Upload
shahidul
View
213
Download
1
Embed Size (px)
Citation preview
2013
http://informahealthcare.com/jmfISSN: 1476-7058 (print), 1476-4954 (electronic)
J Matern Fetal Neonatal Med, 2013; 26(8): 768–771! 2013 Informa UK Ltd. DOI: 10.3109/14767058.2012.755161
Challenges in the peer review of systematic reviews and meta-analyses
Anthony M. Vintzileos1, Jonathan Carvajal1, and Shahidul Islam2
1Department of Obstetrics and Gynecology, and 2Department of Health Outcomes Research, Winthrop University Hospital, Mineola, NY, USA
Abstract
Objective: To assess the role of the referees in assisting the peer review process of systematicreviews and meta-analyses.Methods: A one-page questionnaire was mailed to 1391 referees of two journals, theAmerican Journal of Obstetrics and Gynecology and Obstetrics and Gynecology. The refereeswere asked how often they verified by their own independent analysis 11 key items relatedto the methodology and statistical analysis of systematic reviews and meta-analyses.Response categories included ‘‘always’’, ‘‘frequently’’ (450% of the time), ‘‘infrequently’’(�50% of the time) and ‘‘never’’. A second and a third mailing was sent to the non-respondents.Results: 42 mailings were returned because of change of address. Of the remaining 1349referees, 272 responded (response rate 20%). Of the 272 respondents, 159 (58%) had previouslyreviewed articles dealing with systematic reviews or meta-analyses. The responses variedaccording to the key items in the questions but the referees used their own independentanalyses ‘‘always’’ in only 2%–17% of the time. The rates of ‘‘infrequently’’ or ‘‘never’’ responsescombined together ranged from 51% to 86% for the various key items.Conclusion: The overwhelming majority of the referees do not verify, by their own independentanalysis, key items related to methodology and statistical analysis of submitted systematicreviews and meta-analyses.
Keywords
Evidence-based medicine, peer-review,questionnaire, survey
History
Received 15 August 2012Revised 19 October 2012Accepted 6 November 2012Published online 15 January 2013
Introduction
Systematic reviews and meta-analyses are tools used to
summarize the evidence emanating from studies on the same
topic. Approximately 1 in 10 level A recommendations in the
obstetrical bulletins are based on meta-analyses [1]. Since the
results of such systematic reviews and meta-analyses are
frequently used by physicians and policy makers to imple-
ment health care changes, these reports should have
transparency. In order to improve the quality of the reporting,
reporting guidelines have been mandated by most journals.
Such reporting guidelines for systematic reviews and meta-
analyses include the QUOROM (QUality Of Reporting Of
Meta-analysis) [2] and the PRISMA (Preferred Reporting
Items for Systematic reviews and Meta-Analyses) [3].
Although these guidelines may increase the quality of
reporting of systematic reviews and meta-analyses they do
not address the quality of the peer review process which
precedes the publication.
The purpose of this study was to determine the manner in
which the referees peer review articles labeled as systematic
reviews and meta-analyses.
Material and methods
A packet containing a one-page questionnaire along with an
explanatory cover letter, a stamped postcard with the referee’s
ID number and a stamped return envelope, were mailed to
1391 referees across the United States. The referees were
chosen from the published lists of the ‘‘editorial consultants’’
of two leading journals in obstetrics and gynecology: the
American Journal of Obstetrics and Gynecology [4] and
Obstetrics and Gynecology [5]. The exact mailing addresses
of the referees were extracted from ‘‘Google’’ using the first
and last name of each referee as key words and the city and
state which appeared under the editorial consultants’
published lists. The names and mailing addresses of the
referees were compiled into a list and each referee was
assigned an ID number. The same ID number was placed on
the return postcard.
The cover letter to the referees stated that the purpose of
the survey was to identify the role of the referees in assisting
with the peer review process of systematic reviews and meta-
analyses and that the survey was anonymous and confidential.
