sepsis represents a key public health burden—every 3–5 s, someone in the world dies of sepsis (1). Most epidemiologic and descriptive studies have focused on sepsis-associated mortality, with much less attention given to longer term sepsis-associated morbidity experienced by children who survive sepsis. Recently, the Life After Pediatric Sepsis Evaluation (LAPSE) investigation prospectively evaluated long-term mortality and morbidity among children admitted to PICUs with septic shock and reported critical illness risk factors associated with these adverse outcomes (2,3). Significant health-related quality-of-life (HRQL) impairment compared with baseline status persisted 1 year after hospitalization in one-third of the LAPSE cohort.

The Pediatric Sepsis Biomarker Risk Model (PERSEVERE) is a multibiomarker stratification tool for estimating a reliable baseline risk of mortality from pediatric septic shock (4–7). The PERSEVERE biomarkers are serum proteins measured during the first 24 hours following a septic shock diagnosis. They were originally identified using unbiased, discovery-oriented transcriptomic studies seeking to identify gene expression programs and pathways associated with poor outcome (8). PERSEVERE has been validated internally (4,7,9) but has not been tested in an external cohort nor has it been further evaluated as a tool for estimating the risk of poor HRQL among children surviving septic shock.

The current study represents a secondary analysis of a subset of the LAPSE cohort for whom enrollment blood samples were obtained for the measurement of the PERSEVERE biomarkers. This investigation included three broad goals. First, we tested the performance of PERSEVERE for estimating the risk of hospital mortality in the LAPSE cohort. Second, we examined the association between the PERSEVERE baseline mortality risk and the composite outcome of mortality or deterioration of persistent, serious deterioration of HRQL (PSD-HRQL) below baseline among the survivors of septic shock. Third, we used the PERSEVERE biomarkers to derive a new model to estimate the risk of PSD-HRQL at 3 months among children surviving septic shock.

MATERIALS AND METHODS

LAPSE Cohort

The LAPSE study protocol was approved by the Institutional Review Boards (IRBs) of the 12 participating institutions, either locally or through a central IRB, including approval for this secondary analysis involving the PERSEVERE biomarkers. Extensive details of the study protocol were previously published (2,3). Briefly, after signed informed consent from a parent or legal guardian, children 1 month to 18 years old with community-acquired septic shock were enrolled in this prospective longitudinal observational cohort outcome investigation to quantify sepsis-associated long-term mortality and morbidity among survivors.

Morbidity was assessed primarily using developmentally appropriate measures of HRQL, namely, the Pediatric Quality-of-Life Inventory (10–12) and the Stein Jessop Functional Status Scale (13) at study entry (to reflect preillness status), study day 7, and 1, 3, 6, and 12 months following the initial admission for septic shock. The protocol used prespecified criteria to dichotomize the presence of PSD-HRQL, at each study point, relative to preillness status.

Three predefined outcomes from the original LAPSE investigation were evaluated in the current analyses: all-cause hospital mortality, and the composite outcome of mortality or PSD-HRQL (> 25% below baseline) among surviving children at 1 or 3 months following admission at a PICU for a sepsis event. Blood samples were obtained for the measurement of the PERSEVERE biomarkers in a subset of enrolled patients. These were obtained within 24 hours of enrollment and 48 hours later (day 3 samples), consistent with the original derivation and subsequent internal validation of PERSEVERE (4,6,9). PERSEVERE data reflecting the day 1 samples were used for all analyses in the current study, unless otherwise specified. Obtaining blood samples was optional for each of the participating study sites. All blood samples were obtained from remnant blood in the clinical laboratories.

PERSEVERE Biomarkers and Inhospital Mortality Risk Stratification

The PERSEVERE biomarkers include C-C chemokine ligand 3 (CCL3), interleukin 8 (IL8), heat shock protein 70 kDa 1B, granzyme B, and matrix metallopeptidase 8 (6). Serum concentrations of these biomarkers were measured using a multiplex magnetic bead platform designed for the PERSEVERE research program by the EMD Millipore Corporation (Billerica, MA). Biomarker concentrations were measured using a Luminex 100/200 System (Luminex Corporation, Austin, TX), according to the manufacturers’ specifications. Assay performance data were previously published (6).

