should systematic reviews include searches for published errata?

14

copy Health Libraries Group 2004

Health Information and Libraries Journal

21

pp14ndash20

Blackwell Publishing Ltd

Should systematic reviews include searches for published errata

Pamela

Royle

amp Norman

Waugh Department of Public Health University of Aberdeen Aberdeen UK

Abstract

Our objective was to perform a pilot study to estimate the proportion of pub-lished errata linked to randomized controlled trials (RCTs) that are worthwhileobtaining when doing a systematic review

was searched for recordsthat had both lsquorandomized-controlled-trialrsquo in the publication type field andlsquoerratumrsquo in the comments field One hundred records from four general medicaljournals were examined independently from two different perspectives Fromthe information specialistrsquos perspective 74 of the errata were consideredworthwhile obtaining these were mainly errors in tables or figures Another 9described less serious errors but were worth obtaining if easily available Theother 17 were minor errors From the perspective of the experienced reviewerpublic health consultant 5 of errata were classified as likely to affect a meta-analysis and 10 as having significant errors that would affect the interpretationof the RCT but no effect on a meta-analysis 85 were not considered importantenough to affect either About 5 of errata to RCTs appeared to matter in termsof changing the final conclusions of a systematic review However the majorityof errata were considered to be worthwhile obtaining on the basis that havingfull and accurate data can reduce confusion and save reviewers time

Introduction

The volume of medical literature is expanding sorapidly that keeping up with new studies is almostimpossible for individual clinicians and otherusers Consequently the need for good-qualitysystematic reviews of studies has increased andsuch reviews are increasingly used to help informdecisions about health care In the 15 years sinceMulrowrsquos landmark paper on the poor quality ofmedical reviews

1

considerable progress led bythe Cochrane Collaboration

2

has been made onimproving the standard of reviews The method-ology of systematic reviews has been documented

3

and the resulting improvement has been reported

45

This applies particularly to reviews and meta-analyses of randomized controlled trials (RCTs)as these are considered to provide the best evidencefor addressing questions regarding therapeuticefficacy

3

In order to provide as unbiased a sourceof references as possible for those doing systematicreviews the Cochrane Collaboration has developedthe Cochrane Controlled Trials Register (CCTR)the worldrsquos most comprehensive bibliographicdatabase of controlled trials

6

Although there have been studies of the impactof retracted articles on the biomedical literature

78

one issue which seems to have escaped scrutiny sofar is the role of published errata The erratummay appear in print some time after the originalpaper and it may report various errors includingthose in the results These errors might affect not

Correspondence Dr Pamela Royle Department of Public HealthUniversity of Aberdeen Foresterhill Aberdeen AB25 2ZD UKE-mail PRoyleabdnacuk

Systematic reviews and errata

Pamela Royle amp Norman Waugh



21

pp14ndash20

15

only the conclusions of the original study but alsothose of any systematic reviews which include thatstudy As RCTs and systematic reviews may beread and cited by many people and can sometimesform the basis of health policy it would seemimportant that readers are aware of and have easyaccess to these errata

Errata have been defined as published notices ofsubstantive typographical or other inadvertenterrors discovered after an article was printed

9

Since 1987 when a publisher editor or authorhas published a labelled citable erratum to anarticle that was cited in the

databasethe National Library of Medicine (NLM) hasamended the citation of the article with a biblio-graphic reference to the erratum notice in order toalert users and refer them to the source of therevised information

10

When a citable erratum ispublished for an article indexed in the

database the NLM amends the citation of thatarticle with a bibliographic reference to the erratumnotice This particular method of providing a linkbetween an article and its erratum is unique to the

database Brief errata notices are not gen-erally indexed as independent articles Some sub-stantive articles or letters may however comprisepublished errata If so these items will be indexedwith the Publication Type of Published Erratum

It is NLMrsquos policy that errata will be acknow-ledged only if they are printed in a citable formthat is an erratum notice must appear on a num-bered page in an issue of the journal that originallypublished the article

Our aim was to do a pilot study to determine thecharacteristics of published errata linked to RCTsin the

database and to estimate the pro-portion of these that are worthwhile obtainingwhen doing a systematic review

Methods

Obtaining the RCTs and errata in

MEDLINE

(using the WebSPIRS interface bySilverPlatter) was searched from 1995 to June2001 To identify RCTs a search was first madefor records that had the phrase randomized-controlled-trial in the PT (Publication Type) fieldTo identify errata a second search was made for

records with erratum in the CM (comments) fieldA third search was done to select articles publishedin the four general medical journals The