In order to ensure the anonymity and confidentiality, the
referees were instructed to mail back the completed
questionnaire and the postcard with their ID number
separately. This methodology allowed for the anonymity and
confidentiality and at the same time prevented us from
sending repeat reminders to those who had already completed
and returned the questionnaire. The referees were given the
Address for correspondence: Dr Anthony Vintzileos, MD, Department ofObstetrics and Gynecology, Winthrop University Hospital, Mineola,USA. E-mail: [email protected]
J M
ater
n Fe
tal N
eona
tal M
ed D
ownl
oade
d fr
om in
form
ahea
lthca
re.c
om b
y B
iblio
teka
Uni
wer
syte
tu W
arsz
awsk
iego
on
10/2
9/14
For
pers
onal
use
onl
y.
option of either responding to the questionnaire by regular
mail using the enclosed stamped envelope or by using a
website (http://www.surveymonkey.com/s/Q3FKLHD) (last
assessed 9 Jan 2012). In order to increase the response rate
a second and third mailing was sent to non-respondents.
The questionnaire included questions related to the
referees’ age, gender, medical or other degrees, number of
years of reviewing and number of reviewed papers in the prior
12 months. Those referees who had previously reviewed
systematic reviews or meta-analyses articles were asked how
often they verified by their own independent analysis 11 key
items related to methodology and statistical analysis
(Table 1). For each of the items there were four possible
responses: ‘‘always’’, ‘‘frequently’’ (450% of the time),
‘‘infrequently’’ (�50% of the time) and ‘‘never’’. There were
no open-ended questions. The study was approved by the
Institutional Review Board of Winthrop University Hospital.
The results were reported as descriptive statistics.
Continuous data were presented as medians (range) and
categorical data were presented as number (%). In order to
determine if the responses were influenced by age, gender,
degree, years of reviewing or number of reviewed papers
during the prior year, the responses were collapsed into two
groups (‘‘always’’ and ‘‘frequently’’ combined versus
‘‘infrequently’’ and ‘‘never’’ combined) and then applied a
random-intercept logistic regression model. In these models,
we used survey question as the fixed effects and ID variable
as the random effect. This statistical analysis was performed
by the SAS (SAS Institute, Cary, NC). The results were
considered statistically significant when p50.05.
Results
Of the 1391 initial mailings, 42 were returned because of
incorrect address. Of the remaining 1349 referees, 272
responded (response rate 20%). The majority of responses
(237 or 87%) were received via regular mail; the remaining
responses (35 or 13%) were received electronically. Of the
272 respondents, 113 (42%) had not reviewed any systematic
reviews or meta-analyses, whereas 159 (58%) had previously
reviewed such articles. The demographic characteristics of the
responders are shown in Table 2. Among those who
responded via regular mail, there was wide geographic
representation from 40 states across the United States
(Northeast 11%; Mid-Atlantic 18%; Midwest East North
Central 17%; West North Central 6%; South Atlantic 13%;
East South Central 9%; West South Central 6%; West
Mountain 9%; and West Pacific 11%). The responses with
respect to the frequency that the referees verified by their own
independent analysis of the 11 key items are shown in Table 1.
The referees used their own independent analyses ‘‘always’’
in only 2%–17% of the time. In contrast, the rates of
‘‘infrequently’’ or ‘‘never’’ responses combined together
ranged from 51% to 86%. The regression analysis showed that
the responses were not influenced by age, gender, degree,
years of reviewing, number of reviewed papers during the
prior year or response media (regular versus electronic)
(Table 3).
Discussion
It is a commonly held belief that systematic reviews and
meta-analyses are valuable tools to summarize evidence
relating the efficacy and safety of health care interventions.
However, systematic reviews and meta-analyses often lack
clarity resulting in poor reporting [3]. This has led to the
creation of reporting guidelines for authors, such as
QUOROM and PRISMA [2,3] which have been adopted by
several peer review journals in order to increase clarity of
reporting. Although these guidelines improve communication
between author(s) and reader(s), they were not developed to
improve the quality of the peer review process. There are
virtually no data regarding the quality of the peer review by
the referees in general and how often they verify by their own
Table 1. Referees’ responses on how frequently they use their own independent analyses in verifying key items in methodology and statisticalanalysis.
AlwaysFrequently
(450% of the time)Infrequently
(�50% of the time) Never
Verify the initial set of articles by using the specified key words andsearch engine(s) (i.e. PubMed) that the authors used
19 (12%) 26 (16%) 58 (37%) 56 (35%)
Verify all additional articles, identified by the authors, by using othersources (i.e. abstracts, conference proceedings, etc.)
