Vol. 87 No.
2 February 1999
ORAL SURGERY,
ORAL MEDICINE,
ORAL PATHOLOGY,
REVIEW ARTICLE
Meta-analysis in oral health care
Andrew J. Palmer, BSc, MB, BS,a and Peter P. Sendi, MD,b Basel, Switzerland
INSTITUTE FOR MEDICAL INFORMATICS AND BIOSTATISTICS AND UNIVERSITY HOSPITAL
(Oral Surg Oral Med Oral Pathol Oral Radiol Endod 1999;87:135-41)
Meta-analyses are increasingly reported in the medical
literature.1 Publication of meta-analyses occurs with
some regularity in dental periodicals, but until now
such publication has not yet manifested the explosive
growth that has been seen in the medical literature. In
the future, the application of meta-analytic techniques
in the field of dental research is expected to become
more prominent.2,3 Meta-analyses combine and
summarize the results of multiple studies and describe
the true clinical effects of interventions more accu-
rately than do small individual studies. Meta-analysis
has been both hailed1,4 and vilified5-7 within the scien-
tific community. The following article briefly outlines
the history and background of meta-analysis and
describes its methods, strengths, and weaknesses. This
article should help dentists to assess the quality of
published meta-analyses and to understand, interpret,
and apply the results of them in clinical practice.
DEFINITION OF META-ANALYSIS
Meta-analysis applies scientific strategies that limit
bias to the systematic identification and collection,
critical appraisal, and synthesis of all relevant studies
on a specific topic. It uses statistical methods to
combine and summarize the results of several
studies.8 Meta-analyses are therefore useful tools in
decision-making and health technology assessment.
Meta-analysis (also spelled met-analysis and meta-
analysis) may be referred to in the literature under a
aMedical Director, Institute for Medical Informatics and
Biostatistics, Riehen/Basel, Switzerland.
bInternal Medicine Outpatient Department, University Hospital,
Basel, Switzerland.
Received for publication July 6, 1998; returned for revision Aug 28,
1998; accepted for publication Sept 14, 1998.
Copyright © 1999 by Mosby, Inc.
1079-2104/99/$8.00 + 0 7/12/94508
number of synonyms, including statistical overview,
quantitative synthesis, combining, pooling, and
weighted averaging.
A BRIEF HISTORY OF META-ANALYSIS
The first simplistic forms of meta-analyses,
published in agricultural science literature in the
1920s, pooled the results of studies assessing the
effects of different fertilizers on crop yields. During
the 1930s, the first statistical methods for combining
data from different studies were described.9-12 The
first medical meta-analysis was published in 1955
and assessed the size of the placebo effect in various
medical conditions.13 The word meta-analysis first
appeared in 1976 in an education study that assessed
academic achievement in relation to class size.14 The
first meta-analysis in medicine using modern tech-
niques was published in 1977. 15 The Potsdam
Consultation on meta-analysis in 1994 was a land-
mark debate in the modern era of meta-analysis; at
that conference, methodologic standards for the
meta-analysis of randomized controlled trials were
established and guidelines were published.8 In oral
health care, the first meta-analysis indexed in the
Medline database was published in 1989; it evaluated
the effectiveness of tolonium chloride in oral cancer
screening.16 This was followed shortly thereafter by
a meta-analysis that investigated the effects of
different fluoride delivery systems. 17 A search of
Medline for the years 1986 to 1997 using, alone and
in combination, the medical subject head (MeSH)
indexing terms “meta-analysis” and “dentistry” and
using the publication type “meta-analysis,” as well as
free-form topic searches using the words “meta-
analysis,” dentistry,” and “dental,” revealed a total of
47 dentistry-related meta-analyses published in this
time span (Fig 1).