NewEngland Journal of Medicine

(NEJM)

Journal ofthe American Medical Association

(JAMA)

TheLancet

and the

British Medical Journal

(BMJ)were selected by searching for the journal namein the SO (Source) field These four journalswere chosen because they were the four generalmedicine journals with the highest number of totalcitations and they were all available on site

To select RCTs with errata that were publishedin only these four journals the three searchesabove were combined using the AND operator Allrecords thus obtained were downloaded into thebibliographic management software packageReference Manager v95 One hundred RCTs fromthe four journals plus their erratum were thenrandomly selected using computer-generatedrandom numbers for further analysis

Determining characteristics of the errata

The following data were recorded for each RCTand its attached errata the number of errorsmentioned per published erratum the time betweenpublication of the RCT and its erratum whetherthe RCTs and errata were indexed in the CochraneControlled Trials Register (CCTR) and thenumber of citations to each RCT and its erratumin the Science Citation Index The number ofcitations per year was calculated by dividing thenumber of citations by the number of years sincethe article had been published

Classifying the errata

The full papers of the 100 RCTs and their erratawere examined independently by both authors

First from the perspective of an informationspecialist asking whether it would seem worth-while spending the time and money acquiring thepublished erratum for the reviewers in order tominimize their time spent on trying to interpreterroneous or confusing data Each erratum wasassigned to one of three possible categories

1

Worthwhile obtaining

2

Possibly worthwhile

3

Not worthwhile





21

pp14ndash20

16

Second from the perspective of a public healthconsultant who was also an experienced reviewerasking whether the error (or errors) mentionedwould be significant enough to affect either theinterpretation of the paper or results of a meta-analysis (a systematic review where results fromseveral studies are combined and summarizedquantitatively) Each erratum was assigned to oneof three categories

1

Might affect a meta-analysis

2

Would affect the interpretation of the RCT only

3

Would not be important for eitherIn cases where a published erratum mentioned

more than one error the classification was basedon what was judged to be the most important errorin terms of its impact on a systematic review

Statistical analyses

Statistical results are presented as

P

-values and95 confidence intervals A

P

-value of less than005 was judged significant

Results

Calculating the percentage of articles in

MEDLINE

that have errata

The

search retrieved 666 RCTs witherrata added to their bibliographic record Theywere published in 295 different journals It wasfound that 130 (195) of the 666 RCTs with erratawere retrieved from four journals The numbersfrom each journal were Lancet (59) NEJM (35)JAMA (19) and BMJ (17)

The proportion of

articles with errataattached was found to be 06 when all publica-tion types were included When only those withthe publication type of RCT were included theproportion with errata was found to be 12 The

proportions of RCTs assigned errata in the fourjournals were 84 in Lancet and JAMA 83 inNEJM and 56 in BMJ

Measuring characteristics of RCTs and errata

The data in Table 1 shows data measuringcharacteristics of RCTs and their errata The tableshows the maximum and minimum plus both themean and median values for four characteristicsBoth the mean and median are reported becausein situations where the distribution is skewedthe median will be more representative of thedistribution than the mean ie when there areextreme scores the mean will be boosted con-siderably by one very unusual value (or lsquooutlierrsquo)while the median remains unaffected

It can be seen that mean time between publica-tion of the RCT and errata was 35 months butthe range from 1 to 26 months was wide Also itcan be seen that some erratum notices mentionmore than one error with the mean number being16 and the median 1 It was found that 64 of the100 records mentioned only a single error in thepublished erratum therefore 36 mentioned two ormore separate errors per erratum Two of these 36mentioned six different errors and another twomentioned five errors Three RCTs had a seconderratum published a few months after the first (butthese were only counted as one erratum in thisstudy)

There was a large difference in the number ofcitations per year to the RCTs compared with theerrata and in both cases the distribution of thecitations was skewed and the range was wide Itshould be noted that the mean numbers of cita-tions per year to the RCTs and errata are probablyunderestimates because 15 had group authorshiponly in the

database that is they had nopersonal authors in the author field It has recently

Table 1 RCTs and their errata

Characteristic measured Mean Median Max Min

Number of months between publication of the RCT and erratum 35 3 26 1Number of errors mentioned per published erratum 16 1 6 1Number of citations per year to the RCT 28 123 250 0Number of citations per year to the erratum 06 0 8 0