7 (4%) 25 (16%) 60 (38%) 67 (42%)
Verify the articles which needed to be removed because they wereduplicates
17 (11%) 15 (9%) 51 (32%) 76 (48%)
Verify that the authors’ exclusion criteria were applied appropriatelyfor each excluded study
20 (13%) 28 (18%) 54 (34%) 57 (36%)
Verify that the selection criteria were applied appropriately byreviewing all full-text articles which were included in the finalanalysis
18 (11%) 26 (16%) 53 (34%) 62 (39%)
Verify the appropriateness of studies included in the qualitativesynthesis (systematic reviews)
26 (16%) 53 (33%) 44 (28%) 36 (23%)
Verify the appropriateness of studies included in the quantitativesynthesis (meta-analysis)
27 (17%) 49 (31%) 40 (25%) 43 (27%)
Verify heterogeneity of the included studies 17 (11%) 33 (21%) 47 (29%) 62 (39%)Verify publication bias of the included studies 21 (13%) 32 (20%) 41 (26%) 65 (41%)Verify that the data synthesis from the included individual studies was
appropriate23 (14%) 33 (21%) 44 (28%) 59 (37%)
Verify that the statistical analyses were appropriate by using your ownsoftware or statistical package
4 (2%) 19 (12%) 30 (19%) 106 (67%)
DOI: 10.3109/14767058.2012.755161 Reviews of systematic reviews and meta-analyses 769
J M
ater
n Fe
tal N
eona
tal M
ed D
ownl
oade
d fr
om in
form
ahea
lthca
re.c
om b
y B
iblio
teka
Uni
wer
syte
tu W
arsz
awsk
iego
on
10/2
9/14
For
pers
onal
use
onl
y.
independent analysis key items of the methodology and
statistical analysis. The quality of the peer review process is
extremely important when comes to reviewing systematic
reviews and meta-analyses because the results of such articles
are frequently used by physicians to treat individual patients
and by policy makers to make health care changes in entire
populations.
In general, referees do not undergo any special training on
how to conduct their peer reviews. Since there is no formal
training, most referees ‘‘learn on the job’’. To assist young
referees, we recently published the guidelines on how to peer
review an original research paper [6]. In our guidelines, we
emphasized that a referee should fulfill certain prerequisites
including the willingness to do a thorough review and
availability of time, so that a constructive review can be
accomplished within a reasonable period of time. These two
prerequisites are not difficult to fulfill when referees are asked
by journal editors to review an original research paper
involving a single study.
In reviewing systematic reviews or meta-analyses, a
plethora of studies have to be reviewed by the referee in
order to verify the appropriateness of the studies that were
included (or excluded) from the qualitative or quantitative
synthesis. In addition, the referee should make sure that
heterogeneity, publication bias and appropriate statistical
analyses were done correctly. Such a task would require an
enormous amount of time and it would most likely be
disruptive to the referee’s daily work and life. Thus, the
expectation that the referee will repeat the analysis is not
reasonable. Although our questionnaire did not address
directly the question why the referees do not verify the
methodology and the statistical analysis of the systematic
reviews or meta-analyses articles, it is logical to assume that
the main reason is time constraints. Other possible reasons
could be the lack of statistical knowledge or being unaware of
such an expectation.
It has long been recognized that there are serious
discrepancies between the results of published meta-analyses
and subsequent large randomized-controlled trials in the same
topics [7,8]. Despite the introduction of reporting guidelines
for systematic reviews and meta-analyses articles since 1999,
the contradictions between meta-analyses and randomized-
controlled trials continue to exist [9,10]. Although there are
several possible explanations for these discrepancies, one of
the possibilities may be differences in the quality or reliability
of the peer review process.
There is a fundamental difference between a single study
and a systematic review or meta-analysis with respect to the
complexity of the methodology. In peer-reviewing single
study articles, verification of the methodology and conse-
quently its conduct and statistical analysis are not feasible
because the raw data are not available to the referee; however,
the chance for error(s) is much less as compared to systematic
reviews or meta-analyses. In contrast, in systematic reviews
and meta-analyses, the methodology is more complex relative
to single studies and therefore more prone to errors.
