Does the White Cell Matter in septic shock?

Canadian Critical Care Forum, 2019November 13, 2019

Ryan Zarychanski MD MSc FRCPC

Divisions of Critical Care & Hematology/Medical Oncology

Manitoba, Canada

Disclosures

None

Objectives

1. Review limitations of prognostic tools

2. Introduce group-based trajectory analysis, a method to identify variability in a patient’s illness or response to treatment

3. Present research research results exploring trajectories of white cells in patients with septic shock

Background

• Infection is the 3rd leading cause of preventable mortality1

• Incidence of septic shock ~3/1000 per year2

• 2016 Sepsis-3 definition of sepsis:

– life-threatening organ dysfunction due to a dysregulated host response to infection

1. Mathers CD et al. Global and regional causes of death. Br Med Bull. 2009;92(7-32) 2. Martin et al. The epidemiology of sepsis in the United States from 1979 through 2000. NEJM. 2003;348(16):1546-54.

Pathophysiology

• Complex host response

• Balance of pro-inflammatory and anti-inflammatory responses1

• WBC play an important role

–Pro-inflammatory phase (neutrophils/monocytes)

–Anti-inflammatory phase (lymphocytes)

Prognosis – baseline characteristics

• APACHE II score

–12 physiologic variables, max score 71

–Most abnormal value in first 24 hours of admission

Knaus et al. APACHE II: A severity of disease classification system. Crit Care Medcine. 1985;13(818-29)

• SOFA score

• 6 organ systems

• Score 0-4, worse value per day used

• Possible score 0-24

– Score >15 associated with mortality of 90%

– Increase in SOFA score over first 48 hrs associated with >50% mortality

Ferreira F, et al.. Serial evaluation of the sofa score to predict outcome in critically ill patients. JAMA. 2001;286(14):1754-8.

Prognosis – baseline characteristics

Limitations of current scoring systems

• Assessments are typically static, based on the first 24 hours of ICU admission

• Fail to consider clinical phenotypes, individual genotypes, or response to treatment

Precision Medicine

• An emerging approach for disease treatment and prevention that takes into account individual variability in genes, environment, and lifestyle

• Identify “like” groups

1. Inflammatory biomarkers

2. Genetic testing or genetic array

Metabolic profiles in SIRS (white) and sepsis (black)

Crit Care Med 2016; 44:1649–1662

Genetic Array in pediatric sepsis

• Genes corresponding to glucocorticoid receptor signalling

• 28-day mortality 22% (subclass A) vs. 10% (subclass B), p<0.05

Am J Respir Crit Care Med Vol 191, Iss 3, pp 309–315, Feb 1, 2015

Precision Medicine

• An emerging approach for disease treatment and prevention that takes into account individual variability in genes, environment, and lifestyle

• Identify “like” groups1. Inflammatory biomarkers

2. Genetic testing or genetic array

3. Trajectory analysis

Trajectory analysis

• Describes the pattern of a dependent variable over time

–unique subgroups exist within a population that follow distinct trajectories over time

• Data that evolves over time, such as the temporal trend of the complete blood count

Nagin and Odgers. Group-Based Trajectory Modeling in Clinical Research. Annu Rev Clin Psychol. 2010;6:109-38.

Recidivism trajectories in sex-offenders

• Population separated into risk categories

• Unique, modifiable characteristics of group membership can be explored

Jennings W et al. Journal of Crime and Justice. Vol. 35, No. 3, November 2012, 356–364

Trajectory analysis of HIV virologic outcomes

• Identify and predict HIV ‘career’ trajectories

• Models incorporate social and SES variables

A primer on trajectory analysis

• PROC TRAJ – SAS procedure that estimates multiple groups in a population

• Identifies clusters of individuals following similar progressions over time by fitting a group based model

• Subjects are grouped and assumed that everyone in group follows same trajectory

Nagin DS. Group-Based Modeling of Development. Cambridge, Massachusetts: Harvard University Press; 2005

Methods – Trajectory analysis

1. Determine the maximum number of groups

2. Fit number of groups to data

– Start at one, then 2, then 3, and so on

3. Select the maximum shape of the pattern of change for each group over time (linear or other)

– Can model up to 5th order polynomials

Methods – Model choice and diagnostics

• Model choice based on:

– Bayesian Information Criterion (BIC)• Log-likelihood adjusted for number of parameters and sample size

– Clinical judgment

– Parsimony

• Model diagnostics– Ratio between the probability of group membership to the actual group

assignment

– Average posterior probability

– Odds of correct classification

White Blood Cell Count and Septic Shock

• CBC very commonly ordered test in ICU

• WBC integrates many parameters that are pertinent to the host response

–e.g. granulopoiesis, mobilization of white cells from the marrow reserve, margination and egress into tissues, and apoptosis

• Specific changes in WBC over time in septic shock are not well described

White blood cell count derangements and outcomes septic shock• First 48 hours1

– Initial leukocytosis due to neutrophilia

– Lymphocyte count low

• Persistent lymphopenia associated with poor outcome2

• Baseline WBC <4 associated with increased mortality3

–OR 1.85 (95%CI 1.38-2.48)1. Venet F, et al. Early assessment of leukocyte alterations at diagnosis of septic shock. Shock. 2010;34(4):358-632. Le Tulzo Y, et al. Early Circulating Lymphocyte Apoptosis in Human Septic Shock is Associated with Poor Outcome. Shock. 2002;18(6):487-94.3. Kumar A. Critical Care Medicine. 2006;34(Suppl):A103#377.