135
136 Palmer and Sendi
REASONS FOR PERFORMING META-
ANALYSIS
Because of the small number of patients in many
clinical trials, studies often do not have enough statis-
tical power to assess the statistical relationships
between interventions and outcomes. Combination of
the results of a number of small studies may increase
the statistical power to detect small to moderate but
significant effects. Meta-analysis may allow a single
estimation of the effect of an intervention in the face of
contradictory results from different studies assessing
outcomes for the same intervention. The summarizing
effects of meta-analysis provide a method for dealing
with the explosion of information that has been seen in
recent decades, thus providing a comprehensive
overview of results in a given area of research. Through
the rigorous process of identification of all existing
data relevant to a particular topic, meta-analyses may
be useful to quantify existing levels of data, thereby
identifying areas in which questions have already been
answered, and to identify data gaps in particular areas
of research. Thus meta-analysis can assist in the plan-
ning of future research. Some government guidelines
recommend meta-analysis as the preferred method of
summarizing evidence of effectiveness and safety of
health technologies in the face of multiple study
results.18 Clinicians, policy-makers, and consumers
can all potentially benefit from an increased under-
standing and application of meta-analyses.8,19,20
The alternative to meta-analysis, the standard narra-
tive review, is fraught with problems and is regarded as
an inadequate quantitative method of summarizing the
effects of interventions.
AN ALTERNATIVE TO META-ANALYSIS: THE
NARRATIVE REVIEW
The standard narrative review, at its worst, is a nonsys-
tematic collection of evidence representing the views of
an author—views based on a possibly biased selection
from the published literature.21 The main limitations of
the narrative review are as follows: (1) subjectivity—no
formal inclusion or exclusion criteria defined for the
studies considered and no rules for analyzing and
assessing the results of the included studies; (2) a scien-
tifically unsound method that often ignores sample size,
effect size, and study quality and design when combining
study results and frequently involves simply comparing
the total numbers of positive and negative studies; (3)
inefficiency, leading to difficulties in accurately dealing
with large numbers of studies covering a particular area
and in determining the relationships and interactions
among the many variables within the studies; (4) nonex-
haustivity with regard to the identification of eligible
studies for inclusion; (5) bias—the tendency to report
ORAL SURGERY ORAL MEDICINE ORAL PATHOLOGY
February 1999
studies written by the reviewer, studies published in
major journals, and studies reporting significant results
that support the reviewer’s claims; and (6) failure,
usually, to produce a quantifiable overall statistical
conclusion or set of meaningful summary results.
Meta-analysis was developed partly as an attempt to
overcome the weaknesses of the standard narrative
review as they have been outlined here. However, a
meta-analysis that is not performed according to a
stringent method may be prone to the same weaknesses
as the narrative review.
META-ANALYSIS’ GREATEST WEAKNESS: BIAS
Because of the retrospective nature of meta-analysis,
it is subject to the dangers of bias. Bias can take many
forms, and each may result in a distortion of the results.
Publication bias is the preferential publication of posi-
tive or statistically significant results over negative or
nonsignificant results because of selective submission
of papers, selective acceptance of papers by journal
editors, and/or selective reporting of results that in the
study were found not to be favorable. Citation bias is
the tendency of review articles to quote supporting
evidence, thereby increasing the likelihood that these
studies will be identified for inclusion in a meta-
analysis as a result of searches through the references
of review articles and that nonsupporting evidence will
not be cited (thus decreasing the likelihood of identifi-
cation of such evidence for inclusion in the meta-
analysis). Multiple publication bias occurs when
results from a single study are included in a meta-
analysis 2 or more times. This may occur when care is
not taken to identify “studies” that report results from
subgroups of a single study in different articles or when
the same study is reported in articles written by
different authors, sometimes with slightly different
results. Inclusion of multiple publications of the results
of the same trial may result in an overestimation of the
benefits of the intervention that is being assessed by
means of the meta-analysis, inasmuch as significant
results from one study are more likely to be reported
multiple times. In addition, large studies are also more
likely to be reported multiple times, which again can
lead to overestimation of an intervention’s effect if
results are inadvertently included a number of times.
