The Genomics of Septic Shock

The Genomics of Septic Shock

Hector R. Wong, MDDivision of Critical Care Medicine

Cincinnati Children’s Hospital Medical CenterCincinnati Children’s Hospital Research Foundation

1st International Symposium on AKI in Children at the 7th International Conference on Pediatric Continuous

Renal Replacement TherapySeptember 2012

Disclosures

• The Cincinnati Children’s Hospital Research Foundation and the Speaker have submitted patent applications for biomarker-based stratification model presented in this lecture.

• The Speaker serves on the Scientific Advisory Board for DxTerity and is compensated with stock options.

Nine years of genome-level expression profiling in pediatric septic shock…..

Discovery-oriented, exploratory genome-wide expression studies in children with septic shock

FUNDAMENTAL OBSERVATIONS REGARDING THE GENOME-LEVEL BIOLOGY OF PEDIATRIC SEPTIC SHOCK

DISCOVERY OF NOVEL CANDIDATE THERAPEUTIC TARGETS

DISCOVERY OF GENE EXPRESSION-BASED CLASSES OF SEPTIC SHOCK WITH CLINICALLY RELEVANT PHENOTYPIC DIFFERENCES

DISCOVERY OF NOVEL BIOMARKERS• STRATIFICATION• DIAGNOSIS

Discovery-oriented, exploratory genome-wide expression studies in children with septic shock

FUNDAMENTAL OBSERVATIONS REGARDING THE GENOME-LEVEL BIOLOGY OF PEDIATRIC SEPTIC SHOCK

DISCOVERY OF NOVEL CANDIDATE THERAPEUTIC TARGETS

DISCOVERY OF GENE EXPRESSION-BASED CLASSES OF SEPTIC SHOCK WITH CLINICALLY RELEVANT PHENOTYPIC DIFFERENCES

DISCOVERY OF NOVEL BIOMARKERS• STRATIFICATION• DIAGNOSIS

Nine years of genome-level expression profiling in pediatric septic shock…..

Stratification

• Early assessment (i.e. within 24 hours of admission) of who is at risk for good or poor outcome.

Why Do We Care?

• Reliable outcome risk stratification is fundamental for effective clinical practice and clinical research.

• The oncology paradigm.• Stratification for clinical trials.• Informing individual patient decision making.• Allocation of ICU resources.• Quality metric.• There is no reliable and validated outcome risk

stratification tool for septic shock.

Discovery of candidate stratification biomarkers for septic shock

Mining of genome-wide expression data to identify genes associated with 28-day mortality in children with septic shock.

117 genes with predictive capacity for mortality

12 gene products meeting the following criteria:• Biological plausibility regarding sepsis biology.• Gene product (i.e. protein) can be measured in serum/plasma.

Final list of candidate stratification biomarkersGene Symbol DescriptionCCL3 C-C chemokine ligand 3; a.k.a. MIP-1LCN2 Lipocalin 2; a.k.a. NGALMMP8 Matrix metallopeptidase 8; a.k.a. neutrophil collagenaseRETN ResistinTHBS Thrombospondin 1GZMB Granzyme BHSPA1B Heat shock protein 70kDa 1BCCL4 C-C chemokine ligand 4; a.k.a. MIP-1IL8 Interleukin-8LTF LactotransferrinELA2 Neutrophil elastase 1IL1A Interleukin 1

PERSEVERE

• PEdiatRic SEpsis biomarkEr Risk modEl.• Multi-biomarker-based risk model to predict

outcome in septic shock.

Derivation of PERSEVERE

• 220 patients with septic shock.• 10.5% mortality.• Measured 12 candidate stratification

biomarkers from serum.• Serum samples represent the first 24 hours of

admission to the PICU.• “CART” analysis.

CART Analysis

• Classification and Regression Tree.• Decision tree building technique.• “Binary recursive partitioning”.• Binary: splitting of patients into 2 groups.• Recursive: can be done multiple times.• Partitioning: entire dataset split into sections.• Has the potential to reveal complex interactions

between candidate predictor variables not evident using traditional approaches.

Derivation Cohort CART Analysis Results Overview

• Included 5 of the 12 candidate biomarkers.– CCL3: MIP-1α– Heat shock protein-70– IL-8– Elastase– NGAL

• 5 decision rules• 10 daughter nodes

HSP70 ≤ 3310450 N = 181Outcome Number Rate

Death 6 0.033Survived 175 0.967

ROOT N = 220Outcome Number Rate


CCL3 ≤ 358 N = 195Outcome Number Rate


CCL3 > 358 N = 25Outcome Number Rate


HSP70 > 3310450 N = 14Outcome Number Rate


Elastase ≤ 344596 N = 24Outcome Number Rate


IL8 ≤ 356 N = 133Outcome Number Rate


IL8 > 356 N = 48Outcome Number Rate


Elastase > 344596 N = 24Outcome Number Rate


NGAL > 8712 N = 10Outcome Number Rate


NGAL ≤ 8712 N = 14Outcome Number Rate


Derivation Cohort Tree

































































































































































Low Risk Terminal NodesN = 171























High Risk Terminal NodesN = 49

Non-survivor Survivor

Predicted Non-Survivor

Predicted Survivor

Test characteristics based on terminal nodes. All subjects in low risk nodes predicted as survivors. All subjects in high risk nodes predicted as non-survivors.