Therefore, verification of the methodology (and subsequent
statistical analysis) are of paramount importance and indeed
feasible since all the necessary data are available to the
referees. The problem is that in order to accomplish a review
of systematic review or meta-analysis papers within a
reasonable period of time, most referees accept the entire
methodology and statistical analysis, exactly as proposed by
the author(s) of the paper without verification despite the
availability of the actual data. Thus, any errors in the
qualitative or quantitative synthesis of the studies by the
authors will escape the detection. Errors in the qualitative or
quantitative synthesis by the author(s) may be unintentional or
due to personal bias. Unchecked inappropriate manipulation
or inclusion (or exclusion) of studies has raised concerns
about the political uses of meta-analysis, a phenomenon
which has been characterized as ‘‘tyranny of meta-analysis’’
[11]. Yet, it is impossible to expect the referee to uncover such
inappropriate manipulations or biases without dedicating an
enormous amount of time. Thus, guidelines for the peer
review of systematic reviews and meta-analyses are required
to avoid the publication of undetected bias because of
suboptimal peer review. In order for the systematic reviews,
meta-analyses or any other qualitative or quantitative
syntheses to be reliable, the review of the evidence should
not be done by the journal’s individual referees but perhaps by
a dedicated group of experts. One example is the Cochrane
Table 3. Logistic regression analysis for clustered data: effects of referee characteristics on the survey responses.
Parameter Estimate Standard error Odds ratio (95% C.I.) p Value
Intercept 0.7271 1.5284 – 0.6343Referee characteristicsAge �0.0071 0.0220 0.99 (0.95, 1.04) 0.7479Gender (female versus male) �0.3085 0.3328 0.73 (0.38, 1.41) 0.3539Degree (MD versus MDþ other degrees) �0.3215 0.2945 0.73 (0.41, 1.29) 0.2750Number of years of reviewing �0.0006 0.0244 1.00 (0.95, 1.05) 0.9806Number of papers reviewed prior year (411 versus55) �0.2643 0.3496 0.77 (0.39, 1.52) 0.4495Number of papers reviewed prior year (6–10 versus55) �0.0666 0.3849 0.94 (0.44, 1.99) 0.8627Response media (web versus hard copy) �0.3479 0.3516 0.71 (0.35, 1.41) 0.3224
Table 2. Characteristics of the responders (n¼ 159).
n (%)
Male 129 (81%)Female 30 (19%)MD degree only 133 (84%)MD and additional degree 26 (16%)Number of papers reviewed the prior year
11 or more 68 (43%)6–10 60 (38%)5 or less 31 (19%)
Age (median, range) 53 (34–80)Number of years reviewing papers (median, range) 17 (3–50)
770 A. M. Vintzileos et al. J Matern Fetal Neonatal Med, 2013; 26(8): 768–771
J M
ater
n Fe
tal N
eona
tal M
ed D
ownl
oade
d fr
om in
form
ahea
lthca
re.c
om b
y B
iblio
teka
Uni
wer
syte
tu W
arsz
awsk
iego
on
10/2
9/14
For
pers
onal
use
onl
y.
Collaboration’s contributors group, which is a mix of
volunteers and paid staff dealing with the methodology of
systematic reviews. Many of these dedicated members are
world leaders in their field of medicine, health policy,
research methodology or consumer advocacy, and are work-
ing in some of the world’s best academic and medical
institutions. Such groups consisting of expert individuals may
have all the time available for a thorough qualitative and
quantitative review and analysis.
One of the important observations in this study is the
relatively low response rate (20%) despite a robust methodol-
ogy using well-established methods to maintain confidenti-
ality and anonymity [12]. Low response rates usually raise
questions regarding the reliability of the survey results.
However, the effect of non-response depends on the extent
and direction to which those not responding are biased. In our
study, a possible bias is that those reviewers who do not use
their own independent analyses may be embarrassed to
respond. However, this bias would suggest that in real life, the
percentage of referees that they do not use their own
independent analyses may even be greater than in our study.