Limitations of existing data

• Modeling the Average WBC

–Might not be representative of ANYONE in the population

• Value on first day

• Pre-determined groups based on expert opinion

• Small sample sizes

• Studies are not contemporaneous and to not reflect current standards of care or mortality rates

Hypotheses

• There are distinct clinical of individuals with septic shock with different disease biology, genetics or response to treatment that can be identified using trajectory analysis

• Identified trajectories will be able to predict mortality

Objectives our study

• Identify subgroups of patients with septic shock with differing patterns of WBC counts over time

• Identify characteristics of patients and pathogens associated with the unique WBC count trajectories identified

• Evaluate the association between the unique WBC count trajectories and 30-day mortality in patients with septic shock

Study Design: Retrospective Cohort (n = 917)

• Inclusion criteria:– Adult patients with septic shock at HSC or SBH

– Identified from CATSS database

– Linked to LIS for complete study records

• Exclusion criteria:– Malignancy, HIV, immunosuppression, liver disease, neutropenia

– ICU length of stay <48 hours

– <3 WBC measurements

– Re-admission for septic shock within study period

Results: Trajectory analysis of WBC count

1 2 3 4 5 6 7

Day

WBC

18

16

14

12

10

20

Mean WBC overall recorded every 12 hours.

Summary of WBC trajectory

Group Number Summary of WBC trajectory

1 Normal, flat

2 Normal, rising

3 Moderate, gradual decline

4 High, gradual decline

5 High, rising

6 High, rapid decline

7 Significant elevation

Selecting the best model using the BICGroup BIC (0bservation) (BIC subjects)

1

2 -24834.8 -24818.6

3 -24176.8 -24152.34 -23919.4 -23886.95 -23710.1 -23669.46 -23677.5 -23628.67 -23499.2 -23442.2• 2*(Δ BIC) = 2* (-23499.2-(-23677.5)) = 356.6

• Strong evidence to support group 7

Testing model robustness

Trajectory j (%) Actual(%) ratio j odds AvgPP AvgPP odds OCC

1 35 35.4 0.99 0.54 0.88 7.6 14.1

2 7.9 6.2 0.82 0.09 0.85 5.6 64.9

3 34.6 35.8 0.97 0.53 0.83 4.8 9.0

4 14 14.1 0.99 0.16 0.87 6.7 40.7

5 3.1 2.9 0.94 0.03 0.95 18.4 572.5

6 4.5 4.6 0.98 0.05 0.95 19.3 411.0

7 0.8 0.8 1 0.01 1.00 172412.8 2251348

Final model – 7 groups

• Face validity: meaningful and potentially expected trajectories for patients with septic shock

• BIC value: less negative than the preceding groups

• Predicted group assignment with high probability

Methods - Multinomial regression

• Evaluate patient, illness and pathogen characteristics association with trajectory group

• Trajectory 1 reference group

• Multivariable model

–Age, sex, APACHEII*WBC, bacteremia, time to antibiotics, use of appropriate antibiotics, number of organ failures at time of ICU admission and baseline platelet count

Creation of new variable APACHEII*WBC

• If WBC >=40 or <= 1, then subtract 4 from APACHE II score

• If WBC 20-39.9 or 1-2.9, then subtract 2 from APACHE II score

• If WBC 15-19.9, then subtract 1 from APACHE II score

• If WBC 3-14.9, then do not alter the APACHE II score

Results - Multinomial regression1

Normal, flat(ref)

2Normal, rising

3Moderate, gradual

decline

4High, gradual

decline

5High, rising

6High, rapid decline

7Significantly

elevated

Predictor OR (95% CI)

Age 0.99 (0.97, 1.01) 1.01 (1, 1.02)* 1 (0.98, 1.01) 0.98 (0.96, 1.01) 1 (0.98, 1.02) 0.96 (0.91, 1.01)

Female sex 0.89 (0.48, 1.65) 1.26 (0.9, 1.75) 1.87 (1.18, 2.97)* 2.22 (0.89, 5.53) 1.94 (0.92, 4.08) 6.79 (0.72, 64.33)