Each eligible LAPSE subject was classified according to the PERSEVERE decision tree using predefined biomarker-based decision rules and one age-based decision rule (Supplemental Fig. 1, Supplemental Digital Content 1, http://links.lww.com/PCC/B545; legend, Supplemental Digital Content 3, http://links.lww.com/PCC/B547). The baseline mortality probability for an individual subject reflects assignment to one of the eight terminal nodes (TNs) of the PERSEVERE decision tree. These were the PERSEVERE-based mortality probabilities used for analyses.

Data Analysis

Performance of PERSEVERE for estimating the risk of hospital mortality was evaluated by constructing a receiver operating characteristic curve and calculating the 95% CIs of the area under the curve. Diagnostic test characteristics, with 95% CIs, were calculated by constructing a 2 × 2 contingency table, wherein the PERSEVERE mortality probability was dichotomized to reflect “predicted dead” or “predicted alive,” relative to the actual hospital mortality. Based on the PERSEVERE decision tree rules, subjects with a PERSEVERE mortality probability less than or equal to 0.025 were classified as “predicted alive,” whereas those with a PERSEVERE mortality probability greater than or equal to 0.182 were classified as “predicted dead,” as previously reported (6,7).

We used logistic regression to measure further the association between the PERSEVERE-based mortality probability (independent variable) and the three outcomes listed above (dependent variables). Because PERSEVERE reflects day 1 of a septic shock diagnosis, the other independent variables were selected based on this temporal limit. These included the Pediatric Risk of Mortality (PRISM) score, Version IV (14), the maximum day 1 Pediatric Logistic Organ Dysfunction (PELOD) score, Version 2 (15), and the maximum day 1 Vasoactive-Inotrope Score (VIS) (16). The independent variables were first tested using univariable analyses. Those with a significant association with the outcome of interest (p < 0.05) were subsequently included in the corresponding multivariable analyses.

Classification and Regression Tree Modeling

Classification and regression tree analysis (Salford Predictive Modeler, Version 8.2; Minitab, State College, PA) was used to derive a model estimating the risk of the dichotomous outcome variable, PSD-HRQL at 3 months following admission at the PICU for the septic shock event, among those who survived to hospital discharge. The candidate-predictor variables considered in the modeling procedures included the days 1 and 3 PERSEVERE biomarker concentrations, age, maximum day 1 PELOD score, and day 1 platelet count. None of the candidate-predictor variables were weighted. TNs that did not improve classification and TNs that contained less than 5% of the subjects in the root node were pruned. Performance of the tree was evaluated by calculating the area under the receiver operating characteristic (AUROC) curve and calculating diagnostic test characteristics. The tree was further evaluated using a 10-fold cross validation procedure, calculating a summary AUROC curve.

Results

Study Cohort

Among the 389 subjects enrolled in the original LAPSE cohort, 173 (44%) had blood samples available for the measurement of the PERSEVERE biomarkers. Supplemental Table 1 (Supplemental Digital Content 2, http://links.lww.com/PCC/B546) compares the clinical and demographic characteristics of this subgroup with that of the remaining LAPSE subjects without PERSEVERE data. The subjects with PERSEVERE data tended to be older and were less medically complex at baseline, relative to those without PERSEVERE data. No other differences were noted.

Performance of PERSEVERE in the LAPSE Subgroup

The 173 LAPSE subjects with PERSEVERE data were classified according to the predefined rules of the PERSEVERE decision tree to test the ability of PERSEVERE to estimate the risk of hospital mortality (Supplemental Fig. 1, Supplemental Digital Content 1, http://links.lww.com/PCC/B545; legend, Supplemental Digital Content 3, http://links.lww.com/PCC/B547). PERSEVERE risk estimation generated an AUROC curve of 0.73 (95% CI, 0.59–0.87; p = 0.002) for distinguishing between the hospital survivors and nonsurvivors in the LAPSE cohort. Table 1 summarizes the diagnostic test characteristics of PERSEVERE for predicting hospital mortality.

