Using Secondary Data in

Statistical Analysis

Charles Natanson M.D.

Critical Care Medicine Department
Clinical Center
National Institutes of Health
Bethesda, MD
 Meta-analysis Definition
• Glass 1976, “the statistical analysis of a large
collection of results from individual literature
for the purpose of integrating their respective
findings.”
• Two basic purposes:
1. Determine if similar treatment effects exist for a
therapy in independent studies to estimate a net
effect for this therapy
2. Alternatively, if treatment effects differ
substantially for a therapy among independent
studies, to examine factors that may explain
these differing effects
Education Research 1976
 Techniques of Meta-Analysis

Step One: Formulating the Question

• Validity and importance are contingent on
this step
• Poorly conceived research hypothesis will
usually lead to an analysis of dubious value
 Techniques of Meta-Analysis

Step Two: Defining Eligibility Criteria

• Protocols for study inclusion should be
prospective, systematic, and explicit
• Ideally, randomized trials similar in diagnosis,
outcome, patient characteristics, and
treatment groups
 Techniques of Meta-Analysis

Step Two: Defining Eligibility Criteria Cont’d

• Including all available studies, regardless of
size, design, or quality results in an analysis
that is broadly representative but may
compromise accuracy
• Alternatively, exclusion of poorly done
studies may increase the statistical validity
but limit the ability to generalize findings
 Techniques of Meta-Analysis

Step Three: Identifying Studies and Data

Abstraction
• Usually begins with a search of online
databases such as MEDLINE, Current
Contents, Best Evidence, Cochrane, and
HealthSTAR
• Title and abstract perused to exclude studies
 Techniques of Meta-Analysis

Step Three: Identifying Studies and Data

Abstraction Cont’d
• Full texts of the remaining articles retrieved
and thoroughly studied
• Reference lists of these articles are reviewed
• Once a study selected for inclusion, data
should be extracted by more than one
reviewer onto structured forms
 Techniques of Meta-Analysis

Step Four: Analysis

• A common measure of treatment effect must
be determined
• Fixed versus random effect model used to
combine data
 Techniques of Meta-Analysis

Step Four: Analysis Cont’d

• Cochran’s Q statistic and I2 calculated
• Consider metaregression when I2 > 30% and
P < 0.10
• Publication bias examined
– Funnel Plot
 Techniques of Meta-Analysis

Step Five: Reporting and Interpreting Results

• To improve overall quality of reporting for
meta-analysis, a checklist and a flow chart
should be constructed
• Quality of Reporting of Meta-analyses
(QUOROM) conference provides guidelines
for reporting searches, study selection,
validity assessment, data abstraction, study
characteristics and data synthesis
Lancet, 1999.
Meta-analysis of Clinical Trials in Sepsis
 Septic Shock Management

Early recognition
The right antibiotics right away
Rapid fluid resuscitation
Judicious use of vasopressors
Promptly address removable nidi
Benefits of Starting Appropriate
Antibiotic Therapy
McCabe 1962
Fried 1968
Bryant 1971
Kroger 1980
Ishpahani 1987
Chow 1991
Weinstein 1993
Vidal 1996
Schiappa 1996
Leibovici 1997
Leibovici 1997
Leibovici 1998
Kulley 1999
Ibrahim 2000
Faragofa 2003
Garnacho-Montero 2003
Horbarth 2003
Leone 2003
0.001 0.01 0.1 1 10 100 1000
Odds Ratio of Survival (95% CI)
 Benefit of Early versus Late Antibiotics
Author Year N Diagnosis
Miner 2001 171 Meningitis
Larche 2002 88 Bact/pneumonia*
Houck 2004 13,771 Pneumonia
Proulx 2005 118 Meningitis
Meehan 1997 14,069 Pneumonia
Gacouin 2002 213 Legionella
Iregui 2006 107 VAP
Lodis 2003 167 S. aureus
Kang 2003 123 P. aeruginosa

0.01 0.1 1 10 100

Harm Benefit
* Patients with cancer Odds Ratio of Survival (95% CI)
The Host Inflammatory
Response Hypothesis:
What went wrong?
Pathogenesis of Septic Shock
Host Defenses
Pathogens
Release
Infection and
Inflammatory
Toxins
Mediators