21

pp14ndash20

17

been shown that due to indexing problemscitations to articles with group authorship areundercounted in SCI

11

Classification of errata by an information specialist

The results in Table 2 show that the majority ofthe errata (74) were considered lsquoworthwhilersquoobtaining by an information specialist in thecontext of searching for RCTs for use by sys-tematic reviewers Of the remaining 26 9were considered lsquopossibly worthwhilersquo if they werereadily obtainable free of charge and 17 wereconsidered trivial and lsquonot worthwhilersquo obtainingin this context

The 74 erratum classified as lsquoworthwhilersquoobtaining consisted of a large variety of differenttypes of errors but the majority (4374) were in theResults section in the tables or figures 25 were intables and 19 were in figures The remaining 31erratum considered lsquoworthwhilersquo obtaining con-sisted either of errors in the Methods section or inthe text section of the Results

Analysis of those nine trials classified as lsquopossiblyworthwhilersquo obtaining showed that the majority(59) were errors in the Discussion section orIntroduction The remainder comprised eithervery small errors in tables or inaccurate references

Finally for those 17 classified as lsquonot worth-whilersquo acquiring in this context 14 were errors in

authorship (either omission of names and mis-takes) or errors in affiliation addresses or profes-sional details of authors The remaining threeerrata comprised trivial errors in the Acknow-ledgement Introduction and References

The mean time between publication of the RCTand errata for those classified as (1) lsquoworthwhilersquoobtaining was 37 months versus those classifiedas (3) lsquonot worthwhilersquo was 30 months This dif-ference was not significant (

P

= 043)

Classification of errata by a public health consultantexperienced reviewer

Table 3 shows the results of the classification ofthe erratum by the second author It can be seenthat 5 were classified as possibly affecting ameta-analysis although the relative impactwould be likely to be dependent on the number ofRCTs included Another 10 were classified ashaving significant errors that would affect theinterpretation of the RCT but would have noeffect on a meta-analysis The remaining 85were classified as errors that were not importantbecause the error was (i) not related to the results(ie it was in the author details) or (ii) the error wassignificant but correct data was given elsewhere inthe paper or (iii) the error was in the Discussiononly

Analysis of the five RCTs that had erratum thatcould possibly affect the results of a meta-analysis

Table 2 Results of classification of errata by an information specialist

Table 3 Results of classification of errata by a public health consultantexperienced reviewer

Classification Comments

1 74 Yes worthwhile acquiring The majority comprised errors in data contained in tables or figures2 9 Possibly worthwhile acquiring Mostly contained errors in the introduction or discussion3 17 Not worthwhile acquiring Mostly consisted of errors in authorship such as omissions of authors

or incorrect contact details


1 5 Might affect a meta-analysis (more likely if only a few RCTs included in the review)2 10 Had significant errors that would affect the interpretation of the RCT but no effect on a meta-analysis3 85 Errors were not important because the error was (i) not related to the results or (ii) was significant but

correct figures given elsewhere or (iii) was in the Discussion only





21

pp14ndash20

18

revealed that the errors included such things asomission of a sentence under sample characteris-tics incorrect data that affects size of study popu-lation and has effect on relative risk estimatesmistakes in the y-axis on two figures mistakes inthe results in eligibility criteria and errors in safetyand toxicity section on differences between groups(which reversed the numbers in the two groups)

The erratum of the 10 studies classified ashaving significant errors that would affect theinterpretation of the RCT included incorrect ormissing data in tables and figures incorrect dataon dosages and treatment regimens and errors inthe text which would give the opposite or wrongmeaning

The 15 RCTs assigned as classification (1) or (2)(see Table 2) were cited a total of 1499 times butthe erratum only received 24 total citations Also1015 of these erratum had received no citations

The mean time between publication of the RCTand errata for those classified as either (1) or (2)was 71 months versus 29 months for thoseclassified as 3 This difference was significant (

P

=0000 95 CI = 23ndash60 months) One of the fiveRCTs classified as possibly affecting a meta-analysis took 26 months to appear

Investigation of errata to RCTs in the CCTR database

As CCTR is the single most important datasource of RCTs for those doing systematic reviewsthe presence of the bibliographical details of thepublished erratum in this database was investigatedIt was found that at least one copy of each of the100 RCTs with their erratum information attachedwere in CCTR However in 62 of cases CCTRcontained one additional copy of the RCT lackingthe bibliographical details of the erratum and in15 of cases there were two additional copies ofthe RCT without the erratum details