Language bias occurs when studies are included on the
basis of the language in which they are written. When
included studies are limited to those published in
English, the results of the meta-analysis may be
distorted because of the omission of studies written in
other languages.22 Finally, researchers performing the
meta-analysis may be consciously or subconsciously
influenced by the source of funding for the study,
which may influence decisions about the studies that
ORAL SURGERY ORAL MEDICINE ORAL PATHOLOGY
Volume 87, Number 2
Fig 1. Number of oral health–related meta-analyses identified in Medline search for years 1986-1997.
are to be included or excluded or the way in which data
are abstracted or interpreted in the meta-analysis.23
STEPS INVOLVED IN PERFORMING A META-
ANALYSIS
1. A priori definition of meta-analysis protocol
The meta-analysis must address a specific clinical
question, which subsequently influences the type of
studies that will be relevant and the type of data to be
synthesized. Therefore, it is essential that the protocol
be developed and clearly stated before the commence-
ment of the processes involved in meta-analysis. The
question(s) to be answered by the meta-analysis should
be formulated in terms that address parameters of clin-
ical significance, such as effectiveness and safety
outcomes. It must specify the study identification and
retrieval methods to be used, as well as criteria for
study inclusion or exclusion from the meta-analysis.
The inclusion/exclusion criteria must stipulate the
study design to be included (usually randomized,
controlled trials), the patient population to be assessed,
the type of disease, the classification of the disease to
be investigated by stage or severity and duration, the
treatment or intervention to be assessed (in terms of
treatment /intervention type and duration and whether
the treatment involves monotherapy or a combination
of therapies), prior or concomitant treatment received
by the patient population, minimum amount of effec-
tiveness and/or safety data to be included in the study
report, and minimum quality standard. Multiple publi-
Palmer and Sendi 137
cations—ie, the same study reported by different
authors and/or under different titles and/or in different
journals—must be identified, documented, and
excluded, as must any report that involves a patient
group that is a subset of a patient population identified
in another published report. The inclusion of multiple
publications may bias the outcome of the meta-
analysis.24 The reason(s) for inclusion or exclusion of
each study assessed must be explained. The methods
by which decisions on inclusion or exclusion from the
meta-analysis were made must be transparent, and the
results must be reproducible. The protocol must
specify the data to be abstracted from each source and
the form the abstracted data must take—ie, a standard-
ized data extraction form must be developed. The
protocol must specify the methods of source quality
rating. Any changes to the protocol after the
commencement of the meta-analysis process should be
reported and clearly justified.
2. Study search and data identification
The main goal of the study search is to identify all
appropriate and relevant data, published and unpub-
lished, for possible inclusion in the meta-analysis.
Omission of studies through failure to identify important
relevant studies leaves the way open for criticism and
accusations of bias or deliberate manipulation of results.
Sources available for searching include electronic data-
bases such as Medline, Grateful Med, Embase, and
Cancer-Lit. Each of these electronic databases differs
138 Palmer and Sendi ORAL SURGERY ORAL MEDICINE ORAL PATHOLOGY
February 1999

Table I. Checklist of appropriate steps in performance of meta-analysis

1. A priori definition of meta-analysis protocol
Addresses specific question
Specifies study search procedure
Specifies inclusion/exclusion criteria
2. Study search and data identification
Multiple electronic database searching, hand-searching, reference list scanning, consultation with experts
3. Data extraction
Extracted by more than 1 person following written protocol, using standardization data extraction forms
4. Quality assessment of retrieved articles
Uses validated source quality rating score
5. Test for heterogeneity
Defines a priori significant level of heterogeneity
Uses relevant statistical test for heterogeneity—eg, chi-square test
Identifies/explains causes of significant heterogeneity
Uses random effects model if studies shown to be heterogeneous
Interprets results cautiously if studies heterogeneous
6. Statistical combination of data
Justifies use of fixed or random effects model in presence/absence of heterogeneity
Calculates clinically relevant outcomes—eg, odds ratios, relative risks, number needed to treat
Calculates confidence intervals/measures of variance of outcomes
7. Tests for publication bias
Funnel plots
8. Sensitivity analysis
Cumulative meta-analysis based on publication date, event rates, study size, effect size, study quality
9. Presentation of results
Includes tabulated summary of included/excluded trials, tabulated and graphic presentation of results
10. Interpretation and discussion of results
Interprets results in terms of current clinical practice
Discusses strengths and limitations
Discusses significance of results in comparison with individual included studies and other meta-analyses

from the others in content, time frame covered, and
primary focus. Medline, for example, focuses mainly on
clinical medicine journals published since 1964;
Embase, on the other hand, indexes journals, conference
proceedings, books, and dissertations since 1974.