Multiorgan Failure
and Death Shock and
Injury
Human Clinical Trials of Anti-Inflammatory Therapies in Sepsis
Ranked by Size
Number of Patients Enrolled
0 1000 2000

Odds Ratio of Survival

95% CI

0.1 0.3 0.5 1 2.0 4.0 8.0

Increasing Harm Increasing Benefit
No Effect Crit Care Med 1998
Human Clinical Trials of Anti-Inflammatory Therapies in Sepsis
Ranked by Size
Number of Patients Enrolled
0 1000 2000

BAY x 1351 Anti-TNF Antibodies

BAY x 1351

BAY x 1351

MAK 195F
BN52021
Odds Ratio of Survival
95% CI
MAK 195F

CB0006
CDP571
MAK 195F

0.1 0.3 0.5 1 2.0 4.0 8.0

Increasing Harm Increasing Benefit
No Effect Crit Care Med 1998
Human Clinical Trials of Anti-Inflammatory Therapies in Sepsis
Ranked by Size
Number of Patients Enrolled
0 1000 2000

BAY x 1351 Anti-TNF Antibodies

P-55
BAY x 1351
Soluble TNF Receptors

BAY x 1351

P-55

MAK 195F

Odds Ratio of Survival

P-80 95% CI
MAK 195F

CB0006
CDP571
MAK 195F

0.1 0.3 0.5 1 2.0 4.0 8.0

Increasing Harm Increasing Benefit
No Effect Crit Care Med 1998
Human Clinical Trials of Anti-Inflammatory Therapies in Sepsis
Ranked by Size
Number of Patients Enrolled
0 1000 2000

BAY x 1351 Anti-TNF Antibodies

P-55
BAY x 1351
Soluble TNF Receptors
Antril IL-1ra
Antril

BAY x 1351

P-55

MAK 195F

Odds Ratio of Survival

P-80 95% CI
MAK 195F
Antril
CB0006
CDP571
MAK 195F

0.1 0.3 0.5 1 2.0 4.0 8.0

Increasing Harm Increasing Benefit
No Effect Crit Care Med 1998
Human Clinical Trials of Anti-Inflammatory Therapies in Sepsis
Ranked by Size
Number of Patients Enrolled
0 1000 2000

BAY x 1351 Anti-TNF Antibodies

P-55
BAY x 1351
Soluble TNF Receptors
Antril IL-1ra
Antril
BN52021
PAFra
BAY x 1351

P-55

MAK 195F
BN52021
Odds Ratio of Survival
P-80 95% CI
MAK 195F
Antril
CB0006
CDP571
MAK 195F

0.1 0.3 0.5 1 2.0 4.0 8.0

Increasing Harm Increasing Benefit
No Effect Crit Care Med 1998
Human Clinical Trials of Anti-Inflammatory Therapies in Sepsis
Ranked by Size
Number of Patients Enrolled
0 1000 2000

BAY x 1351 Anti-TNF Antibodies

P-55
BAY x 1351
Soluble TNF Receptors
Antril IL-1ra
Antril
BN52021
PAFra
BAY x 1351 Anti-Prostaglandin
P-55
Ibuprofen
MAK 195F
BN52021
Odds Ratio of Survival
P-80 95% CI
MAK 195F
Antril
CB0006
CDP571
MAK 195F
Ibuprofen
Ibuprofen

0.1 0.3 0.5 1 2.0 4.0 8.0

Increasing Harm Increasing Benefit
No Effect Crit Care Med 1998
Human Clinical Trials of Anti-Inflammatory Therapies in Sepsis
Ranked by Size
Number of Patients Enrolled
0 1000 2000

BAY x 1351 Anti-TNF Antibodies

P-55
BAY x 1351
Soluble TNF Receptors
Antril IL-1ra
Antril
BN52021
PAFra
BAY x 1351 Anti-Prostaglandin
CPO-127
Anti-Bradykinin
P-55
Ibuprofen
MAK 195F
BN52021
Odds Ratio of Survival
CPO-127
P-80 95% CI
MAK 195F
Antril
CB0006
CDP571
MAK 195F
Ibuprofen
Ibuprofen

