Publication Bias

Report
Publication Bias
Emily E. Tanner-Smith
Associate Editor, Methods Coordinating Group
Research Assistant Professor, Vanderbilt University
Campbell Collaboration Colloquium
Copenhagen, Denmark
May 29th, 2012
The Campbell Collaboration
www.campbellcollaboration.org
Outline
• What is publication bias
• Avoiding publication bias
• Methods for detecting publication bias
• Detecting publication bias in Stata
• Summary & recommendations
The Campbell Collaboration
www.campbellcollaboration.org
Publication Bias
Publication bias refers to bias that occurs when research
found in the published literature is systematically
unrepresentative of the population of studies (Rothstein et
al., 2005)
• Publication bias is often referred to as the file drawer
problem where:
•
“…journals are filled with the 5% of studies that show Type I errors,
while the file drawers back at the lab are filled with the 95% of the
studies that show non-significant (e.g. p < 0.05) results” (Rosenthal,
1979)
The Campbell Collaboration
www.campbellcollaboration.org
Reporting Biases
Types of Reporting Biases
Definition
Publication bias
The publication or non-publication of research findings, depending on the nature
and direction of results
Time lag bias
The rapid or delayed publication of research findings, depending on the nature and
direction of results
Multiple publication bias
The multiple or singular publication of research findings, depending on the nature
and direction of results
Location bias
The publication of research findings in journals with different ease of access or levels
of indexing in standard databases, depending on the nature and direction of results
Citation bias
The citation or non-citation publication of research findings, depending on the nature
and direction of results
Language bias
The publication of research findings in a particular language, depending on the
nature and direction of results
Outcome reporting bias
The selective reporting of some outcomes but not others, depending on the nature
and direction of results
Source: Sterne et al. (Eds.) (2008: 298)
The Campbell Collaboration
www.campbellcollaboration.org
Why Publication Bias Matters
• Systematic reviews and meta-analyses are often used to
inform policy and practice
• Omitting unpublished studies from a review could yield a
biased estimate of an intervention effect
– Biased results could lead decision-makers to adopt practices
that may ultimately cause harm, increase adverse events, or
prevent treatment of life-threatening diseases or disorders
The Campbell Collaboration
www.campbellcollaboration.org
Avoiding Publication Bias
As Primary Researchers:
– Ethical imperative for primary researchers to publish
null/negative findings
– Prospective registration of trials
As Systematic Reviewers/Meta-analysts:
– Prospective meta-analysis of studies identified prior to reporting
of study results
– Extensive grey literature searching
– Transparent assessment of possible bias
The Campbell Collaboration
www.campbellcollaboration.org
Avoiding Publication Bias: Grey Literature Searching
• An ounce of prevention is worth a pound of cure…
•
•
•
•
•
Conference proceedings
Technical reports (research, governmental agencies)
Organization websites
Dissertations, theses
Contact with primary researchers
The Campbell Collaboration
www.campbellcollaboration.org
Avoiding Publication Bias: Grey Literature Searching
Databases
Australian Education Index Current Controlled Trials Index to Theses
PAIS Archive
Bib of Nordic Criminology Directory of OA Journals INASP
PAIS International
British Education Index
Dissertation Abstracts
ISI Conf Proceedings Index PolicyFile
Canadian Eval Society
DissOnline
LILACS Latin American
Project Cork
CBCA Education
DrugScope DrugData
NTIS
PsycArticles
CERUK
EconLit
NCJRS Abstracts Database PsycEXTRA
Child Welfare Info Gateway Educ Research Global
NLM Gateway
Social Care Online
ClinicalTrials.gov
ERIC
NARCIS
SSRN eLibrary
CORDIS Library
HINARI
NBBF
Theses Canada
CRD
HMIC
NY Acad of Med
TRID
CrimDOC
HUD User Database
Open Grey
WHO Trials
The Campbell Collaboration
www.campbellcollaboration.org
Detecting Publication Bias
Methods for detecting publication bias assume:
• Large n studies are likely to get published regardless of results
due to time and money investments
• Small n studies with the largest effects are most likely to be
reported, many will never be published or will be difficult to
locate
• Medium n studies will have some modest significant effects
that are reported, others may never be published
The Campbell Collaboration
www.campbellcollaboration.org
Funnel Plots
• Exploratory tool used to visually assess the possibility of
publication bias in a meta-analysis
• Scatter plot of effect size (x-axis) against some measure of
study size (y-axis)