TABLE 1. - Diagnostic Test Characteristics of Pediatric Sepsis Biomarker Risk Model for Predicting Hospital Mortality in the Life After Pediatric Sepsis Evaluation Cohort (n = 173)
Variable Value 95% CI
Area under the curve 0.73 0.59–0.87
True positives, n 13
True negatives, n 106
False positives, n 49
False negatives, n 5
Sensitivity 72% 46–89
Specificity 68% 60–75
Positive predictive value 21% 12–34
Negative predictive value 95% 89–98
(+) Likelihood ratio 2.3 1.6–3.3
(–) Likelihood ratio 0.4 0.2–0.9

Association of PERSEVERE With Outcome

The association between the PERSEVERE baseline mortality risk and the outcomes was further assessed using logistic regression. Three separate outcomes were evaluated: inhospital mortality (n = 173) and the composite outcome of mortality or PSD-HRQL assessed at 1 month (n = 125) or 3 months (n = 117). The other independent variables considered in the logistic regression procedures were PRISM, maximum PELOD on day 1, and maximal VIS on day 1 (Table 2).

TABLE 2. - Univariable and Multivariable Analyses of Candidate Variable-Associated With Outcome
Outcome Variable n OR (95% CI) p OR (95% CI) p
Univariable Analysis Multivariable Analysis
In-hospital mortality PERSEVEREa 173 1.9 (1.3–2.7) < 0.001 1.7 (2.5–4.1) 0.010
PRISM 1.1 (1.0–1.1) 0.021 1.0 (1.0–1.1) 0.638
PELOD 1.2 (1.1–1.4) < 0.001 1.2 (1.0–1.4) 0.053
VIS 1.0 (1.0–1.0) 0.050 1.0 (1.0–1.0) 0.748
Mortality or PSD-HRQLb at 1 mo PERSEVEREa 125 1.4 (1.0–1.9) 0.039 1.2 (0.9–1.7) 0.194
PRISM 1.0 (1.0–1.1) 0.389
PELOD 1.1 (1.0–1.2) 0.030 1.1 (1.0–1.2) 0.143
VIS 1.0 (1.0–1.0) 0.762
Mortality or PSD-HRQLb at 3 mo PERSEVEREa 117 1.6 (1.1–2.2) 0.005 1.4 (1.0–2.0) 0.051
PRISM 1.0 (1.0–1.1) 0.242
PELOD 1.2 (1.1–1.3) 0.002 1.1 (1.0–1.3) 0.016
VIS 1.0 (1.0–1.0) 0.121
OR = odds ratio, PELOD = Pediatric Logistic Organ Dysfunction, PERSEVERE = Pediatric Sepsis Biomarker Risk Model, PRISM = Pediatric Risk of Mortality, PSD-HRQL = persistent, serious deterioration of health-related quality of life, VIS = Vasoactive-Inotrope Score.
aThe raw PERSEVERE mortality probability was transformed by a factor of 10 for the logistic regression analyses, such that the PERSEVERE-related ORs reflect a 0.1 increase in the PERSEVERE mortality probability.
bPSD-HRQL > 25% below baseline.

Among the 173 LAPSE subjects with PERSEVERE data, 18 (10%) died during hospitalization. All four independent variables were associated with hospital mortality in univariable analyses. When the four independent variables were considered in a multivariable analysis, only PERSEVERE was independently associated with hospital mortality. PELOD was marginally associated with hospital mortality in the multivariable analysis.

Among the 125 LAPSE subjects with PERSEVERE data and HRQL data at 1 month, 49 (39%) also had the composite outcome of mortality or PSD-HRQL. PERSEVERE and PELOD were associated with this composite adverse outcome in univariable analyses. However, when both variables were considered in a multivariable analysis, neither was independently associated.

Among the 117 LAPSE subjects with PERSEVERE data and HRQL data at 3 months, 37 (32%) also had the composite outcome of mortality or PSD-HRQL. PERSEVERE and PELOD were associated with this composite outcome in univariable analyses. However, when both variables were considered in a multivariable analysis, only PELOD was independently associated with this composite outcome. PERSEVERE was marginally associated with this composite outcome in the multivariable analysis.