Predicted Non-Survivor 21 28

Predicted Survivor 2 169

Sensitivity91%

CI 70 to 98%

Specificity86%

CI 80 to 80%

PPV 43% (CI 29 to 58%)+LR 6.4 (CI 4.5 to 9.3)

NPV 99% (CI 95 to 100%)-LR 0.10 (CI 0.03 to 0.4)

Test characteristics based on terminal nodes. All subjects in low risk nodes predicted as survivors. All subjects in high risk nodes predicted as non-survivors.

AUC = 0.885

Testing PERSEVERE

• 135 different patients with septic shock.• 13.3% mortality.• Measured the same candidate biomarkers.• “Dropped the patients through the tree”.



Predicted Survivor

Test characteristics in the test cohort




Sensitivity89%

CI 64 to 98%

Specificity64%

CI 55 to 73%

PPV 28% (CI 17 to 41%)+LR 2.5 (CI 1.8 to 3.3)

NPV 97% (CI 90 to 99%)-LR 0.18 (CI 0.05 to 0.69)

Test characteristics in the test cohort

AUC = 0.759

Updating PERSEVERE using the combined derivation and test cohorts (n = 355).

Updated Model

• Included 3 of the 5 candidate biomarkers from the initial model.– CCL3: MIP-1α– Heat shock protein-70– IL-8

• Eliminated 2 of the 5 candidate biomarkers from the original model.– Elastase– NGAL

• Added granzyme B, MMP-8, & age as decision rules.• 7 decision rules.• 14 daughter nodes.

HSPA1B ≤ 3.27E6 N = 207Outcome Number Rate








HSPA1B > 3.27E6 N = 27Outcome Number Rate










GZMB > 55 N = 36Outcome Number Rate


GZMB ≤ 55 N = 30Outcome Number Rate


MMP8 > 47513 N = 15Outcome Number Rate


MMP8 ≤ 47513 N = 40Outcome Number Rate


Age ≤ 0.5 years N = 8Outcome Number Rate


Age > 0.5 years N = 22Outcome Number Rate
































High risk terminal nodesN = 119Death risk: 18.2 to 62.5%































Low risk terminal nodesN = 236Death risk: 0.0 to 2.5%



Predicted Survivor

Test characteristics of updated model




Test characteristics of updated model

Sensitivity93%

CI 79 to 98%

Specificity74%

CI 69 to 79%

PPV 32% (CI 24 to 41%)+LR 3.6 (CI 2.9 to 4.4)

NPV 99% (CI 96 to 100%)-LR 0.1 (CI 0.0 to 0.3)

AUC = 0.883




Biologically Plausible?

False Positives

True Negatives

False positives should be “sicker” than true negatives.




Persistence of ≥2 organ failures at 7 days after ICU admission

False Positives: 30%

True Negatives: 9%

P < 0.001




Median PICU Length of Stay

False Positives: 11 days

True Negatives: 7 days

P = 0.003

Potential questions you may have...

• Manuscript in press: Crit Care.• Derived an analogous model in adults.• Outperforms PRISM.• Have evaluated the performance of the updated

tree in 54 new patients (13% mortality).– Correctly predicted 6 of 7 deaths (86% sensitivity).– 33 of 34 predicted survivors actually survived (97%

NPV).

Potential applications of PERSEVERE

• Stratification for clinical trials.• Inform individual patient decision making.• Allocation of ICU resources.• Quality improvement.

Acknowledgements: Contributing Centers• Natalie Cvijanovich, MD: Children’s Hospital & Research Center Oakland, Oakland, CA. • Thomas Shanley, MD: University of Michigan, C.S. Mott Children’s Hospital, Ann Arbor, MI. • Geoffrey Allen, MD: Children’s Mercy Hospitals & Clinics, Kansas City, MO. • Neal Thomas, MD: Penn State Hershey Children’s Hospital, Hershey, PA. • Robert Freishtat, MD: Children’s National Medical Center, Washington, DC. • Nick Anas, MD: Children’s Hospital of Orange County, Orange, CA. • Keith Meyer, MD: Miami Children’s Hospital, Miami, FL. • Paul Checchia, MD: Texas Children’s Hospital, Houston, TX.• Richard Lin, MD: The Children’s Hospital of Philadelphia, Philadelphia, PA. • Michael Bigham, MD: Akron Children’s Hospital, Akron, OH. • Mark Hall, MD: Nationwide Children’s Hospital, Columbus, OH. • Anita Sen, MD: New York-Presbyterian, Morgan Stanley Children’s Hospital, Columbia

University Medical Center, New York, NY. • Jeffery Nowak, MD: Children’s Hospital and Clinics of Minnesota, Minneapolis, MN.• Michael Quasney, MD, PhD: Children’s Hospital of Wisconsin, Milwaukee, WI. • Jared Henricksen, MD: Primary Children’s Medical Center, Salt Lake, UT. • Arun Chopra, MD: St. Christopher’s Hospital for Children, Philadelphia, PA.

Funding Acknowledgement

• NIH R01GM064619 • NIH RC1HL100474• NIH R01GM096994

Thank you

Documents

The Genomics of Septic Shock