assessing the impact of searching fewer databases in rapid · of searching fewer databases in rapid...
TRANSCRIPT
Trusted evidence.
Informed decisions.
Better health.
Irma Klerings
Cochrane Austria
Assessing the impact
of searching fewer
databases in rapid
reviews
2
Potential conflicts of interest
• This research was funded by Cochrane Austria.
• No other potential or actual conflicts of interest in
relation to this presentation.
Decision making needs
reliable evidence syntheses
4
Systematic Reviews
• Most reliable &
valid support for
decision-making
• Synthesis of all
evidence about a
research question
• Systematic methods
minimize bias
• Takes 6-12 months
to complete
Rapid Reviews
• Based on systematic
review methods:
processes are
accelerated and
methods are
streamlined
• Takes 5–12 weeks to
complete
• Reliability of
conclusions?
5
Do bodies of evidence that are
based on abbreviated literature searches
lead to different conclusions
compared with those based on
comprehensive, systematic literature
searches?
Research question
6 METHODS
7
Sample selection
60 randomly selected Cochrane
reviews
Main inclusion criteria
• Authors were able to draw a conclusion
• Summary of findings tables
• Reproducible Meta-analyses
• Used MEDLINE, EMBASE, and Central
8
14 search abbreviated search approaches
compared to original comprehensive search
Search of reference lists of relevant publications
MEDLINE EMBASE CentralMEDLINE + EMBASE
MEDLINE + Central
Central + EMBASE
MEDLINE + Central + EMBASE
+
Database
coverage# of included references indexed in a database
# of included references cited in the reviewx100
Recall# of included references retrieved by a search approach
# of included references cited in the reviewx100
9
Research Process
Run searches
Check partiallyfoundstudies
RecalculateMA & createnew SoF-tables
Web-basedsurveyCochraneauthors
Non-inferiorityanalysis
10
Non-inferiority analysis
Scenario 2:
non-inferior
Scenario 1:
inferior
Non-inferiority margin(Wagner et al. 2017)
11 RESULTS
12
Review characteristics
Type of intervention
• Pharmacological (drugs, vaccines) 30 (50%)
• Non-pharmacological (psychological, educational,
dietary, physical exercise, complex interventions,
screening, surgery, management strategies) 30 (50%)
Information sources
• Medline, Embase, Central 60 (100%)
• At least 2 other types of information sources 60 (100%)
Other bibliographic databases 56 (93%)
Grey literature and unpublished data 59 (98%)
Other sources (reference lists, citation tracking,
handsearch)
56 (93%)
Study design of included studies
• RCT only or RCT+quasi-RCT 53 (88%)
• RCT+controlled clinical trial, before-after study or
interrupted time series 7 (12%)
13
Database coverage
Were the included references indexed in any of the databases?
72%
63%
75% 75%
89% 89%
92%93% 92%
100% 100% 100% 100%100%
87%
82%
91%94%
96% 97%99%
40%
50%
60%
70%
80%
90%
100%
Medline Embase Central M+E M+C C+E M+E+C
Refe
ren
ce
s ind
exe
d
Median & interquartile range of database coverage (n=60)
MedianM=Medline E=Embase C=Central
14
Recall
Were the included references found by the search strategies?
57%
71%
47%
63%
65%
75%
67%
77%
74%
79%
73%
79%78%
85%
87%
90%
84%
90%
93%
95% 95%97%
94%
98% 98%100% 100% 100%
76%
84%
71%
80%
75%
82%
85%
90%
86%
90% 89%92%
90%
94%
40%
50%
60%
70%
80%
90%
100%
Re
fere
nce
s fo
un
d
Median & interquartile range of search strategy recall without/with addedreference list checking (n=60)
MedianM=Medline E=Embase C=Central R=Reference list checking
15
Impact on overall conclusion if
Discordant conclusion = any change in conclusion
• less certainty, but the same direction of
conclusion
• opposite conclusion (= changed direction of
conclusion)
• no conclusion possible
Conclusions of
abbreviated searches
M M+R E E+R C C+R M+EM+E+
RM+C
M+C+
RC+E
C+E+
R
M+C+
E
M+C+
E+R
conclusion
does not
change48 50 44 49 47 47 50 53 53 53 50 54 52 55
Same
conclusion
with less
certainty
6 4 6 5 8 8 5 2 3 3 6 2 4 1
Opposite
conclusion2 2 3 1 1 1 1 1 1 1 1 1 1 1
conclusion
is no
longer
possible
4 4 7 5 4 4 4 4 3 3 3 3 3 3
M=Medline E=Embase C=Central R=Reference list checking
17
Any change in conclusion
11%
8%
16%
10%
12% 12%
8%
5% 5% 5%
8%
6%
4%3%
32%
29%
40%
30%
34% 34%
29%
23% 23% 23%
29%
25%
21%
18%20%
17%
27%
18%
22% 22%
17%
12% 12% 12%
17%
13%
10%8%
0%
5%
10%
15%
20%
25%
30%
35%
40%
45%
Proportion of conclusions with any changes and 95% confidenceinterval for each search approach (without/with added reference list
checking) (n=60)
Proportion of discordant conclusion
non-inferiority margin
