Making a decision in clinical practice

1
Making a decision in clinical practice

• Clinical experience
• Clinical trials
• Review of the literature
• Meta-analyses

2
Definition

• Meta-analysis is a means of quantitatively
  combining the results of research studies
  addressing similar questions to provide overall
  summary statistics

3
Different types of meta-analysis

• Summary data extracted from publications
• Summary data obtained from trialists
• Central collection, validation and re-analysis of
  updated individual patient data (IPD) obtained
  from trialists

4
Conduct of meta-analyses
Write protocol Objectives, inclusion criteria,
analyses
Identify all relevant trials
Establish Secretariat and Trialist Groups
Assemble the most complete dataset possible
Collect and validate data
Analyse individual studies and perform
meta-analysis
Collaborators Conference
Processes the same for summary data and IPD
Processes are similar for summary data and IPD,
but methodology and practical aspects differ
Prepare structured report
5
Benefits of meta-analyses

• Size
• High statistical power
• Reliable detection of rare events
• Heterogeneity
• Identification of sources of heterogeneity
• Prognostic factor and subset analyses
• Replication
• Identification of biases in individual trials
• Further analyses possible

6
Additional benefits of IPD MAs

• Quality of data
• Absence of publication and selection bias
• Checks possible on patient data
• Optimality of analyses
• Use of appropriate statistical methods (e.g.
  survival analyses)
• Avoidance of grossly inadequate methods(e.g.
  meta-regression)
• Consensus
• Investigators involvement

7
Simple (approximate) formulas for meta-analysis
of binary data
Observed number of events (O) 45 Expected
number of events (E) 95/200 100
47.5 Variance of number of events (V) _at_ 50 / 4
12.5
8
Observed minus Expected and its Variance provide
estimate of OR and X² test statistic
log OR _at_ (O - E) / V -2.5 / 12.5
-0.20   OR exp (-0.20) 0.82 X² (O - E)² /
V Note estimate of OR is biased when OR is far
from 1, and/or when numbers in treated and
control are very imbalanced
9
Statistics for one trial
Observed minus expected
-
(
)
O
E
-
(
)
O
E
Variance
-
V
O
E
(
)
Odds ratio
-
(
)
é
ù
O
E
OR
ê
ú
V
ë
û
e
95 CI of OR
-
é
ù
O
E
1
96
(
)
.
IC
OR
95

ê
ú
V
ë
û
V
e
X² test
-
2
(
)
O
E
2
X
V
10
Statistics for n trials
Observed minus expected
Variance
Odds ratio
å
é
ù
OR
-
O
E
(
)
ê
ú
å
ê
ú
V
ë
û
e
95 CI of OR
-
é
ù
O
E
1
96
(
)
.

ê
ú
V
ë
û
V
e


X² test
å
2
-
(
)
O
E
å
V
11
Presentation of results the  forest  plot
12
Heterogeneity assessment and generation of
hypotheses
13
Selection of trials for inclusion in meta-analyses

• Meta-analyses based on trials selected from the
  literature may be biased
• Only properly randomized trials should be
  included
• All trials should be included

14
Predictive factors(tests for interaction)

• Predictive factors
• Factors that modify the effects of treatment
  (measured in relative terms, e.g. through odds
  ratio)
• Such factors can be identified by tests for
  interaction

15
Lancet 19983511451-67, 352930-42
16
Predictive factors(tests for trend)

• Predictive factors
• Predictive factors can also be identified far
  more effectively by tests for trend if they have
  ordered categories (e.g. age)

17
Lancet 19983511451-67, 352930-42
18
(No Transcript)
19
Only what is exhaustive is of interest. The
truth comes from a build-up of details. Thom
as Mann