To determine if hospital mortality was the primary factor influencing the association between PERSEVERE and the composite adverse outcomes, the univariable analyses were repeated after excluding subjects who did not survive to hospital discharge. Among 109 subjects who survived to hospital discharge, 33 (30%) had PSD-HRQL at 1 month. PERSEVERE was not associated with this measure of HRQL morbidity in this subgroup of subjects who survived to hospital discharge (OR, 1.1; 95% CI, 0.7–1.5; p = 0.766). Among 99 subjects who survived to hospital discharge, 19 (19%) had PSD-HRQL at 3 months. PERSEVERE was not associated with this measure of HRQL morbidity at 3 months in this subgroup who survived to hospital discharge (OR, 1.2; 95% CI, 0.8–1.8; p = 0.375).

Estimating the Risk of Poor HRQL Outcome Among Survivors of Sepsis

Since the association between PERSEVERE and the composite poor outcomes was primarily influenced by hospital mortality, a new decision tree was derived to estimate the risk of PSD-HRQL at 3 months among septic shock survivors. Figure 1 shows the derived decision tree, which includes the day 1 CCL3 concentration, the day 1 IL8 concentration, and age. None of the other candidate-predictor variables added to predictive capacity. The decision tree consists of three low-risk TNs (TN1, TN3, and TN4; probability of poor outcome was 0.000–0.059) and two high-risk TNs (TN2 and TN5; probability of poor outcome was 0.320–0.692). The decision tree had an AUROC curve of 0.87 (95% CI, 0.80–0.95). Ten-fold cross validation of the decision tree yielded a summary AUROC curve of 0.74. Table 3 summarizes the diagnostic test characteristics of the decision tree for predicting PSD-HRQL at 3 months among children who survive septic shock.

TABLE 3. - Diagnostic Test Characteristics of Newly Derived Model to Estimate the Risk of Persistent, Serious Deterioration Health-Related Quality of Life at 3 mo (n = 99)
Variable Value 95% CI
AUC 0.87 0.80–0.95
10-fold cross validation AUC 0.74
True positives, n 17
True negatives, n 59
False positives, n 21
False negatives, n 2
Sensitivity 89% 65–98
Specificity 74% 63–83
Positive predictive value 45% 29–62
Negative predictive value 97% 88–99
(+) Likelihood ratio 3.4 2.3–5.1
(–) Likelihood ratio 0.1 0.04–5.3
AUC = area under the curve.

Figure 1.
Figure 1.:
The derived decision tree for estimating the risk of persistent, serious deterioration of health-related quality of life (PSD-HRQL) at 3 mo, among subjects who survived to hospital discharge. The primary outcome is dichotomized as “poor” or “good” to reflect whether the study subject had PSD-HRQL deterioration at 3 mo after enrollment. The root node contains all subjects (n = 99) and provides the number of subjects with good or poor outcome, and the respective rates. Subsequent to the root node, subjects are allocated to daughter nodes according to decision rules reflecting either a biomarker concentration (pg/mL) or age. Each daughter node provides the decision rule used to generate the respective daughter nodes, and the number of subjects with good or bad outcome, and the respective rates, which correspond to the respective probabilities of good or poor outcome. The decision tree contains two C-C chemokine ligand 3 (CCL3)-based decision rules, one interleukin 8-based decision rule, and one age-based decision rule. The assignment of risk probability for good or poor outcome is based on allocation to one of the five terminal nodes (TNs). Subjects allocated to TN1, TN3, and TN4 have a low probability of poor outcome (0.000–0.059), whereas subjects allocated to TN2 and TN5 have a higher probability of poor outcome (0.320–0.692). These probabilities are used to calculate the area under the receiver operating characteristic curve and the diagnostic test characteristics.