– x-axis: use logged values of effect sizes for binary data, e.g.,
ln(OR), ln(RR)
– y-axis: the standard error of the effect size is generally
recommended (see Sterne et al., 2005 for a review of additional
y-axis options)
• Not recommended in very small meta-analyses (e.g., n < 10)
The Campbell Collaboration
www.campbellcollaboration.org
Funnel Plots
• Precision of estimates increases as the sample size of a
study increases
– Estimates from small n studies (i.e., less precise, larger
standard errors) will show more variability in the effect size
estimates, thus a wider scatter on the plot
– Estimates from larger n studies will show less variability in effect
size estimates, thus have a narrower scatter on the plot
• If publication bias is present, we would expect null or
‘negative’ findings from small n studies to be suppressed
(i.e., missing from the plot)
The Campbell Collaboration
www.campbellcollaboration.org
0
- Note x & y axes
Asymmetry in
small n studies
provides
evidence of
possible bias
mean
- Pseudo 95%
confidence limits
1.5
1
.5
- Centered around FE
-4
The Campbell Collaboration
-2
Favors treatment
0
Log odds ratio (lor)
2
Favors control
4
www.campbellcollaboration.org
0
1.5
1
.5
Symmetric
funnel plots
indicate a
possible
absence of
bias
-4
-2
Favors treatment
The Campbell Collaboration
0
Log odds ratio (lor)
2
4
Favors control
www.campbellcollaboration.org
0.00
0.25
0.20
0.15
0.10
0.05
Symmetry
difficult to
assess with
<10 studies
-0.40
The Campbell Collaboration
-0.20
0.00
0.20
Log risk ratio (lrr)
0.40
0.60
www.campbellcollaboration.org
Interpreting Funnel Plots
• Asymmetry could be due to factors other than publication
bias, e.g.,
– poor methodological quality (smaller studies with lower quality
–
–
–
–
may have exaggerated treatment effects)
Other reporting biases
Artefactual variation
Chance
True heterogeneity
• Assessing funnel plot symmetry relies entirely on subjective
visual judgment
The Campbell Collaboration
www.campbellcollaboration.org
Contour Enhanced Funnel Plots
• Funnel plot with additional contour lines associated with
‘milestones’ of statistical significance: p = .001, .01, .05, etc.
– If studies are missing in areas of statistical non-significance,
publication bias may be present
– If studies are missing in areas of statistical significance,
asymmetry may be due to factors other than publication bias
– If there are no studies in areas of statistical significance,
publication bias may be present
• Can help distinguish funnel plot asymmetry due to
publication bias versus other factors (Peters et al., 2008)
The Campbell Collaboration
www.campbellcollaboration.org
0
1
.5
1.5
-4
The Campbell Collaboration
-2
Favors treatment
0
Log odds ratio (lor)
2
Favors control
4
www.campbellcollaboration.org
0
.5
Asymmetry may
be due to factors
other than
publication bias
Studies
p < 1%
1% < p < 5%
5% < p < 10%
p > 10%
1
1.5
-4
The Campbell Collaboration
-2
0
Log odds ratio (lor)
2
4
www.campbellcollaboration.org
Tests for Funnel Plot Asymmetry
• Several regression tests are available to test for funnel plot
asymmetry
– Attempt to overcome subjectivity of visual funnel plot inspection
• Framed as tests for “small study effects”, or the tendency for
smaller n studies to show greater effects than larger n studies;
i.e., effects aren’t necessarily a result of bias
The Campbell Collaboration
www.campbellcollaboration.org
Egger Test
• Recommended test for mean difference effect sizes (d, g)
• Weighted regression of the effect size on standard error
ES i  1   0 sei   i
H0 : 0  0
– β0 = 0 indicates a symmetric funnel plot
– β0 > 0 shows less precise (i.e., smaller n) studies yield bigger effects
– Can be extended to include p predictors hypothesized to potentially
explain funnel plot asymmetry (see Sterne et al., 2001)
The Campbell Collaboration
www.campbellcollaboration.org
0
.2
.4
.6
-3.00
The Campbell Collaboration
-2.00
-1.00
Standardized mean difference (d)
0.00
www.campbellcollaboration.org
0.00
Fail to reject the null
hypothesis of no small study
effects (b = -1.14; p = .667)
No evidence of “small study
bias”
-3.00
-2.00
-1.00
Standardized mean difference (d)
ES  .61  1.14 se
.6
.4
.2
0
The Campbell Collaboration
www.campbellcollaboration.org
Egger Test
• Limitations
– Low power unless there is severe bias and large n
– Inflated Type I error with large treatment effects, rare event
data, or equal sample sizes across studies
– Inflated Type I error with log odds ratio effect sizes
The Campbell Collaboration
www.campbellcollaboration.org
Peters Test
• Modified Egger test that for use with log odds ratio effect
sizes
– Weighted regression of ES on 1/total sample size
ESi  1   0 (1 / n)   i
(nevents * nnonevents )
w
n
The Campbell Collaboration
www.campbellcollaboration.org
0
1
.5
1.5
-4
The Campbell Collaboration
-2
Favors treatment
0