Accordingly, there may be only 30% to 40% overlap
between Medline-referenced and Embase-referenced
articles (data presented during a course at Deutsches
Institut für Medizinische Dokumentation und
Information [German Institute for Medical
Documentation and Information], Cologne, Germany,
February 1995). Review articles, conference proceed-
ings, symposia proceedings, industry reports, books,
memoranda from expert consultations, and results of
manual searching of literature lists of retrieved articles
should be scanned for additional potential data sources.
In cases in which language limitations have been applied
to the search, the nature of these restrictions must be
specified. The protocol for the search must be docu-
mented and transparent and the results reproducible.
3. Data extraction
Data extraction should be performed by 2 or more
blinded extractors following a written protocol.
Multiple extraction, in which 2 or more investigators
abstract the data from each study independently and
compare their results on completion, facilitates repro-
ducibility of the data extraction process, identifies data
entry errors, and identifies areas of ambiguity or confu-
sion, which may be resolved through discussion at the
completion of data extraction. The method of data
extraction must be documented a priori, and extraction
should be completed through use of a standardized data
extraction form. Such a form is used to record and
present the extracted data in a systematic way under the
broad headings of a summary of trial characteristics,
effectiveness parameters, clinical safety parameters,
and laboratory safety parameters; it should also include
the source quality rating score. Where data are incom-
plete or unclear, discussion with the study authors is to
be encouraged. Where missing information is unob-
tainable, it should be explicitly stated that this infor-
mation was not stated or was not provided.
4. Quality assessment of retrieved articles
Each data source should be assessed in terms of the
quality of the data presented. Source quality rating
should be performed through use of a validated stan-
ORAL SURGERY ORAL MEDICINE ORAL PATHOLOGY
Volume 87, Number 2
dardized score that assesses the quality of study design,
statistical analysis of the effectiveness and safety para-
meters, and presentation of the study results.25 The
method for the source quality rating must be trans-
parent and reproducible. The source quality rating
score can be used to exclude studies with quality scores
below a certain predefined value, or it can be used as a
weighting factor in the analysis. Finally, the source
quality rating score can be used in the sensitivity
analysis, in which studies of lower quality are removed
from the meta-analysis and the effects on the meta-
analysis results are examined. (Sensitivity analysis is
discussed in more detail in step 8.)
5. Tests for heterogeneity
A group of studies is considered to be homogeneous if
all of them are attempting to estimate or observe the
same true effect and if variability between results of the
different studies is due to random error only (intrastudy
variability). Heterogeneity occurs when the true effect
differs between studies and is therefore represented by a
distribution of values rather than a single number. A
statistical test for heterogeneity calculates the proba-
bility that differences between the observed results of
the individual studies could have occurred by random
variation alone. If that probability is below a certain
level—eg, if it is less than 0.05—then the studies are
said to be heterogeneous. A statistical definition of
significant heterogeneity (eg, P < .05) should be speci-
fied a priori; an assessment of heterogeneity of the
studies included in the meta-analysis should be then
made through use of an appropriate statistical test, such
as the chi-square test.8 Where significant heterogeneity
is found to exist, the sources of heterogeneity should be
identified and explained. Potential sources of hetero-
geneity include differences between studies in any of the
following areas: patient baseline and demographic char-
acteristics; study protocols, doses, and the timing of
administration of interventions; rules for reporting, clas-
sification, and grading of side effects experienced by
patients; rules for dropping out because of ineffective-
ness and/or toxicity; and study conditions, clinical tech-
niques, and patient management over the time period
elapsed.26 Pooling of results and interpretation should be
done only with caution and only after the existence of
heterogeneity between included studies has been identi-
fied.26 If heterogeneity cannot be explained, this should
be acknowledged, and the subsequent limitations
imposed by this should be identified and explained.
