Professional Documents
Culture Documents
Partini Pudjiastuti
Child Health Department
Faculty of Medicine University of Indonesia
Background
Facts: INFORMATION EXPLOTION
information?
How do we summarize
medical information?
Traditional Approach
Expert Opinion
Narrative review articles (Literature review)
Consensus statements (group expert opinion)
Narrative Review
Systematic Review
Usually evidence-based
Systematic Review
Meta-analysis
Meta-Analysis
Definition:
A specific subset of systematic reviews that
Meta-Analysis
Definition:
Meta-analysis is a name that is given to any
Meta-analysis
Meta-Analysis
Initially developed in social sciences in mid-1960s
Adapted to medical studies in early 1980s
Initially applied to RCTs esp. when individual
10
a topic
to resolve conflicting research reports in the literature
to clarify or quantify the strengths and weaknesses of
studies on a topic
to document the need for a major clinical trial
to avoid the time and expense of conducting a clinical
trial
reported
9. to investigate variations in treatment effects through
subgroup (or stratified) analysis
10. to investigate or improve the generalizability (external
validity) of known treatment effects
Hierarchy of studies
/ Justification
Research question(s)
Objectives
Hypothesis
METHODS
Steps in meta-analysis
1. Identification:
Formulate question
Develop search strategy
Searching relevant studies
2. Selection:
3. Abstraction:
4. Analysis:
Quantitative pooling
Sensitivity analysis
Investigate heterogeneity & biases
5. Write up:
Clear presentation
1. IDENTIFICATION
The research question must be clearly defined:
Defined patient population
Clear & consistent definitions of the disease,
IDENTIFICATION - Sources
M-As use systematic, explicit search procedures (cf.
MEDLINE
4100 journals
1966 - present
EMBASE
Identification - sources
Cochrane Collaboration Controlled Trials Register
Over 160,000 trials, including abstracts (+ translations)
by subscriptions.. MSU Electronic Library database
includes
MEDLINE,
EMBASE
non-English publications
non-indexed publications
hand-search of journals
Other MEDLARS
Index Medicus
important if searching before 1966
hand-search only
Identification - steps
Search own personal files
Search electronic databases
Review titles and on-line abstracts to eliminate irrelevant
Retrieve remaining articles, review, and determine if meet
inclusion/exclusion criteria
Identification
find relevant study
eligibility!
dissertations
drug
company studies
book
chapters
non-indexed
conference
proceedings
government
reports
pre-MEDLINE
(1966)
2. SELECTION
TYPICAL INCLUSION CRITERIA:
study design (e.g., RCTs?, DBPC?, Cohort & CCS?)
setting (emergency department, outpatient, inpatient)
age (adults only, > 60 only, etc)
year of publication or conduct (esp. if technology or typical dosing
changes)
similarity of exposure or treatment (e.g., drug class, or dosage)
similarity of outcomes (case definitions)
minimum sample size or follow-up
languages?
complete vs incomplete (abstracts)
published vs fugitive?
pre-1966?
Selection Other
Issues.
Multiple publications from same study?
Include only one!
Selection process should be done independently by
at least 2 reviewers
Measure agreement (K) and resolve
discrepancies
for exclusion
Guidelines
A Good Checklist
Use it for reporting
Meta Analysis
Systematic reviews
3. ABSTRACTION
Goal: to abstract reliable, valid and bias free information from all
written sources
Should expect a degree of unreliability
intra- and inter- rater reliability is rarely if ever 100%!!
Event rate
Homogeneity (inverse of the variance)
Quality
Other factors
Abstraction
Typical process
2 independent reviewers
Practice with 2 or 3 articles to
calibrate
Use a 3rd reviewer or consensus
meeting to resolve conflicts
Measure agreement (K) and resolve
discrepancies
Abstraction investigator
bias
Abstractor may be biased in favor of (or against!)
Berlin (97):
compared blinded vs non-blinded reviews
Found discrepancy in which studies to include but
Time consuming
Probably can avoid esp. if use well defined abstraction
procedures
Abstract Data:
Create a spreadsheet (Excel)
For each study, create the following columns:
name of the study
name of the author, year published
number of participants who received intervention
number of participants who were in control arm
number who developed outcomes in intervention
number who developed outcomes in control arm
Spreadsheet Data
We created seven
columns:
trial: trial identity code
trialname: name of trial
year: year of the study
pop1: study population
deaths1: deaths in study
pop0: control population
deaths0: deaths in control
4. ANALYSIS
Many techniques
Still controversial
Ask expert (a statistician) in M-A
Focus on variability (heterogeneity)
Analysis - heterogeneity
What is Heterogeneity?
No two studies are exactly the same! Heterogeneity (or Diversity) is
the extent to which individual studies vary from one another.
Types of Heterogeneity
1. Clinical (for example, studies with different age
criteria for enrollment)
2. Methodological (variability in design for example,
parallel group vs. crossover placebo controlled trial)
3. Statistical (observed differences in treatment effects among
studies are greater than those one would expect by chance alone)
Analysis - heterogeneity
Causes of heterogeneity
Differences between studies with respect to:
Patients: diagnosis, prognostic factors, in- and exclusion
criteria, etc.
Interventions: type, dose, duration, etc.
Outcomes: type, scale, cut-off points, duration of follow-up,
etc.
Quality and methodology: randomised or not, allocation
Consequences of
heterogeneity
When the results of studies in a meta-analysis are
ANALYSIS
Focus on variability (heterogeneity)
Fixed effects model
o only intra-study, ignore interstudy var
o narrower CI
Random effects model
o considers also inter-study var
o broader CI
Meta-analysis (Forest)
plot
45
Test of sensitivity:
Was the result robust?