0.1 0.3 0.5 1 2.0 4.0 8.0

Increasing Harm Increasing Benefit
No Effect Crit Care Med 1998
Human Clinical Trials of Anti-Inflammatory Therapies in Sepsis
Ranked by Size
Number of Patients Enrolled
0 1000 2000

BAY x 1351 Anti-TNF Antibodies

P-55
BAY x 1351
Soluble TNF Receptors
Antril IL-1ra
Antril
BN52021
PAFra
BAY x 1351 Anti-Prostaglandin
CPO-127
Anti-Bradykinin
P-55
Ibuprofen
MAK 195F
BN52021
Odds Ratio of Survival
CPO-127
P-80 95% CI
MAK 195F
Antril
CB0006
CDP571
MAK 195F
Ibuprofen
Ibuprofen

0.1 0.3 0.5 1 2.0 4.0 8.0

Increasing Harm Increasing Benefit
No Effect Crit Care Med 1998
 Control Arm Mortality Rates by
100
Type of Anti-inflammatory Agent
90
Mean Percent Mortality

80

70
(±SEM)

60

50

40

30

20 # of # of
Deaths Patients 276/783 76/210 218/438 602/1493 256/853 102/258
10
# of Trials N=3 N=2 N=2 N=7 N=3 N=3
0

Crit. Care Med 1997

 Preclinical Studies of
Anti-inflammatory Agents
Favors Therapy

1000
P < 0.0001
100
Beneficial
Odds Ratio

10

1
Favors Control

Patients
0.1
Harmful
0.01 Animals
25
0.001 50
100
0.0001
0.01 0.1 1 10 100
Am J Respir Crit Care Med 2002;166:1197 Control Odds of Dying
 Prospective Animal Studies of
Selected Agents
Favors Control Favors Therapy

1000
P < 0.0001 P-80 TNF soluble receptor
100 Tyrosine kinase inhibitor
Odds Ratio

10

0.1
Endotoxin
0.01 Gram-negative Rats Dogs Patients
Gram-positive 25
0.001 i.v. challenge 50
Pneumonia 100
0.0001
0.01 0.1 1 10 100
Control Group Odds of Dying
Am J Respir Crit Care Med 2002;166:1197
 Summary
Anti-Inflammatory Agents in Sepsis
• Meta-analysis:
– treatment effects are small (3%),
but statistically significant
• Meta-regression analysis:
– Efficacy dependent on risk of
death
» Beneficial at high risks of death,
» ineffective or harmful when risk
was moderate or low
 Paradox of
Corticosteroids in Sepsis

Less may have benefits,

but only in sickest patients
Corticosteroids in Sepsis

• Investigated since the 1960s

• By early 1990s, shown to be
ineffective or possibly harmful
• Renewed interest and new
trials over the last decade
 Trials of Corticosteroids in Sepsis
Author Year Published Odds ratio

95% CI

0.14 1.00 7.40 55.00

Ann Intern Med 2004; 141:47 Favors Control Favors Steroids
Clin Microbiol Infect 2009; 15: 308 Odds Ratio of Survival
 Trials of Corticosteroids in Sepsis
Author Year Published Odds ratio
Bennett 1963
Klastersky 1971 95% CI
Schumer 1976
Thompson 1976
Lucas 1984
Sprung 1984 I2 = 70%
Bone 1987
VA 1987
Luce 1988

0.14 1.00 7.40 55.00

Ann Intern Med 2004; 141:47 Favors Control Favors Steroids
Clin Microbiol Infect 2009; 15: 308 Odds Ratio of Survival
 Trials of Corticosteroids in Sepsis
Author Year Published Odds ratio
Bennett 1963
Klastersky 1971 95% CI
Schumer 1976
Thompson 1976
Lucas 1984
Sprung 1984 I2 = 0%
Bone 1987
VA 1987
Luce 1988

Short course (24 h) high dose

corticosteroids (median -
24,000 mg of hydrocortisone
equivalents) worsened survival
Summary
Pre 1989 studies (N = 8) P = 0.008