In some cases, after literature searching, data extrac-
tion, data source quality rating, and assessment for
heterogeneity, meta-analysis may not be possible.
Among the possible reasons for this are that insufficient
information is available, that data are of poor quality,
Palmer and Sendi 139
Table II. Minimum requirements for good-quality
meta-analysis
Identification and inclusion of all relevant studies/data
Exhaustive search for both published and unpublished data
Similar treatments in included trials
Are all interventions comparable and results combinable?
Similar doses of drugs/intensities of interventions
Similar timing of administration of interventions
Comparable patients in included trials
Age
Gender
Disease duration
Disease severity
Coexisting illnesses
Concomitant medications influencing effectiveness and
safety
Comparable outcomes in included trials
Studies assess same outcome measures of patient response
Outcomes must be measured by comparable methods
Similar follow-up lengths in included trials
Decreasing effectiveness of interventions over time
Increasing dropouts because of ineffectiveness over time
Increasing percentage of patients experiencing side effects
over time
Increasing dropouts because of side effects over time
Adequate quality of included trials
“Garbage in = garbage out”
and that available data are not similar enough to allow
synthesis. This result, with the reason for the inability to
perform the meta-analysis, should be reported.
6. Statistical combination of data
The statistical analysis of the data abstracted from the
studies included in the meta-analysis should cover all
relevant and clinically useful measures of treatment
effectiveness and safety and should be expressed in terms
of a clinically meaningful outcome, such as odds ratio,
relative risk, absolute probabilities, or number needed to
treat (ie, how many patients need to be treated to prevent
one patient from experiencing an event). The type of
statistical model used should reflect the nature of the
trials analyzed, and the use of a fixed-effects or random-
effects model should be specified and justified. The
fixed-effects method provides an estimate of single
underlying effects, taking only intrastudy variability into
account when calculating pooled values weighted by
study size. It assumes that all studies are observing the
same true effect and that variability is due to random
error only. The random-effects model, which takes inter-
study variation into account, provides a more conserva-
tive estimate (usually with wider confidence intervals)
and may be used when included studies are more hetero-
geneous in nature. The random-effects method assumes
that the true effect differs between studies and is therefore
represented by a distribution of values rather than by a
140 Palmer and Sendi
single number. The results should be expressed with
confidence intervals and should include an analysis and
statement of the significance level of the results obtained.
7. Tests for publication bias
Publication bias arises whenever studies with positive
and/or statistically significant results are more likely to
be published than those with negative and/or nonsignifi-
cant results; it is brought about mainly by selective
submission and acceptance/publication of papers. This
may lead to an overestimation of the positive effects and
an underestimation of the negative effects of interven-
tions assessed. A simple graphic test for publication bias
is the funnel plot, in which individual trial effect esti-
mates are plotted against sample size.27 In the absence of
publication bias, the funnel plot of a large number of
trials can be expected to be symmetrically distributed in
an inverted funnel shape around a chosen reference
point.; studies with small sample sizes will be scattered
broadly and symmetrically around the base of the
funnel, with the degree of scatter lessening with
increasing study size (and accuracy). If, however, publi-
cation bias is present, the plot will be asymmetrical,
typically with an absence of small studies at the base of
the funnel on the side representing no effect of the inter-
vention. However, publication bias may not be the only
cause of asymmetry, which can also arise as a result of
language bias, citation bias, multiple publication bias,
poor methodologic design (partiuclarly in small studies),
and heterogeneity of the included studies. Moreover, in
meta-analyses with relatively small numbers of trials in
each treatment arm, asymmetry may arise by chance.27
8. Sensitivity analysis
Sensitivity analysis should be performed to assess
the robustness of the outcomes of the meta-analysis
with respect to changes in the studies included. This
can be performed in a number of ways, including
assessment of the effect of sequential removal of trials
identified as a cause of heterogeneity (outliers) and
cumulative meta-analysis, in which the effects of
adding studies sequentially to the meta-analysis on the
basis of publication date, control group event rate,
study size, size of difference between treatment and
control, or study quality are assessed.28
9. Presentation of results
The results should include a structured summary or
abstract. The key elements, including study design,
number of patients starting and completing each
included trial, form of treatment and treatment regime,
and effectiveness and safety outcome measures of the
individual studies included in and excluded from the
meta-analysis, should be presented in tabulated form.
