Professional Documents
Culture Documents
MENU
IS T
Data ANALYSIS
ANALYSIS
MAURISH FATIMA (KEMU)
OUTLINE OF SESSION
• Principles of a meta-analysis
• Steps in a meta-analysis
• Presenting your results
ANALYS CONTAC
MENU
IS T
Study level
● Study A Outcome data Effect
measure
Review level
● Effect
Study C Outcome data
measure
ANALYS CONTAC
MENU
IS T
Data ANALYSIS
What is a meta-
anlaysis?
BRAINSTORMING!!!
ANALYS CONTAC
MENU
IS T
Data ANALYSIS
OPTIONAL
03 PART OF A
META-ANALYSIS
SYSTEMATIC
REVIEW
ANALYS CONTAC
MENU
IS T
Data ANALYSIS
WHY META-
AnAlYSIS?
• Quantify treatment effects and their
uncertainty
• Increase power
• Increase precision
• Explore differences between the results
• Settle controversies from conflicting
studies
• Generate new hypotheses
ANALYS CONTAC
MENU
IS T
Data ANALYSIS
SOFTWARES
YOU NEED!!
• R-package
• STATA
• Revman
• Statsdirect
• Meta-XL
• Open-MetAnalyst
ANALYS CONTAC META
MENU
IS T ANALYSIS
NOT TO
?
PERFORM
A META-
ANALYSIS ?
ANALYS CONTAC
MENU
IS T
Data ANALYSIS
•
Each included study must address
the same question
Consider outcomes and
comparisons
Requires your subjective •
Garbage in-
garbage out
A meta-anlaysis is only as good as
judgement
2
the studies in it
• Combining a broad mix of stuides • If studies included are biased then
answers broad questions meta-analysis will be incorrect
• Answer maybe meaningless and • If serious reporting bias is present
genuine effects maybe obscured if then unrepresentative set of
studies are too diverse studies will give misleading
results
ANALYS CONTAC
MENU
IS T
Data ANALYSIS
Steps in meta-analysis
Step 1 Step 2 Step 3
Identify comparisons to be Identify outcomes to be Collect data from each
made reported and statistics to be relevant study
used
Selecting COMPARISONS
Hypothetical review: Caffeine for day-time
drowsiness Break your topic 70
into pairwise %
comparisons
Caffeinated
Decaffeinated Use your own
coffee judgement to 30
decide what to
%
group and what
should be a
separate
comparison
ANALYS CONTAC
MENU
IS T
Data ANALYSIS
rr
For each For each
comparison, outcome, select
select an effect
outcomes measure
Depends on Depends on
data available
in included
studies
data available
in included
studies
HEADACHES
ANALYS CONTAC
MENU
IS T
Data ANALYSIS
EXAMPLES??
ANALYS CONTAC
MENU
IS T
Data ANALYSIS
RISK RATIO
RISK
ODD RATIO DIFFERENC
E
ANALYS CONTAC
MENU
IS T
Data ANALYSIS
MEAN
DIFFERENC
E
STANDARDI
ZED MEAN
DIFFERENC
E
ANALYS CONTAC
MENU
IS T
Data ANALYSIS
WHAT IS Risk???
Number of EVENTS OF
RISK INTEREST
TOTAL NUMBER OF
OBSERVATIONS
EXAMPLES • 1 day of the week is
MONDAY
• What is the probability • There are 7 days in a
that today is Monday? week
ANALYS CONTAC
MENU
IS T
Data ANALYSIS
EXAMPLES
In a study conducted at RMU, 20 students
were identified taking SSRIs and 5 of them
developed emotional blunting
EXAMPLES
In a study conducted at RMU, 20 students
were identified taking SSRIs and 5 of them
developed emotional blunting
OLL!!
ANALYS CONTAC
MENU
IS T
Data ANALYSIS
BETW
EEN
GROU
ANALYS CONTAC
MENU
IS T
Data ANALYSIS
Treatment a b a+b
Control c d c+d
Treatment a b a+b
Control c d c+d
Odds in treatment = a/ b
Odd in control = c/d
Odd Ratio = Odd in treatment/ Odds in control
=
ANALYS CONTAC
MENU
IS T
Data ANALYSIS
Treatment 10 90 100
Control 14 86 100
Treatment 10 90 100
Control 14 86 100
DIRECTION OF EFFECTS!!
• Is a risk ratio or odds ratio defined as
treatment over control or control over
treatment?
• Is rd defined as treatment – control or
control minus treatment?
CONTINUOUS DATA
MEAN = MEAN OUTCOME IN
difference INTERVENTION – MEAN
Haemoglobin OUTCOME INFortifie
CONTROLUnfortified group
levels are on d
average 0.5 g/L group
larger in the n 271 269
fortified group Mean Hb 121.0 120.5
compared with
SD 10.1 9.5
the unfortified
group Mean Difference (95% CI) 0.5 (-1.1, 2.2)
ANALYS CONTAC
MENU
IS T
Data ANALYSIS
CONTINUOUS DATA
STANDARDIZE MEAN OUTCOME IN INTERVENTION –
D MEAN = MEAN OUTCOME
STANDARD IN CONTROL
DEVIATION OF
difference OUTCOME AMONG PARTICIPANTS
SMD expresses the size of the intervention effect
relative to the variability observed
ANALYS CONTAC
MENU
IS T
Data ANALYSIS
CONTINUOUS DATA
WHEN TO
USE MD
AND SMD??
same scale??? MEAN
DIFFERENCE!!
different STANDARDIZ
scale??? ED MEAN
DIFFERENCE!!