0.14 1.00 7.40 55.00

Ann Intern Med 2004; 141:47 Favors Control Favors Steroids
Clin Microbiol Infect 2009; 15: 308 Odds Ratio of Survival
 Trials of Corticosteroids in Sepsis
Author Year Published Odds ratio
Bennett 1963
Klastersky 1971 95% CI
Schumer 1976
Thompson 1976
Lucas 1984
Sprung 1984 I2 = 0%
Bone 1987
VAStress dose corticosteroids
Luce
1987
1988
(median - 1200 mg of
Bollaert
Briegel
1998
1999
Chawla 1999
hydrocortisone equivalents)
Yildiz 2002
Annane 2002
tapered over 6 days were
Confalonieri
Mussack
2005
2005
associated with improved
Oppert
Tandan
2005
2005
Rinaldi 2006
survival
Cicarelli
Sprung
2007
2008
Summary
Pre 1989 studies (N = 8) P = 0.008
Post 1997 studies (N = 12) P < 0.001

0.14 1.00 7.40 55.00

Ann Intern Med 2004; 141:47 Favors Control Favors Steroids
Clin Microbiol Infect 2009; 15: 308 Odds Ratio of Survival
 Trials of Corticosteroids in Sepsis
Author Year Published Odds ratio
Bennett 1963
Klastersky 1971 95% CI
Schumer 1976
Thompson 1976
Lucas 1984
Sprung 1984 I2 = 0%
Bone 1987
VA 1987
Luce 1988
Bollaert 1998
Briegel 1999
Chawla 1999
Yildiz 2002 I2 = 25%
Annane 2002
Confalonieri 2005
Mussack 2005
Oppert 2005
Tandan 2005
Rinaldi 2006
Cicarelli 2007
Summary Sprung 2008 Corticus Trial
Pre 1989 studies (N = 8) P = 0.008
Post 1997 studies (N = 12) P < 0.001

0.14 1.00 7.40 55.00

Ann Intern Med 2004; 141:47 Favors Control Favors Steroids
Clin Microbiol Infect 2009; 15: 308 Odds Ratio of Survival
 Trials of Corticosteroids in Sepsis
Author Year Published Odds ratio
Bennett 1963
Klastersky 1971 95% CI
Schumer 1976
Thompson 1976
Lucas 1984
Sprung 1984 I2 = 0%
Bone 1987
VA 1987
Luce 1988
Bollaert 1998
Briegel 1999
Chawla 1999
Yildiz 2002 I2 = 0%
Annane 2002
Confalonieri 2005
Mussack 2005
Oppert 2005
Tandan 2005
Rinaldi 2006
Cicarelli 2007
Sprung 2008
Summary
Pre 1989 studies (N = 8) P = 0.008
Post 1997 studies (N = 12) P < 0.001
Post 1997 studies (N = 11) P < 0.001
0.14 1.00 7.40 55.00
Ann Intern Med 2004; 141:47 Favors Control Favors Steroids
Clin Microbiol Infect 2009; 15: 308 Odds Ratio of Survival
 How is Corticus Different
from the 11 Other Trials of
Low-Dose Steroids?
 Effect of Corticosteroids During Sepsis
Dependent on the Severity of Illness
Odds Ratio of Survival
Favors Steroids

20.00
P = 0.03
7.40

2.70

1.00
Favors Control

0.37
Corticus Trial
0.14
0.1 1.0 10.0
Control Odds of Death
 Funnel Plot of Sepsis Trials of
Low Dose Steroids
6

5
Precision (1/Std Err)

0
-3 -2 -1 0 1 2 3
Log OR of death
Favors Steroids Favors Control
Corticosteroid Effect on Shock
 Low Dose Corticosteroids Trials
Reporting Shock Reversal
Authors
Briegel I2 = 0%
Bollaert
Chawla
Annane
Mussack
Tandan
CORTICUS
Overall (N = 7) P < 0.001