Log odds ratio (lor)
2
Favors control
4
www.campbellcollaboration.org
0
ES  .03  118(1 / n)
-1.5
-1
-.5
Reject the null
hypothesis of no small
study effects
(b=-118.21;p=.002)
-2.5
-2
Possible evidence of
“small study bias”
0
The Campbell Collaboration
.005
.01
1/total sample size
.015
.02
www.campbellcollaboration.org
Tests for Funnel Plot Asymmetry
• Other recommended tests for use with the log odds ratio effect
size:
– Harbord test (Harbord et al., 2006) if τ2 < .10
– Rücker test (Rücker et al., 2008)
• Numerous other tests available (see Sterne et al., 2008 for a
review)
The Campbell Collaboration
www.campbellcollaboration.org
Other Methods
• Selection modeling (Hedges & Vevea, 2005)
– Incorporate biasing selection mechanism into your model to get
an adjusted mean effect size estimate
– Selection model is rarely known; do sensitivity analysis with
alternative selection models
– Relatively complex to implement, performs poorly with small
number of studies
The Campbell Collaboration
www.campbellcollaboration.org
Other Methods
• Trim and fill analysis (Duval & Tweedie, 2000)
– Iteratively trims (removes) smaller studies causing asymmetry
– Uses trimmed plot to re-estimate the mean effect size
– Fills (replaces) omitted studies and mirror-images
– Provides an estimate of the number of missing (filled) studies
and a new estimate of the mean effect size
– Major limitations include: misinterpretation of results,
assumption of a symmetric funnel plot, poor performance in the
presence of heterogeneity
The Campbell Collaboration
www.campbellcollaboration.org
Other Methods
• Sensitivity testing
– Comparing fixed- and random-effects estimates
• Cumulative meta-analysis
– Typically used to update pooled effect size estimate with each
new study cumulatively over time
– Can use as an alternative to update pooled effect size estimate
with each study in order of largest to smallest sample size
– If pooled effect size does not shift with the addition of small n
studies, provides some evidence against publication bias
The Campbell Collaboration
www.campbellcollaboration.org
Other Methods
• Failsafe N (Rosenthal 1979)
– Number of additional null studies that would be needed to
increase the p-value to above .05
– Ad hoc rule of thumb that failsafe N less than 5n + 10, results
may not be robust to publication bias
– Several variations of the failsafe N
– Numerous limitations (not recommended for use); see Becker
(2005)
The Campbell Collaboration
www.campbellcollaboration.org
Detecting Publication Bias in Stata
• Several user-written commands are available that automate
the most commonly used methods to detect publication bias
Method
Stata ado-file
Funnel plots
metafunnel
Contour enhanced funnel plots confunnel
Egger, Peters, Harbord tests
metabias
Cumulative meta-analysis
metacum
Trim and fill analysis
metatrim
The Campbell Collaboration
www.campbellcollaboration.org
Summary & Recommendations
• Publication bias deserves careful consideration in systematic
reviews and meta-analyses, given their potentially large
impact on policy and practice
• Narrative and non-systematic reviews are subject to all the
same potential biases as systematic reviews and metaanalyses
– Yet publication bias is rarely if ever acknowledged in narrative reviews
– Meta-analyses have the benefit of being able to empirically assess the
possibility of publication bias and its potential impact on review findings
The Campbell Collaboration
www.campbellcollaboration.org
Summary & Recommendations
• Reporting biases occur when the nature and direction of
research findings influence their dissemination and
availability
• The reality of reporting biases means systematic reviewers
must conduct comprehensive literature searches in attempt
to locate all eligible studies
• Protocols and reviews should be explicit and transparent
about methods used to assess publication bias
The Campbell Collaboration
www.campbellcollaboration.org
Summary & Recommendations
• Funnel plots
– Always examine & report funnel plots when you have 10 or
more studies with some variability in standard errors across
studies
– Always consider publication bias as only one possible source of
funnel plot asymmetry
The Campbell Collaboration
www.campbellcollaboration.org
Summary & Recommendations
• Regression tests
– For continuously measured intervention effects (d, g): Egger
test
– For log odds ratio effect sizes: Peters, Harbord, or Rücker test if
τ2 < .10
– For log odds ratio effect sizes: Rücker test if τ2 > .10
– Acknowledge low power of statistical tests
• Other sensitivity tests
– Comparing FE vs. RE estimates, trim & fill analysis, cumulative
meta-analysis, selection modeling
The Campbell Collaboration
www.campbellcollaboration.org
Summary & Recommendations
• What if you find possible evidence of publication or small
study bias?