ANALYS CONTAC
MENU
IS T
Data ANALYSIS
CALCULATING
SUMMARY
• collect a summary statistic from each contributing study
• STATISTICS
how do we bring them together?
• treat as one big study – add intervention & control data?
• breaks randomiZation, will give the wrong answer
• simple average?
• weights all studies equally – some studies closer to the truth
• weighted average
ANALYS CONTAC
MENU
IS T
Data ANALYSIS
WHAT IS THIS???
A FOREST
PLOT!!
ANALYS CONTAC
MENU
IS T
Data ANALYSIS
weighti
ng
studies
ANALYS CONTAC
MENU
IS T
Data ANALYSIS
more participants,
more events,
narrower confidence More events
interval
calculated using the
effect estimate and
Narrow
its variance Confidence
interval
ANALYS CONTAC
MENU
IS T
Data ANALYSIS
EXAMPLES
HETEROGENEIT
Y
IN META-
ANALYIS
ANALYS CONTAC
MENU
IS T
Data ANALYSIS
?
ANALYS CONTAC
MENU
IS T
Data ANALYSIS
HETEROGENEITY
Heterogeneity in meta-analysis refers
to the variation in study outcomes
between studies
ANALYS CONTAC
MENU
IS T
Data ANALYSIS
HETEROGENITYCLINCAL
heterogeneit
y
TYPES
Statistical heterogeneity
METHODO
LGICAL
heterogenet
y
ANALYS CONTAC
MENU
IS T
Data ANALYSIS
Sources of Heterogeneity
Different
study Different
settings outcomes
EXPLORE
HETEROGENEI 0 0 IGNORE IT
-Fixed effect model ignore
TYanalysis
-subgroup
-Meta-regression
3 4 heterogeneity
-ignoring may mean that an
intervention effect actually doesn’t
exist
Random effects
meta-analysis
0 0 EXCLUDE
Incorporates heterogeneity but is not
a substitution for a thorough 5 6 STudies
Sensitivity analysis
investigation
ANALYS CONTAC
MENU
IS T
Data ANALYSIS
Assessment of a heterogeneity
from a forest plot
ANALYS CONTAC
MENU
IS T
Data ANALYSIS
Significant
Results
DO EXPLAIN
YOUR
HETEROGENEITY!
Heterogeneity
ANALYS CONTAC
MENU
IS T
Data ANALYSIS
Assumes that ea
ch
has its own uniqu study
e
size and true eff effect
ect size
vary across studie
s
Peto
for odds ratios only
good with few events and
small effect sizes (OR close
to 1)
ANALYS CONTAC
MENU
IS T
Data ANALYSIS
“USE RANDOM
EFFECT
MODEL!!”
—SOMEONE
FAMOUS
ANALYS CONTAC
MENU
IS T
FOREST PLOT
effect of vitamin D supplementation on total cancer mortality
ANALYS CONTAC
MENU
IS T
FOREST PLOT
effect of vitamin D supplementation on total cancer mortality
ANALYS CONTAC
MENU
IS T
FOREST PLOT
effect of vitamin D supplementation on total cancer mortality
ANALYS CONTAC
MENU
IS T
FOREST PLOT
effect of vitamin D supplementation on total cancer mortality
ANALYS CONTAC
MENU
IS T
FOREST PLOT
effect of vitamin D supplementation on total cancer mortality
ANALYS CONTAC
MENU
IS T
FOREST PLOT
effect of vitamin D supplementation on total cancer mortality
ANALYS CONTAC
MENU
IS T
FOREST PLOT
effect of vitamin D supplementation on total cancer mortality
ANALYS CONTAC
MENU
IS T
FOREST PLOT
effect of vitamin D supplementation on total cancer mortality
ANALYS CONTAC
MENU
IS T
FOREST PLOT
effect of vitamin D supplementation on total cancer mortality
ANALYS CONTAC
MENU
IS T
FOREST PLOT
effect of vitamin D supplementation on total cancer mortality
ANALYS CONTAC
MENU
IS T
FOREST PLot
continuous outcomes
ANALYS CONTAC
MENU
IS T
FOREST PLot
LOGO
FOR
COCHRA
NE
LIBRARY
ANALYS CONTAC
MENU
IS T
Data ANALYSIS
significance
• if the CI includes the null value
• rarely means evidence of no effect
• effect cannot be confirmed or refuted by the available evidence
consider what level of change is clinically important
ANALYS CONTAC
MENU
IS T
Data ANALYSIS
Publication bias
Positive studies are more likely to be published
than negative studies
Reasons Positive
-Language barriers Studies
-unwillingness to publish negative findings
-hard to publish negative findings
Negative
Studies
OVER ESTIMATE OF EFFECT SIZE!!!
ANALYS CONTAC
MENU
IS T
Data ANALYSIS
FUNNEL PLOT
ANALYS CONTAC
MENU
IS T
Data ANALYSIS
Funnel plots are scatter plots of the effect size estimate versus the
standard error (or sample size) of each study. They can be used to assess
publication bias and heterogeneity
-A symmetric funnel plot indicates low heterogeneity and no publication
bias.
-Asymmetry can indicate heterogeneity or publication bias.
-Statistical tests can then be used to confirm this.
Number of studies usually have to be >10 for funnel plots to be effective
in quantifying heterogeneity. Studies <10 may benefit from Doi plots
ANALYS CONTAC
MENU
IS T
Data ANALYSIS
FUNNEL PLOT
For total
mortality
ANALYS CONTAC
MENU
IS T
Data ANALYSIS
FUNNEL PLOT
For total
CANCER
INCIDENC
E
ANALYS CONTAC
MENU
IS T
Data ANALYSIS