APACHE II score (excl WBC) 1.02 (0.97, 1.07) 0.99 (0.96, 1.02) 1.02 (0.98, 1.05) 1.05 (0.98, 1.12) 1 (0.95, 1.06) 1.03 (0.91, 1.18)

Time to 1st antibiotic (hours) 1 (0.98, 1.02) 1.01 (0.99, 1.02) 0.99 (0.97, 1.01) 0.99 (0.95, 1.03) 0.96 (0.9, 1.03) 1.03 (1, 1.05)*

Appropriate Antibiotic Use 1.26 (0.41, 3.82) 1.73 (0.9, 3.31) 1 (0.43, 2.31) 0.83 (0.19, 3.62) 3.13 (0.37, 26.56) 0.96 (0.08, 11.56)

Number of organ failures on day 1 1.21 (0.96, 1.52) 1.06 (0.93, 1.2) 1.17 (0.98, 1.4) 1.38 (0.99, 1.93) 1.38 (1.05, 1.82)* 1.68 (0.88, 3.2)

Bacteremia 0.98 (0.46, 2.1) 1.51 (1.01, 2.26)* 1.32 (0.76, 2.31) 1.4 (0.51, 3.88) 1.86 (0.81, 4.24) 0.8 (0.08, 7.96)

Platelet count (per 10 increase) 2.01 (0.94, 4.31) 2.78 (1.8, 4.31)* 4.67 (2.75, 7.93)* 2.89 (1.07, 7.84)* 4.94 (2.35, 10.36)* 3.81 (0.72, 20.15)

Group Trajectory and mortality outcomes: Cox proportional hazard model

• Outcome: 30-day mortality

• Variables in multivariable model

– age, sex, APACHE II*WBC, number of organ failures on day 1 of ICU admission, bacteremia, culture positive or negative, time to first antibiotic, the provision of appropriate antibiotics, combination antibiotics

• Time zero = admission to ICU

Mortality

1 1 1 2 2 2 3 3 3 4 4 4 5 5 5

6 6 6 7 7 7

• Unadjusted overall 30-day mortality 26.3%

23

.4

31

.5

26

.1

23

.1

63

28

.6

28

.6

1 2 3 4 5 6 7

30

-DA

Y M

OR

TALI

TY (

%)

GROUP P=0.0013

Mortality - Cox proportional hazard model

Group Hazard Ratio (95% CI) p-value

1 reference reference

2 1.46 (0.84, 2.54) 0.184

3 1.13 (0.81, 1.58) 0.464

4 0.84 (0.51, 1.38) 0.486

5 3.48 (1.91, 6.35) <0.0001

6 1.43 (0.7, 2.92) 0.329

7 1.44 (0.35, 6.02) 0.617

Adjusted for age, sex, APACHE*WBC, co-morbidities, type of infection, time to antibiotics, appropriate antibiotics

Strengths

• Novel method of analysis to demonstrate individual variability in sepsis

• Large database; consecutive patients

• Excluded patients with alternative explanation of WBC count abnormalities

Beyond an unmeaningful mean…

1 2 3 4 5 6 7

Day

WBC

18

16

14

12

10

20

Mean WBC overall recorded every 12 hours.

Limitations

• Proof of concept

• Retrospective

–Unmeasured confounders

• Diverse but finite explanatory variables included

–e.g. missing: Frailty, immune responses, genomics

• Large database but small overall numbers in the trajectory groups

Conclusions – Septic shock

• Group-based trajectory analyses can segregate patients into distinct and clinically relevant WBC trajectories

• Conventionally recognized patient, treatment and illness characteristics poorly predict trajectory assignment

• A rising WBC is independently associated increased risk of death at 30-days

What’s next?

• Validate in larger database

• Trajectory of neutrophils and lymphocytes

• Trajectory of platelet count, bilirubin

• Create a multivariable trajectory model

What’s the vision?

• Real-time integration of comprehensive health data to create dynamic patient trajectories

– Inform prognosis

–Response to treatment

• When to escalate or de-escalate

”Sepsis and ARDS are biologically heterogenous syndromes: they are not diseases per se with singular mechanisms that are plausibly amenable to singular interventions”

Laffey JG, and Kavanagh BP. Negative trials in critical care: why most research is probably wrong. The Lancet Respiratory Medicine. 2018;6(9):659-60.

Beyond Critical Care

• Application of trajectory analysis to other areas of medicine are many:

–Changes in tumor markers

• e.g. M-protein in myeloma or PSA in prostate cancer

–Changes in functional ability

Acknowledgements: Acute Care Hematology Research Cluster

Ryan ZarychanskiEmily RimmerAnand KumarAlexis TurgeonDonald HoustonSylvain Lother

Allan GarlandBrett HoustonSteve DoucetteMurdoch LeeiesChantalle MenardGloria Vazquez-Grande

Thank you

Ryan Zarychanski MD MSc FRCPC

rzarychanski@cancercare.mb.ca