ORAL SURGERY ORAL MEDICINE ORAL PATHOLOGY
February 1999
The results of the meta-analysis should be presented as
clinically useful measures, such as the number of
patients needed to treat to prevent an event,29 and
should be presented in tabulated and graphic forms,
with confidence intervals clearly stated.
10. Interpretation and discussion of the results
The results of the meta-analysis should be discussed
in terms of current clinical practice. The strengths and
weaknesses of both the individual studies included in
the meta-analysis and the limitations of the meta-
analysis itself should be acknowledged and discussed.
The results of the meta-analysis and the power of these
results should be discussed in the context of the results
of the individual studies included and in comparison
with other meta-analyses that have been performed.
CONCLUSION
When reading a meta-analysis and attempting to judge
its quality, the reader can use the points listed in Table I
as a checklist of appropriate steps that should have been
carried out by the study authors. The reader can use the
minimum requirements for a good-quality meta-analysis
listed in Table II to further assist in judging the quality
of a meta-analysis. If one or more steps have been
omitted or if criteria have not been fulfilled, then the
quality of the meta-analysis, and therefore the applica-
bility of the results, conclusions, and recommendations
drawn from it, may be called into question.
Meta-analysis is a state-of-the-art method of summa-
rizing the increasing amounts of information that are
generated from clinical research, but any meta-analysis
shares the weaknesses of the original studies and also has
a potential for bias because of the retrospective nature of
the technique. Understanding the principles and methods
involved in meta-analysis will help the dental practitioner
to assess the value of meta-analyses encountered in the
literature. When high-standard meta-analyses are used in
clinical decision-making, an improvement in the quality
of dental care may be anticipated.
REFERENCES
1. Egger M, Smith GD. Meta-analysis: potentials and promise.
BMJ 1997;315:1371-4.
2. Cohen PA. Meta-analysis: application to clinical dentistry and
dental education. J Dent Educ 1992;56:172-5.
3. Proskin HM, Volpe AR. Meta-analysis in dental research: a para-
digm for performance and interpretation. Journal of Clinical
Dentistry 1994;5:19-26.
4. Mulrow CD, Cook DJ, Davidoff F. Systemic reviews: critical
links in the great chain of evidence. Ann Intern Med
1998;126:389-91.
5. Rosendaal FR. The emergence of a new species: the professional
meta-analyst. J Clin Epidemiol 1994;47:1325-6.
6. Shapiro S. Meta-analysis/shmeta-analysis. Am J Epidemiol
1994;140:771-8.
7. Feinstein AR. Meta-analysis: statistical alchemy for the 21st
century. J Clin Epidemiol 1995;48:71-9.
ORAL SURGERY ORAL MEDICINE ORAL PATHOLOGY Palmer and Sendi 141
Volume 87, Number 2

8. Cook DJ, Sackett DL, Spitzer WO. Methodologic guidelines for
systematic reviews of randomized control trials in health care
from the Potsdam Consultation on meta-analysis. J Clin
Epidemiol 1995;48:167-71.
9. Tippett LHC. The methods of statistics. London: Williams and
Norgate; 1931.
10. Fisher RA. Statistical methods for research workers. London:
Oliver and Boyd; 1932.