– “Solution” will vary; requires thoughtful consideration by the
–
–
–
–
reviewers
Reconsider search strategy, grey literature inclusion
Identify plausible explanations (e.g., study quality, other study
characteristics)
Explore potential explanations with subgroup and moderator
analyses
Explicitly acknowledge all potential biases when discussing the
findings of the review
The Campbell Collaboration
www.campbellcollaboration.org
Recommended Reading
Duval, S. J., & Tweedie, R. L. (2000). A non-parametric ‘trim and fill’ method of
accounting for publication bias in meta-analysis. Journal of the American Statistical
Association, 95, 89-98.
•
Egger, M., Davey Smith, G., Schneider, M., & Minder, C. (1997). Bias in meta-analysis
detected by a simple, graphical test. British Medical Journal, 315, 629-634.
•
Hammerstrøm, K., Wade, A., Jørgensen, A. K. (2010). Searching for studies: A guide to
information retrieval for Campbell systematic reviews. Campbell Systematic Review,
Supplement 1.
•
Harbord, R. M., Egger, M., & Sterne, J. A. C. (2006). A modified test for small-study
effects in meta-analyses of controlled trials with binary endpoints. Statistics in Medicine, 25,
3443-3457.
•
Peters, J. L., Sutton, A. J., Jones, D. R., Abrams, K. R., & Rushton, L. (2008). Contourenhanced meta-analysis funnel plots help distinguish publication bias from other causes of
asymmetry. Journal of Clinical Epidemiology, 61, 991-996.
•
The Campbell Collaboration
www.campbellcollaboration.org
Recommended Reading
Rosenthal, R. (1979). The ‘file-drawer problem’ and tolerance for null results.
Psychological Bulletin, 86, 638-641.
•
Rothstein, H. R., Sutton, A. J., & Borenstein, M. L. (Eds). (2005). Publication bias in
meta-analysis: Prevention, assessment and adjustments. Hoboken, NJ: Wiley.
•
Rücker, G., Schwarzer, G., & Carpenter, J. (2008). Arcsine test for publication bias in
meta-analyses with binary outcomes. Statistics in Medicine, 27, 746-763
•
Sterne, J. A., & Egger, M. (2001). Funnel plots for detecting bias in meta-analysis:
Guidelines on choice of axis. Journal of Clinical Epidemiology, 54, 1046-1055.
•
Sterne, J. A. C., Egger, M., & Moher, D. (Eds.) (2008). Chapter 10: Addressing reporting
biases. In J. P. T. Higgins & S. Green (Eds.), Cochrane handbook for systematic reviews of
interventions, pp. 297 – 333. Chichester, UK: Wiley.
•
Sterne, J. A. C., et al. (2011). Recommendations for examining and interpreting funnel
plot asymmetry in meta-analyses of randomised controlled trials. BMJ, 343, d4002.
•
The Campbell Collaboration
www.campbellcollaboration.org
P.O. Box 7004 St. Olavs plass
0130 Oslo, Norway
E-mail: [email protected]
http://www.campbellcollaboration.org
The Campbell Collaboration
www.campbellcollaboration.org

similar documents