0.02 0.14 1.00 7.39 54.60

Favors Control Favors Corticosteroids
Odds Ratio of Shock Reversal
 Summary
• Corticosteroid effects during
sepsis depend on dose and
severity of illness
• Low dose
High steroids
dose decrease
corticosteroids
vasopressor
increaserequirements
mortality
and enhance shock reversal
Low dose corticosteroids
improve survival in
severely ill patients
 Limitations
• At present, the beneficial effects of low
doses of corticosteroids are based on
small trials (median 40 patients, IQR 41-44)
confounded by publication bias
• The largest trial of low dose cortico-
steroids (CORTICUS, n = 499) studied a
relatively low risk population
• Benefit from low dose corticosteroids has
not been confirmed in a large multicenter
trial of high risk patients
 Conclusions

• Until new data are available, the

decision to administer low dose
steroids for septic shock should
be individualized:
– Severity of illness
– Assessment of risk
 Intensive Insulin Therapy
in Patients with Sepsis
How much risk and how much benefit?
Endorsement of Glycemic Control as
Standard of Care for the Critically Ill
• JCAHO
– Core quality of care - all Medicare hospitals
• American College of Endocrinology
• Volunteer Hospital Association
– Care bundle
• Institute for Healthcare Improvement
– Sepsis bundle
– Post cardiac surgery
• Surviving Sepsis Campaign
– Sepsis bundle
Am J Resp Crit Care Med 2005; 172:1358
 Selected Baseline Characteristics
Conventional Intensive
Insulin: glucose Insulin: glucose
180 - 200 mg/dl 80 - 110 mg/dl
N = 783 N = 765
Men 557 (71%) 544 (71%)

Age (yr) 62.2 ± 13.9 63.4 ± 4.4

Reason for ICU care:

Cardiac Surgery 493 (63%) 477 (62%)
Non-cardiac indications 290 (37%) 288 (38%)

Apache II (median, IQR) 9 (7 - 13) 9 (7 - 13)

N Engl J Med 2001;345:1359
 Mortality Associated with
Conventional versus Intensive Insulin
# of Conventional Intensive ∆
Death in ICU
patients Insulin Insulin deaths
Cardiac Surgery 970 25 (5%) 10 (2%) 15
Thoracic 122 10 (18%) 5 (7.6%) 5
Other 70 6 (17%) 0 (0%) 6
Neuro, Vascular,
386 22 (11%) 20 (11%) 2
Trauma,Transplant

All patients 1548 63 (8%) 35 (5%) 28*

N Engl J Med 2001;345:1359 * P < 0.04

 Limitations
• Single center, unblinded study
• Relatively high mortality among cardiac
surgery patients in control group (5.1%)
• Immediate post-operative i.v. glucose
(200-300 g per day: ~ 2 - 3 L D10 or D20)
and early feeding (enteral or parenteral)
– Not routine care for cardiothoracic
surgery patients

N Engl J Med 2001;345:1359

 Meta-analysis of Tight Glucose Control in Critically Ill
Van den Berghe 2001 Favors Tight Glucose Favors Usual Care
Stecher 2006
Kia 2005
Grey 2004
Bilotta 2007
Bilotta 2008
RR and 95% CI
Chan 2008
Van den Berghe 2006
Fernandez 2005
Bland 2005
Oksanen 2007
Davies 1991
Walters 2006
Gray/GIST-UK 2007
Bruno/THIS 2008
Brunkhorst/VISEP
Devos/GLUCONTROL
2008
2007
27 trials of tight
Mackenzie/GLYCOGENIC
Arabi
2005
2006
glucose control
Wang
Yu
2006
2005 (N = 8315; I2 = 17%)
Mitchell 2006
DeLaRosa 2006
Farah 2007
McMullin/LOGIC 2007
Henderson/SUGAR 2005
Azevedo 2008