Costs of accessing the errata

The Lancet BMJ NEJM and JAMA all providedfree electronic access to errata via their websites Itwas found that 64 (711) of the other journalschecked were found to charge for access to errataSome journals printed the incorrect published

version alongside the corrected one which madechecking for impact easier

Discussion

In this pilot study it was found that errata werepublished for an average of 8 of RCTs in fourmajor general medical journals It was also foundthat about 5 of these errata appeared to matterin terms of changing the final conclusions of ameta-analysis and that another 10 would affectthe interpretation of the RCT However in themajority (74) of cases the errata appeared tocontain information important enough to beworthwhile obtaining for systematic reviewerson the basis that knowing the errors can savereviewersrsquo time spent in trying to resolve poten-tially confusing inconsistencies and discrepanciesin the data Even errata in authorship that wereconsidered minor in this study (including miss-pellings or omission of authorsrsquo names) could beproblematic if searches were for papers by orcitations to a particular author

The impact of an erratum on a meta-analysiswould vary according to the magnitude of theerror and the number of RCTs in a systematicreview A small numerical error might not affecteven the original study a large one might affectconclusions drawn from that study but might notaffect those from a meta-analysis if there was a suf-ficient number of other trials unless the RCT withthe error was a particularly large one

The fact that the RCTs with errata that wereconsidered significant had been cited a total of1499 times but their errata had only been cited 24times implies that the errata were relatively over-looked It is most likely that authors were notaware of the errata or if they were they did notbother to cite them It was interesting to note thatit took significantly longer to publish errors thatcould affect a meta-analysis or the interpreta-tion of the RCT compared with other errors(71 months vs 29 months) We do not know thereason for this difference but perhaps some trivialerrors are more noticeable than others such asthose affecting authorsrsquo names

A large proportion of errata appeared to be indata within tables or figures and this often led toinconsistencies with data in the text This gave us





21

pp14ndash20

19

the impression that tables might be less thoroughlychecked at the proof stage than text

It appeared that about twice as many RCTs haveerrata compared with all other publication typesthat is 12 compared with 06 This may bebecause journals are more likely to publish anerratum to an RCT as this study design is consid-ered the lsquogold standardrsquo in hierarchy of evidenceOther possibilities are that there are often a largenumber of authors involved in RCTs and this leadsto more errors made when writing them up or asRCTs are often widely read and cited errors aremore likely to be picked up

Although there was at least one copy of all theRCTs with the bibliographic details of their erratain CCTR it was found that in 77 of cases therewas at least one additional copy of the RCT whichlacked the errata information Therefore there isa considerable chance that someone searchingCCTR for RCTs will miss the copy of the RCTwith the errata information Where duplicates ofRCTs exist in CCTR and one copy contains theerrata reference this copy should remain inCCTR and the copies lacking the errata referenceshould be removed This would help ensure thatthe errata are not missed when searching

One important barrier to obtaining errata iswhether there is a charge for access It was foundthat the BMJ NEJM Lancet and JAMA all gavefree electronic access to errata on their websiteseven though articles in the latter three journalswere only available to subscribers to the journalHowever this was not the case for all journals asseven of the 11 other journals looked at did notprovide free access to errata on their website

This study has a number of limitations We donot know what proportion of errors identified areactually published in journals and if publishedwhat proportion are eventually indexed by

It is possible that the percentage of RCTs witherrata may be an underestimate This could bechecked by hand searching the errata sections inthe journals and then checking how many of theseare actually indexed in

Also although wecould calculate the period between publication ofthe RCT and its errata we do not know the timelag involved in

uploading the amendedRCT record with the errata details attached intothe

database Finally we do not know

how representative the errata in these 4-weeklygeneral medical journals are of other journals andhence if the results can be extrapolated to otherjournals

As only a small sample was tested in this pilotstudy future studies could involve repeating thisstudy on a larger scale using a larger sample ofRCTs and errata published in a wider range ofjournals than just the four investigated above Alsothe study would be improved by using a largernumber of reviewers with different backgroundsand with different levels of experience with review-ing to independently classify the errata

Another possible study could involve taking arandom sample of Cochrane systematic reviewsand checking the included studies and noting thosethat have had errata published Any errata couldthen be obtained and checked by several reviewersto obtain an estimate of whether they felt that itcould have changed the result of the review