DISCUSSION

This study represents the first external validation of PERSEVERE. External validation is critical to assessing generalizability and reliability of stratification models. The LAPSE cohort reflects patients with community-acquired septic shock, whereas the original derivation and testing of PERSEVERE involved children who acquired septic shock in both the outpatient and inpatient settings, and who were enrolled with no exclusion criteria other than failure to obtain informed consent. In this context, PERSEVERE had modest performance in the LAPSE cohort. We note that there were five subjects who did not survive to hospital discharge but who were classified by PERSEVERE as having a low risk of mortality. Among these five false-negative subjects, four were allocated to TN2 of the PERSEVERE decision tree (Supplemental Table 1, Supplemental Digital Content 2, http://links.lww.com/PCC/B546). This is a low-risk TN with a baseline mortality risk estimate of 0.011. A majority of LAPSE subjects (59%) were allocated to TN2, yielding an overall mortality rate of 0.039 among subjects allocated to TN2. This actual mortality rate is not overly discordant with the PERSEVERE-estimated mortality risk of 0.011.

When we considered PERSEVERE, PRISM, PELOD, and VIS in a multivariable logistic regression, PERSEVERE was the only variable independently associated with hospital mortality, suggesting that PERSEVERE might provide mortality risk information beyond that provided by well-established clinical and physiologic parameters. We note, however, that these results are in contrast to previous studies, showing that PELOD and PRISM performed well among children with sepsis (17,18). Whether these discrepancies reflect our small sample size, the fact that LAPSE enrollment was limited to community-acquired septic shock or other unknown variables requires further exploration.

This study also assessed the association between the PERSEVERE baseline mortality risk and PSD-HRQL, as originally measured and defined in the LAPSE investigation (2,3). The PERSEVERE baseline mortality risk was associated with poor outcome, defined as the composite outcome of mortality or PSD-HRQL. However, this association was no longer apparent when subjects who did not survive to hospital discharge were excluded, suggesting that PERSEVERE per se has limited utility for estimating the risk of long-term HRQL morbidity among children surviving sepsis.

It nonetheless stands to reason that the biological pathways associated with increased risk of mortality from pediatric septic shock, as reflected by the PERSEVERE biomarkers, might also be associated with increased risk of long-term HRQL morbidity. We tested this possibility by deriving a new model to estimate the risk of PSD-HRQL at 3 months following admission for the sepsis event. The new model had excellent test characteristics and demonstrated acceptable performance upon cross validation.

Among the various candidate-predictor variables considered, the newly derived model included the day 1 concentrations of CCL3 and IL8, and age, indicating that these variables have the strongest association with increased risk of poor, long-term HRQL. These same variables contribute to the original PERSEVERE decision tree (4–6). CCL3 concentrations inform the first-level decision rule of both PERSEVERE and the current model for estimating the risk of sepsis-associated long-term HRQL morbidity. Collectively, these similarities suggest biological commonalities in the pathways for both mortality and poor HRQL outcome among children with community-acquired septic shock.

We note the limitations of this study, including that it is a secondary analysis. Due to the availability of blood samples, less than 50% of the original LAPSE cohort was included in the analysis. The resulting small sample size is at risk for generating unreliable estimates of the diagnostic test characteristics. This also raises the possibility of selection bias, although the differences between LAPSE participants with and without PERSEVERE blood samples were minimal (Supplemental Table 1, Supplemental Digital Content 2, http://links.lww.com/PCC/B546). In addition, although the newly derived model performed well upon cross validation, it nonetheless requires prospective testing in an independent cohort that also includes a broader population of patients not limited to those with community-acquired septic shock. Finally, we were unable to analyze HRQL outcomes at 1 year after surviving sepsis due to a limited sample size.

CONCLUSIONS

In summary, PERSEVERE had modest performance for estimating hospital mortality in an external cohort of children with community-acquired septic shock. The PERSEVERE biomarkers, measured early during the acute phase of septic shock, appear to have utility for estimating the risk of PSD-HRQL up to 3 months after surviving sepsis, albeit with a different decision tree than the PERSEVERE mortality prediction model. This ancillary investigation suggests an opportunity to develop a clinical tool for early assignment of HRQL morbidity risk, and therefore enable more efficient and targeted implementation of interventions to mitigate the risk of HRQL morbidity and to target rehabilitation among children surviving septic shock.