Sensitivity
Analysis
Overall
Modified from Wiener RS,
0.01 0.1 1 10 100
et al. JAMA 2008; 300:933 Relative Risk of Hospital Mortality
 Meta-analysis of Tight Glucose Control in Critically Ill
Van den Berghe 2001 Favors Tight Glucose Favors Usual Care
Stecher 2006
Kia 2005
Grey 2004
Bilotta 2007
Bilotta 2008
RR and 95% CI
Chan 2008
Van den Berghe 2006
Fernandez 2005
Bland 2005
Oksanen 2007
Davies 1991
Walters 2006
Gray/GIST-UK 2007
Bruno/THIS 2008
Brunkhorst/VISEP 2008
Devos/GLUCONTROL 2007
Mackenzie/GLYCOGENIC 2005
Arabi 2006
Wang 2006
Yu 2005
Mitchell 2006
DeLaRosa 2006
Farah 2007
McMullin/LOGIC 2007
Henderson/SUGAR 2005
Azevedo 2008
Glucose < 110 mg/dl 14 trials very tight control
Sensitivity
Analysis
Overall
Modified from Wiener RS,
0.01 0.1 1 10 100
et al. JAMA 2008; 300:933 Relative Risk of Hospital Mortality
 Meta-analysis of Tight Glucose Control in Critically Ill
Van den Berghe 2001 Favors Tight Glucose Favors Usual Care
Stecher 2006
Kia 2005
Grey 2004
Bilotta 2007
Bilotta 2008
RR and 95% CI
Chan 2008
Van den Berghe 2006
Fernandez 2005
Bland 2005
Oksanen 2007
Davies 1991
Walters 2006
Gray/GIST-UK 2007
Bruno/THIS 2008
Brunkhorst/VISEP 2008
Devos/GLUCONTROL 2007
Mackenzie/GLYCOGENIC 2005
Arabi 2006
Wang 2006
Yu 2005
Mitchell 2006
DeLaRosa 2006
Farah 2007
McMullin/LOGIC 2007
Henderson/SUGAR 2005
Azevedo 2008
Glucose < 110 mg/dl
Glucose < 150 mg/dl 13 trials moderate control
Sensitivity
Analysis
Overall
Modified from Wiener RS,
0.01 0.1 1 10 100
et al. JAMA 2008; 300:933 Relative Risk of Hospital Mortality
 Meta-analysis of Tight Glucose Control in Critically Ill
Van den Berghe 2001 Favors Tight Glucose Favors Usual Care
Stecher 2006
Kia 2005
Grey 2004
Bilotta 2007
Bilotta 2008
RR and 95% CI
Chan 2008
Van den Berghe 2006
Fernandez 2005
Bland 2005
Oksanen 2007
Davies 1991
Walters 2006
Gray/GIST-UK 2007
Bruno/THIS 2008
Brunkhorst/VISEP 2008
Devos/GLUCONTROL 2007
Mackenzie/GLYCOGENIC 2005
Arabi 2006
Wang 2006
Yu 2005
Mitchell 2006
DeLaRosa 2006
Farah 2007
McMullin/LOGIC 2007
Henderson/SUGAR 2005
Azevedo 2008
Glucose < 110 mg/dl
Glucose < 150 mg/dl
Sensitivity Surgical 7 trials in SICUs
Analysis
Overall
Modified from Wiener RS,
0.01 0.1 1 10 100
et al. JAMA 2008; 300:933 Relative Risk of Hospital Mortality
 Meta-analysis of Tight Glucose Control in Critically Ill
Van den Berghe 2001 Favors Tight Glucose Favors Usual Care
Stecher 2006
Kia 2005
Grey 2004
Bilotta 2007
Bilotta 2008
RR and 95% CI
Chan 2008
Van den Berghe 2006
Fernandez 2005
Bland 2005
Oksanen 2007
Davies 1991
Walters 2006
Gray/GIST-UK 2007
Bruno/THIS 2008
Brunkhorst/VISEP 2008
Devos/GLUCONTROL 2007
Mackenzie/GLYCOGENIC 2005
Arabi 2006
Wang 2006
Yu 2005
Mitchell 2006
DeLaRosa 2006
Farah 2007
McMullin/LOGIC 2007
Henderson/SUGAR 2005
Azevedo 2008
Glucose < 110 mg/dl
Glucose < 150 mg/dl
Sensitivity Surgical
Analysis Medical 8 trials in MICUs