Conclusions

While not wishing to increase the considerableworkload of systematic reviews we suggest that itis worthwhile identifying and checking publishederrata Although most errata in this study didnot appear to matter in terms of changing theconclusion of a systematic review it would appearthat just enough might matter to make checkingadvisable Also the majority of the less significanterrors were thought to be potentially misleadingor confusing and obtaining them would appear tobe cost-effective in saving reviewersrsquo time

Therefore the following recommendations aresuggested (i) to ensure that reviewers have accessto as complete and accurate a data set as possibleit is recommended that information specialistsshould endeavour to identify errata to RCTs in thesearch step and ensure that reviewers obtain themespecially in those reviews when only a few trialsare available (ii) where duplicates of RCTs exist inCCTR and one copy contains the errata referencethis copy should remain in CCTR and the copieslacking the errata reference should be removedThis would help ensure that the errata are notmissed when searching (iii) to facilitate access toerrata all journals should provide free electronicaccess to errata to RCTs they have published





21

pp14ndash20

20

Acknowledgements

We wish to thank Liz Hodson for obtaining copiesof the articles used in this study

References

1 Mulrow C D The medical review article state of the science

Annals of Internal Medicine

1987

106

485ndash8

2 Cochrane Collaboration Available from httpwwwcochraneorg

3 Clarke M amp Oxman A D (eds)

Cochrane Reviewerrsquos Handbook 4

1

6 (Updated January 2003)

Available from httpwwwcochranedkcochranehandbookhandbookhtm (accessed 23 April 2003)

4 Olsen O Middleton P Ezzo J Gotzsche P C Hadhazy V Herxheimer A

et al

Quality of Cochrane reviews assessment of sample from 1998

BMJ

2001

323

829ndash32

5 McAlister F A Clark H D van Walraven C Straus S E Lawson F M Moher D

et al

The medical review article

revisited has the science improved


1999

131

947ndash516 Dickersin K Manheimer E Wieland S Robinson K A

Lefebvre C amp McDonald S Development of the Cochrane Collaborationrsquos CENTRAL Register of controlled clinical trials

Evaluation of Health Professionals

2002

25

38ndash647 Budd J M Sievert M amp Schultz T R Phenomena

of retraction reasons for retraction and citations to the publications

JAMA

1998

280

296ndash78 Budd J M Sievert M Schultz T R amp Scoville C Effects of

article retraction on citation and practice in medicine

Bulletin of the Medical Library Association

1999

87

437ndash439 Colaianni L A Retraction comment and errata policies

of the US National Library of Medicine

Lancet

1992

340

536ndash7

10 National Library of Medicine (NLM) Fact sheet errata retraction duplicate publication and comment policy for

Available from httpwwwnlmnihgovpubsfactsheetserratahtml

11 Dickersin K Scherer R Suci E S amp Gil-Montero M Problems with indexing and citation of articles with group authorship

JAMA

2002

287

2772ndash4





21

pp14ndash20

15

only the conclusions of the original study but alsothose of any systematic reviews which include thatstudy As RCTs and systematic reviews may beread and cited by many people and can sometimesform the basis of health policy it would seemimportant that readers are aware of and have easyaccess to these errata

Errata have been defined as published notices ofsubstantive typographical or other inadvertenterrors discovered after an article was printed

9

Since 1987 when a publisher editor or authorhas published a labelled citable erratum to anarticle that was cited in the

databasethe National Library of Medicine (NLM) hasamended the citation of the article with a biblio-graphic reference to the erratum notice in order toalert users and refer them to the source of therevised information

10

When a citable erratum ispublished for an article indexed in the

database the NLM amends the citation of thatarticle with a bibliographic reference to the erratumnotice This particular method of providing a linkbetween an article and its erratum is unique to the

database Brief errata notices are not gen-erally indexed as independent articles Some sub-stantive articles or letters may however comprisepublished errata If so these items will be indexedwith the Publication Type of Published Erratum

It is NLMrsquos policy that errata will be acknow-ledged only if they are printed in a citable formthat is an erratum notice must appear on a num-bered page in an issue of the journal that originallypublished the article

Our aim was to do a pilot study to determine thecharacteristics of published errata linked to RCTsin the

database and to estimate the pro-portion of these that are worthwhile obtainingwhen doing a systematic review