REFERENCES

1. Weiss SL, Fitzgerald JC, Pappachan J, et al.; Sepsis Prevalence, Outcomes, and Therapies (SPROUT) Study Investigators and Pediatric Acute Lung Injury and Sepsis Investigators (PALISI) Network. Global epidemiology of pediatric severe sepsis: The sepsis prevalence, outcomes, and therapies study. Am J Respir Crit Care Med. 2015; 191:1147–1157
2. Zimmerman JJ, Banks R, Berg RA, et al.; Life After Pediatric Sepsis Evaluation (LAPSE) Investigators. Trajectory of mortality and health-related quality of life morbidity following community-acquired pediatric septic shock. Crit Care Med. 2020; 48:329–337
3. Zimmerman JJ, Banks R, Berg RA, et al.; Life After Pediatric Sepsis Evaluation (LAPSE) Investigators. Critical illness factors associated with long-term mortality and health-related quality of life morbidity following community-acquired pediatric septic shock. Crit Care Med. 2020; 48:319–328
4. Wong HR, Cvijanovich NZ, Anas N, et al. Pediatric sepsis biomarker risk model-II: Redefining the pediatric sepsis biomarker risk model with septic shock phenotype. Crit Care Med. 2016; 44:2010–2017
5. Wong HR, Cvijanovich NZ, Anas N, et al. Improved risk stratification in pediatric septic shock using both protein and mRNA biomarkers. PERSEVERE-XP. Am J Respir Crit Care Med. 2017; 196:494–501
6. Wong HR, Salisbury S, Xiao Q, et al. The pediatric sepsis biomarker risk model. Crit Care. 2012; 16:R174
7. Wong HR, Caldwell JT, Cvijanovich NZ, et al. Prospective clinical testing and experimental validation of the pediatric sepsis biomarker risk model. Sci Transl Med. 2019; 11:eaax9000
8. Kaplan JM, Wong HR. Biomarker discovery and development in pediatric critical care medicine. Pediatr Crit Care Med. 2011; 12:165–173
9. Wong HR, Weiss SL, Giuliano JS Jr, et al. Testing the prognostic accuracy of the updated pediatric sepsis biomarker risk model. PLoS One. 2014; 9:e86242
10. Varni JW, Burwinkle TM, Seid M, et al. The PedsQL 4.0 as a pediatric population health measure: Feasibility, reliability, and validity. Ambul Pediatr. 2003; 3:329–341
11. Aspesberro F, Fesinmeyer MD, Zhou C, et al. Construct validity and responsiveness of the Pediatric Quality of Life Inventory 4.0 generic core scales and infant scales in the PICU. Pediatr Crit Care Med. 2016; 17:e272–e279
12. Varni JW, Limbers CA, Neighbors K, et al. The PedsQL™ infant scales: Feasibility, internal consistency reliability, and validity in healthy and ill infants. Qual Life Res. 2011; 20:45–55
13. Stein RE, Jessop DJ. Functional status II®. A measure of child health status. Med Care. 1990; 28:1041–1055
14. Pollack MM, Holubkov R, Funai T, et al.; Eunice Kennedy Shriver National Institute of Child Health and Human Development Collaborative Pediatric Critical Care Research Network. The pediatric risk of mortality score: Update 2015. Pediatr Crit Care Med. 2016; 17:2–9
15. Leteurtre S, Duhamel A, Salleron J, et al.; Groupe Francophone de Réanimation et d’Urgences Pédiatriques (GFRUP). PELOD-2: An update of the PEdiatric logistic organ dysfunction score. Crit Care Med. 2013; 41:1761–1773
16. McIntosh AM, Tong S, Deakyne SJ, et al. Validation of the vasoactive-inotropic score in pediatric sepsis. Pediatr Crit Care Med. 2017; 18:750–757
17. Matics TJ, Sanchez-Pinto LN. Adaptation and validation of a pediatric sequential organ failure assessment score and evaluation of the sepsis-3 definitions in critically ill children. JAMA Pediatr. 2017; 171:e172352
18. Leclerc F, Leteurtre S, Duhamel A, et al. Cumulative influence of organ dysfunctions and septic state on mortality of critically ill children. Am J Respir Crit Care Med. 2005; 171:348–353
Keywords:

biomarkers; enrichment; morbidity; outcome; epsis

Supplemental Digital Content

Copyright © 2020 by the Society of Critical Care Medicine and the World Federation of Pediatric Intensive and Critical Care Societies