Random & fixed effects: similar
Exclusion of low weighted studies: similar
? Publication bias: larger sample size gave smaller
effect size
Exclusion of special cases, e.g. incomplete follow-up
Subgroup analysis
Meta-analyses typically include patients
Subgroup analysis
(cont.)
Analyse Data
Statistically
Two Graphs
Forest Plot
Funnel Plot
are
Use Statistical Packages, several choices
Summary Estimates
Mantel Haenszel OR=0.77
95% Confidence Interval
[0.72, 0.83]
Test of Heterogeneity:
Chi-square (df=21) = 31.5
P Value = 0.07
Forrest Plot
1992
1994
1995
1995
1996
1997
1999
2000
Combined
-0.3
Favor drug
Xe-xc=0
+0.3
Favor placebo
Study results
For nominal outcomes:
Presented as OR for individual study with
corresponding 95% CI
weighted value
1992
1994
1995
1995
1996
1997
1999
2000
Combined
0.1
10
OR = 1
Favor drug
Favor placebo
Forest plot
The label tells you what the comparison and
Forest plot
Each study has an ID
Forest plot
The data for each trial are here, divided into
Forest plot
Results from each trial are also given
numerically
Forest plot
This is the % weight given to each study in
Forest plot
The size of the block for each study is
Forest plot
The label above the graph tells you which
Forest plot
Forest plot
The vertical line in the middle is where
Forest plot
Forest plot
The pooled analysis is given a diamond
Forest plot
Results of the pooled analysis are also given
numerically
Forest plot
The result of a Z test with a p value are also
Forest plot
The forest plot displays the results of the Chi-
Publication bias
Published studies are not representative of all studies
that have been performed
Articles with positive findings (P < 0.05) are more
likely to be published
Hence published studies are a biased sub-set
Publication bias = systematic error of M-A that results
from using only published studies
Study Status
% (P < 0.05)
Published
138
67%
Presented Only
69
55%
Neither
78
29%
Total
285
54%
Results
Published
Registered
16
13
1.16
1.06
95% CI
1.06 1.27
0.97 1.15
P value
0.02
0.24
N
Median Survival
Ratio
Funnel Plots
Funnel plots are a device for checking for publication bias.
Each dot represents the overall effect
from one RCT.
As sample size increases, the width of
the confidence interval should
decrease.
Result should be located in a
symmetric, triangular area centered on
the overall effect for all studies.
Funnel Plots
Missing studies will manifest as an asymmetry in the
funnel plot.
Missing studies will appear
as a gap in the portion of the
funnel plot where you would
expect to find negative
studies.
The unopposed positive
studies will shift the
apparent treatment effect
(blue line) towards a larger
effect size than it really is.
Cumulative Meta-analysis
75
Passive
smoking and
lung cancer
review
Total 37 studies
Is passive
smoking
harmful?
Sub-group analysis
Criticism of Meta-analysis
The file drawer problem/Publication bias: There is always a high
likelihood for studies that did not achieve statistical significance not
to be published. This is called the file drawer problem because it is
assumed that this studies are filed away somewhere and not
accessible to the public. Therefore statistically significant studies
are more likely to be included in the meta-analysis which will result
in an overestimate of the treatment effect.
Potential solutions:
if possible contact all persons known to work in the field to
inquire about studies done, but not published
estimate how many studies it will take to reduce the treatment
effect to non-significance
Criticism of Meta-analysis
The Simpsons Paradox: A reversal in the direction of the
Criticism of Meta-analysis
Biased or flawed studies: If the studies in the meta-analysis are
Evaluating metaanalysis
Validity
Importance
Applicability
COMMENTS
Tests of Homogeneity?
Thank you
Funnel Plots
Treatment effects (for eg. lnOR, lnRR, RD) is plotted
against sample size (or other measures of precision such
as standard error or variance)
5000
10000
Sample size
4000
3000
2000
1000
0
0.01
1000
100
10
0.1
10
0.01
0.1
10
Funnel Plot
Plots the effect size against
the sample size of the study
To study a funnel plot, look at
its LOWER LEFT corner, thats
where negative or null studies
are located
If EMPTY, this indicates
PUBLICATION BIAS
x
x
x x x
x
x x x
x x x x
X
Sample
Size
(precision)
X
X
X
X
X
X
X
X
X
X
Effect Size
X
X
X
X
Example:
If N = 25, and ES = 0.6 then X = 58.2
Almost 60 unpublished negative studies would be required to negate the
meta-analysis of 25 studies.
Growth of Meta-analysis
3000
Number of Publications
2500
2000
1500
1000
500
0
93-94
94-95
95-96
96-97
97-98
98-99
99-00
2000-1
2001-2
2002-3
2003-4
Year of Publications
meta-analysis as publication type from PubMed, from years 1993 through 2004
Example
Let's say we want to know whether
streptokinase is protective for death
from acute myocardial infarction. How
should we set up a search strategy?
The Search
streptokinase[text word] OR acute
myocardial infarction[text word] produces ALL
articles that contain EITHER streptokinase OR
acute myocardial infarction anywhere in the
text inclusive, many
streptokinase [text word] AND acute
myocardial infarction [text word] will capture
only those subsets that have BOTH
streptokinase AND acute myocardial
infarction anywhere in the text restrictive,
few
Summary Estimates
Mantel Haenszel OR=0.77
95% Confidence Interval
[0.72, 0.83]
Test of Heterogeneity:
Chi-square (df=21) = 31.5
P Value = 0.07