Methods

Obtaining the RCTs and errata in

MEDLINE

(using the WebSPIRS interface bySilverPlatter) was searched from 1995 to June2001 To identify RCTs a search was first madefor records that had the phrase randomized-controlled-trial in the PT (Publication Type) fieldTo identify errata a second search was made for

records with erratum in the CM (comments) fieldA third search was done to select articles publishedin the four general medical journals The

NewEngland Journal of Medicine

(NEJM)

Journal ofthe American Medical Association

(JAMA)

TheLancet

and the

British Medical Journal

(BMJ)were selected by searching for the journal namein the SO (Source) field These four journalswere chosen because they were the four generalmedicine journals with the highest number of totalcitations and they were all available on site

To select RCTs with errata that were publishedin only these four journals the three searchesabove were combined using the AND operator Allrecords thus obtained were downloaded into thebibliographic management software packageReference Manager v95 One hundred RCTs fromthe four journals plus their erratum were thenrandomly selected using computer-generatedrandom numbers for further analysis

Determining characteristics of the errata

The following data were recorded for each RCTand its attached errata the number of errorsmentioned per published erratum the time betweenpublication of the RCT and its erratum whetherthe RCTs and errata were indexed in the CochraneControlled Trials Register (CCTR) and thenumber of citations to each RCT and its erratumin the Science Citation Index The number ofcitations per year was calculated by dividing thenumber of citations by the number of years sincethe article had been published

Classifying the errata

The full papers of the 100 RCTs and their erratawere examined independently by both authors

First from the perspective of an informationspecialist asking whether it would seem worth-while spending the time and money acquiring thepublished erratum for the reviewers in order tominimize their time spent on trying to interpreterroneous or confusing data Each erratum wasassigned to one of three possible categories

1

Worthwhile obtaining

2

Possibly worthwhile

3

Not worthwhile





21

pp14ndash20

16


1


2


3





P


P


Results


MEDLINE

that have errata

The


The proportion of















21

pp14ndash20

17


11








P

= 043)
















21

pp14ndash20

18





P







Discussion









21

pp14ndash20

19













Conclusions







21

pp14ndash20

20

Acknowledgements


References



1987

106

485ndash8




1




et al


BMJ

2001

323

829ndash32


et al




1999

131




2002

25



JAMA

1998

280




1999

87



Lancet

1992

340

536ndash7




JAMA

2002

287

2772ndash4





21

pp14ndash20

16


1


2


3





P


P


Results


MEDLINE

that have errata

The


The proportion of















21

pp14ndash20

17


11








P

= 043)
















21

pp14ndash20

18





P







Discussion









21

pp14ndash20

19













Conclusions







21

pp14ndash20

20

Acknowledgements


References



1987

106

485ndash8




1




et al


BMJ

2001

323

829ndash32


et al




1999

131




2002

25



JAMA

1998

280




1999

87



Lancet

1992

340

536ndash7




JAMA

2002

287

2772ndash4





21

pp14ndash20

17


11








P

= 043)
















21

pp14ndash20

18





P







Discussion









21

pp14ndash20

19













Conclusions







21

pp14ndash20

20

Acknowledgements


References



1987

106

485ndash8




1




et al


BMJ

2001

323

829ndash32


et al




1999

131




2002

25



JAMA

1998

280




1999

87



Lancet

1992

340

536ndash7




JAMA

2002

287

2772ndash4





21

pp14ndash20

18





P







Discussion









21

pp14ndash20

19













Conclusions







21

pp14ndash20

20

Acknowledgements


References



1987

106

485ndash8




1




et al


BMJ

2001

323

829ndash32


et al




1999

131




2002

25



JAMA

1998

280




1999

87



Lancet

1992

340

536ndash7




JAMA

2002

287

2772ndash4





21

pp14ndash20

19













Conclusions







21

pp14ndash20

20

Acknowledgements


References



1987

106

485ndash8




1




et al


BMJ

2001

323

829ndash32


et al




1999

131




2002

25



JAMA

1998

280




1999

87



Lancet

1992

340

536ndash7




JAMA

2002

287

2772ndash4





21

pp14ndash20

20

Acknowledgements


References



1987

106

485ndash8




1




et al


BMJ

2001

323

829ndash32


et al




1999

131




2002

25



JAMA

1998

280




1999

87



Lancet

1992

340

536ndash7




JAMA

2002

287

2772ndash4

should systematic reviews include searches for published errata?

Documents