Overall
Modified from Wiener RS,
0.01 0.1 1 10 100
et al. JAMA 2008; 300:933 Relative Risk of Hospital Mortality
 Meta-analysis of Tight Glucose Control in Critically Ill
Van den Berghe 2001 Favors Tight Glucose Favors Usual Care
Stecher 2006
Kia 2005
Grey 2004
Bilotta 2007
Bilotta 2008
RR and 95% CI
Chan 2008
Van den Berghe 2006
Fernandez 2005
Bland 2005
Oksanen 2007
Davies 1991
Walters 2006
Gray/GIST-UK 2007
Bruno/THIS 2008
Brunkhorst/VISEP 2008
Devos/GLUCONTROL 2007
Mackenzie/GLYCOGENIC 2005
Arabi 2006
Wang 2006
Yu 2005
Mitchell 2006
DeLaRosa 2006
Farah 2007
McMullin/LOGIC 2007
Henderson/SUGAR 2005
Azevedo 2008
Glucose < 110 mg/dl
Glucose < 150 mg/dl
Sensitivity Surgical
Analysis Medical
Med / Surg 12 trials in mixed ICUs
Overall
Modified from Wiener RS,
0.01 0.1 1 10 100
et al. JAMA 2008; 300:933 Relative Risk of Hospital Mortality
 Meta-analysis of Tight Glucose Control in Critically Ill
Van den Berghe 2001 Favors Tight Glucose Favors Usual Care
Stecher 2006
Kia 2005
Grey 2004
Bilotta 2007
Bilotta 2008
RR and 95% CI
Chan 2008
Van den Berghe 2006
Fernandez 2005
Bland 2005
Oksanen 2007
Davies 1991
Walters 2006
Gray/GIST-UK 2007
Bruno/THIS 2008
Brunkhorst/VISEP 2008
Devos/GLUCONTROL 2007
Mackenzie/GLYCOGENIC 2005
Arabi 2006
Wang 2006
Yu 2005
Mitchell 2006
DeLaRosa 2006
Farah 2007
McMullin/LOGIC 2007
Henderson/SUGAR 2005
Azevedo 2008
Glucose < 110 mg/dl
Glucose < 150 mg/dl
Sensitivity Surgical
Analysis Medical P = NS for all
Med / Surg
Overall
Modified from Wiener RS,
0.01 0.1 1 10 100
et al. JAMA 2008; 300:933 Relative Risk of Hospital Mortality
 Tight Glucose Control and
the Risk of Hypoglycemia

Favors Tight Control Favors Usual Care

Tight glucose
Hypoglycemia control
RR and 95% CI
increased
Glucose 7 - 8 fold the risk of
goal < 110 mg/dL
hypoglycemia
Glucose goal < 150 mg/dL (< 40 mg/dl)
independent
Surgical ICU of target glucose
Medical ICU
(< 150 or
Medical-Surgical ICU
< 110 mg/dL) or type
of ICU (medical, surgical or I2 = 0%
combined) Overall N = 15
0.01 0.1 1 10 100
Modified from Wiener RS,
et al. JAMA 2008; 300:933 Relative Risk
 NICE-Sugar Trial
Baseline Characteristics
Intensive Conventiona
Insulin l Insulin
Enrolled (N) 3054 3050
Surgical 37% 37%
Apache II > 25 31% 31%
Severe Sepsis 22% 21%
Mech Ventilator 94% 94%
N Engl J Med 2009; 360:1283
 NICE-Sugar Trial
Outcomes

In 6014 critically ill patients,

Intensive Conven tight ORglucose P-
control was associated
Insulin with hypoglycemia
-tional (95% CI) Value
and increased mortality at 90 1.14days.
Mortality (90d) 27.5% 24.9% 0.02
(1.02 - 1.28)
“On the basis of [these] results we do not
recommend use of the lower target 14.7 (81
Hypoglycemia 6.8% 0.5% <0.001
- 110 mg/dL) in critically ill patients.”
(9 - 25.9)

N Engl J Med 2009; 360:1283

Meta-analyses of Sepsis Trials
with at Least
One Significant Benefical Trial
 Summary
Anti-Endotoxin High Dose
J5 Antiserum Monoclonals Corticosteroids

*
*
*
p = 0.59 p = 0.60
p = 0.009

IL-1RA Activated Protein C Intensive Insulin

*
*
p = 0.14
p = 0.55
p = 0.32
*

* Shift from beneficial to last trial, p = 0.003

The randomized control trial minimizes
bias but does not eliminate the need for
reproducibility which is the sine qua non
(i.e. the indispensable and essential
condition) of scientific evidence