M=Medline E=Embase C=Central R=Reference list checking
18
Impact on overall conclusion if
Disconcordant conclusion
= Opposite conclusion only
19
Opposite conclusions only
0% 0%
1%
0% 0% 0% 0% 0% 0% 0% 0% 0% 0% 0%
12% 12%
14%
9% 9% 9% 9% 9% 9% 9% 9% 9% 9% 9%
3% 3%
5%
2% 2% 2% 2% 2% 2% 2% 2% 2% 2% 2%
0%
2%
4%
6%
8%
10%
12%
14%
16%
Proportion of opposite conclusions and 95% confidence interval foreach search approach (without/with reference list checking) (n=60)
Proportion of discordant conclusion
non-inferiority margin
M=Medline E=Embase C=Central R=Reference list checking
20 • Same conclusion,
less certain
• Opposite
conclusion
• Conclusion
unchanged
• No conclusion
possible
Conclusion stays the same
Refe
rences fo
un
d
(n=60)
21
Pharmacological vs non-pharma. reviews
23%
20%
37%
27%
33%
33%
20%
13%
17%
17%
27%
20%
13%
10%
17%
13%
17%
10%
10%
10%
13%
10%
7%
7%
7%
7%
7%
7%
MEDLINE
MEDLINE +Refs
Embase
Embase +Refs
CENTRAL
CENTRAL +Refs
MEDLINE +Embase
MEDLINE +Embase +Refs
MEDLINE +CENTRAL
MEDLINE +CENTRAL +Refs
CENTRAL +Embase
CENTRAL +Embase +Refs
MEDLINE +CENTRAL +Embase
MEDLINE +CENTRAL +Embase+Refs
Proportion of "conclusions with any change" in reviews on non-pharmacological interventions (n= 30)
Proportion of "conclusions with any change" in reviews on pharmacological interventions (n= 30)
22
Recall pharma. vs non-pharma reviews
76%
83%
75%
81% 82%
90%
85%89% 88%
92% 93% 94% 94% 96%
40%
50%
60%
70%
80%
90%
100%
M M+R E E+R C C+R M+E M+E+R M+C M+C+R C+E C+E+R M+C+E M+C+E+R
Median recall of pharmacological reviews (n=30)
76%
84%
61%
72% 71%
78%
85%
92%
80%
88%
81%
85%87%
93%
40%
50%
60%
70%
80%
90%
100%
M M+R E E+R C C+R M+E M+E+R M+C M+C+R C+E C+E+R M+C+E M+C+E+R
Median recall of non-pharma. reviews (n=30)
M=Medline E=Embase C=Central R=Reference list checking
23
Conclusions depending on number of
included studies
36%
27%
41%
32%
32%
32%
27%
18%
18%
18%
27%
23%
14%
14%
11%
11%
18%
11%
16%
16%
11%
8%
8%
8%
11%
8%
8%
5%
MEDLINE
MEDLINE +Refs
Embase
Embase +Refs
CENTRAL
CENTRAL +Refs
MEDLINE +Embase
MEDLINE +Embase +Refs
MEDLINE +CENTRAL
MEDLINE +CENTRAL +Refs
CENTRAL +Embase
CENTRAL +Embase +Refs
MEDLINE +CENTRAL +Embase
MEDLINE +CENTRAL +Embase +Refs
Proportion of "conclusions with any change" in reviews including fewer than ten primary studies (n=22)
Proportion of "conclusions with any change" in reviews including ten or more primary studies (n=38)
24
Recall: number of included studies
67%
75%
63%
70%74%
78%
71%
80%
75%
80%78%
81% 80%83%
40%
50%
60%
70%
80%
90%
100%
M M+R E E+R C C+R M+E M+E+R M+C M+C+R C+E C+E+R M+C+E M+C+E+R
Median recall of reviews with fewer than ten included studies (n=22)
81%
87%
75%
83%80%
88% 87%91%
89%92% 93% 94% 94% 96%
40%
50%
60%
70%
80%
90%
100%
M M+R E E+R C C+R M+E M+E+R M+C M+C+R C+E C+E+R M+C+E M+C+E+R
Median recall of reviews with ten or more included studies (n=38)
M=Medline E=Embase C=Central R=Reference list checking
25 DISCUSSION
26
Conclusion
• If decision-makers are willing to accept less certainty
and a small risk for opposite conclusions, some
abbreviated searches are viable options for rapid
evidence syntheses.
• Decisions demanding high certainty require
comprehensive searches.
• Impact of abbreviated searches depends on type of
intervention, „size“ of the topic, and definition of „changed
conclusion“
27
Discussion
Limitations:
Central is only useful for RCTs
External validity (raw database entries vs. real-life)
Reference entries
Points for discussion:
Limiting the number of databases searched could be
more suitable for rapid reviews of pharmacological
interventions
Different streamlined methods for different
intervention-types?
28
More information
Study protocol:
Nussbaumer-Streit, B., I. Klerings, G. Wagner, et al. (2016). "Assessing the validity of abbreviated literature searches for rapid reviews: protocol of a non-inferiority and meta-epidemiologic study." Syst Rev 5(1): 197.
Main analysis:
Nussbaumer-Streit, B., I. Klerings, G. Wagner, et al. (2018). "Abbreviated literature searches were viable alternatives to comprehensive searches: a meta-epidemiological study." J Clin Epidemiol Article in Press.
29
Principal investigators:
Barbara Nussbaumer-Streit, Gerald Gartlehner
Search specialists:
Irma Klerings, Megan van Noord, Tarquin Mittermayr
Project members:
Andreea Dobrescu, Thomas Heise, Jan Stratil, Gernot Wagner, Susan Armijo Olivo,
Emma Persad, Stefan Lhachimi, Hajo Zeeb, Birgit Teufer
Meta-epidemiology:
Lars Hemkens, Aviv Ladanie
Thank you to
the authors of the Cochrane Reviews &
our team