In addition, the response rate after the third mailing was much
less (approximately 50% less) than after the second mailing;
thus, further mailings would not have increased the number of
respondents or alter the results considerably. Another possible
reason for the low response rate may be the small number of
systematic reviews and meta-analyses, as compared to
original articles, published by the two journals. For all the
aforementioned reasons, we believe that the results of this
survey are valid given the wide geographic representation of
the respondents.
Another possible limitation of the study is that the survey
did not explore how much knowledge or experience the
referees themselves had in conducting systematic reviews and
meta-analyses. This can be a crucial issue; however, it is
likely that Editors choose referees who already have published
systematic reviews or meta-analyses. Another limitation is
that, our survey to the referees did not address the possibility
that some of the results of the published systematic reviews
and meta-analyses may have been verified by the statistical
consultants of the journals.
Our survey included only referees from two major
obstetrics and gynecology journals. However, we have no
reason to believe that the results will be different for referees
of other medical journals or specialties. In our view, the
results of this study highlight the need for academic units,
investigators and Journal editors to develop guidelines for
the peer review of systematic reviews and meta-analyses.
These guidelines can be provided to the referees at the time of
invitation. One suggestion may be that referees should be
guided by the Editors to specify, in a check list form, all those
items which they are expected to verify by their own
independent analysis when reviewing systematic reviews
and meta-analyses. Another possibility is to have the authors
of systematic reviews, upon request by the referee, submit a
copy of the abstracted data, a copy of the articles as well as
data obtained from other sources, so that the referee does not
have to spend a lot of time searching the literature and
abstracting data. Editors should also try, whenever possible, to
employ experienced referees for these types of publications.
This will hopefully lead to better quality of medical
information and ultimately patient care.
Declaration of interest
None of the authors have a conflict of interest.
References
1. Chauhan SP, Berghella V, Sanderson M, et al. Randomized clinicaltrials behind level A recommendations in obstetric practicebulletins; compliance with CONSORT statement. Am J Perinatol2009;26:69–80.
2. Moher D, Cook DJ, Eastwood S, et al. Improving the quality ofreports of meta-analyses of randomized controlled trials: theQUOROM statement. Quality of reporting of meta-analyses.Lancet 1999;354:1896–900.
3. Liberati A, Altman DG, Tetzlaff J, et al. The PRISMA statement forreporting systematic reviews and meta-analyses of studies thatevaluate health care interventions: explanation and elaboration.J Clin Epidemiol 2009;62:e1–34.
4. Consultants for the American Journal of Obstetrics andGynecology, 1 January 2005–31 December 2005. Am J ObstetGynecol 2006;194:e49–e63.
5. Editorial consultants. Obstet Gynecol 2009;114:1165–457.6. Vintzileos AM, Ananth CV. The art of peer-reviewing an original
research paper. J Ultrasound Med 2010;29:513–18.7. Villar J, Carroli G, Belizan JM. Predictive ability of meta-analyses
of randomized controlled trials. Lancet 1995;345:772–6.8. LeLorier J, Gregoire G, Benhaddad A, et al. Discrepancies between
meta-analyses and subsequent large randomized, controlled trials.N Engl J Med 1997;337:536–42.
9. Vas J, Aranda JM, Nishishinya B, et al. Correction of nonvertexpresentation with moxibustion: a systematic review and metaana-lysis. Am J Obstet Gynecol 2009;201:241–59.
10. Guittier MJ, Pichon M, Dong H, et al. Moxibustion for breechversion: a randomized controlled trial. Obstet Gynecol2009;114:1034–40.
11. Klein MC. The tyranny of meta-analysis and the misuse ofrandomized controlled trials in maternity care. Birth 2012;39:80–2.
12. Fowler FJ. Survey research methods. Applied social researchmethods series. 4th ed. Thousand Oaks, CA: Sage Publications, Inc;2009:49–67.
DOI: 10.3109/14767058.2012.755161 Reviews of systematic reviews and meta-analyses 771
J M
ater
n Fe
tal N
eona
tal M
ed D
ownl
oade
d fr
om in
form
ahea
lthca
re.c
om b
y B
iblio
teka
Uni
wer
syte
tu W
arsz
awsk
iego
on
10/2
9/14
For
pers
onal
use
onl
y.