11. Cochran WG. Problems arising in the analysis of a series of
similar experiments. Journal of the Royal Statistics Society
(Series B) 1937;4:102-18.
12. Pearson ES. The probability integral transformation for testing
goodness of fit and combining independent tests of significance.
Biometrika 1938;30:148.
13. Beecher HK. The powerful placebo. JAMA 1955;159:1602-6.
14. Glass G. Primary, secondary and meta-analysis of research.
Educational Research 1976;5:3-8.
15. Chalmers TC, Smith H, Kunzler AM, Matta RJ. Evidence
favoring the use of anticoagulants in the hospital phase of acute
myocardial infarction. N Engl J Med 1977;297:1091-6.
16. Rosenberg D, Cretin S. Use of meta-analysis to evaluate tolo-
nium chloride in oral cancer screening. Oral Surg Oral Med Oral
Pathol Oral Radiol Endod 1989;67:621-7.
17. Beiswanger BB, Stookey GK. The comparative clinical cario-
static efficacy of sodium fluoride and sodium monofluorophos-
phate dentifrices: a review of trials. ASDC J Dent Child
1989;56:337-47.
18. Swiss Federal Office of Social Security. Manual for the stan-
dardisation of clinical and economic evaluation of medical tech-
nology. Bern: The Office; 1996.
19. Bero LA, Jadad AR. How consumers and policymakers can use
systematic reviews for decision making. Ann Intern Med
1997;127:37-42.
20. Cook DJ, Mulrow CD, Haynes RB. Systematic reviews:
synthesis of best evidence for clinical decisions. Ann Intern Med
1997;126:376-80.
21. Gotzsche PC. Reference bias in reports of drug trials. BMJ
1987;295:654-6.
22. Gregoire G, Derderian F, Le Lorier J. Selecting the language of
the publications included in a meta-analysis: is there a Tower of
Babel bias? J Clin Epidemiol 1995;48:159-63.
23. Bero LA. Why review articles on the health effects of passive
smoking reach different conclusions. JAMA 1998;279:1566-70.
24. Tramèr MR, Reynolds DJM, Moore RA, McQuay HJ. Impact of
covert duplicate publication on meta-analysis: a case study. BMJ
1997;315:635-40.
25. Moher D, Jadad AR, Nichol G, Penman M, Tugwell P, Walsh S.
Assessing the quality of randomized controlled trials: an anno-
tated bibliography of scales and checklists. Control Clin Trials
1995;16:62-73.
26. Mulrow CD, Langhorne P, Grimshaw J. Integrating heteroge-
neous pieces of evidence in systematic reviews. Ann Intern Med
1997;127:989-95.
27. Egger M, Smith GD, Schneider M, Minder C. Bias in meta-
analysis detected by a simple, graphical test. BMJ
1997;315:629-34.
28. Lau J, Schmid CH, Chalmers TC. Cumulative meta-analysis of
clinical trials builds evidence for exemplary medical care. J Clin
Epidemiol 1995;48:45-57.
29. McQuay HJ, Moore RA. Using numerical results from systematic
reviews in clinical practice. Ann Intern Med 1997;126:712-20.
Reprint requests:
Andrew J. Palmer, BSc, MB, BS
Medical Director
IMIB (Institute for Medical Informatics and Biostatistics)
Bachtelenweg 3
4125 Riehen/Basel
Switzerland

CALL FOR LETTERS TO THE EDITOR

A separate and distinct space for Letters to the Editor was established by Larry J. Peterson, editor in
chief of Oral Surgery, Oral Medicine, Oral Pathology, Oral Radiology, and Endodontics in his Editorial
in the January 1993 issue.
Dr Peterson also encouraged brief reports on interesting observations and new developments to be
submitted to appear in this letters section as well as Letters commenting on earlier published articles.
Please submit your letters and brief reports for inclusion in this section. Information for authors for
the Journal appears in this issue of Oral Surgery, Oral Medicine, Oral Pathology, Oral Radiology, and
Endodontics.
We look forward to hearing from you.