Systematic review search filter for Ebsco APA PsycInfo

I needed one of these but couldn't find one.   I did find a multi-database filter for Ovid, including APA PsycInfo, from Canada's Drug Agency, CADTH, so modified it to use in EbscoHost.

I could find no relevant publication types to match line 1 of the original filter, so replaced it with a modification of line 17, and then omitted line 17, so my filter has 22 and not 23 lines.    I changed Ovid's adj3 into N2 (and changed any other adjn in the same way).

This is the result.   I have not validated or tested this filter, although the original one is labelled as having had "pragmatic" testing.

S1 MR (systematic review or meta analysis)

S2 DE "systematic review" OR DE "meta analysis"

S3 ((systematic* N2 (review* or overview*)) or (methodologic* N2 (review* or overview*)))

S4 ((quantitative N2 (review* or overview* or synthes*)) or (research N2 (integrati* or overview*)))

S5 ((integrative N2 (review* or overview*)) or (collaborative N2 (review* or overview*)) or (pool* N2 analy*))

S6 (data synthes* or data extraction* or data abstraction*)

S7 (handsearch* or hand search*)

S8 (mantel haenszel or peto or der simonian or dersimonian or fixed effect* or latin square*)

S9 (met analy* or metanaly* or technology assessment* or HTA or HTAs or technology overview* or technology appraisal*)

S10 (meta regression* or metaregression*)

S11 (meta-analy* or metaanaly* or systematic review* or biomedical technology assessment* or bio-medical technology assessment*)

S12 (medline or cochrane or pubmed or medlars or embase or cinahl)

S13 (cochrane or (health N1 technology assessment) or evidence report)

S14 (comparative N2 (efficacy or effectiveness))

S15 (outcomes research or relative effectiveness)

S16 ((indirect or indirect treatment or mixed-treatment or bayesian) N2 comparison*)

S17 (multi* N2 treatment N2 comparison*)

S18 (mixed N2 treatment N2 (meta-analy* or metaanaly*))

S19 umbrella review*

S20 (multi* N1 paramet* N1 evidence N1 synthesis)

S21 (multiparamet* N1 evidence N1 synthesis)

S22 (multi-paramet* N1 evidence N1 synthesis)

S23 S1 OR S2 OR S3 OR S4 OR S5 OR S6 OR S7 OR S8 OR S9 OR S10 OR S11 OR S12 OR S13 OR S14 OR S15 OR S16 OR S17 OR S18 OR S19 OR S20 OR S21 OR S22

Speling, or you say caesarean, I say caesarian

I always mention in literature searching classes - if you don't find many results, check your spelling.   And my slide says speling, ha ha. 

It all started with a tweet from Janice Kung, Health Sciences Librarian, @janicekung, at the University of Alberta, pointing out the need to search for typos and spelling mistakes when searching the literature.  Tom Roper, retired health librarian, @tomroper, who I follow, replied, which is how I saw it, and included his video on the topic.  The video explores the reasons for some misspellings, including authors thinking a word is Latin in origin when it is Greek.   

One of Tom's examples is caesarean, which I have to admit is a word I always have to think about when spelling or searching.  Being midwifery librarian, it's a word I encounter quite often.

Another example mentioned by Megan Kennedy, @MeganSansH, also a Health Sciences Librarian at the University of Alberta, was the various spellings of Cinahl.

I had noticed before when searching that you sometimes retrieve results when you spell something incorrectly.   But I had not thought of the implications of that.   If you are doing a systematic review, shouldn't you therefore include typos and spelling mistakes?   What would you miss if you did not include accidental spelling variants?   How would you know what they were?

In Cinahl, yesterday: 

Cinahl - 48999 results

Cinhal - 664

Cinalh (not thought of that one before) - 34

In Ovid Medline, yesterday: 

Cinahl.ti,ab. - 36010

Cinhal.ti,ab. - 1069

Cinalh.ti,ab. - 46

What about caesarean?   Cinahl seemed to search caesarean when I searched for caesarian, finding the same number of results, so I looked just at Ovid Medline:

Ovid MEDLINE(R) ALL <1946 to February 03, 2022>

1 caesarean.ti,ab. 23038

2 cesarean.ti,ab. 44059

3 caeserean.ti,ab. 32

4 ceserean.ti,ab. 7

5 caesarian.ti,ab. 1436

6 cesarian.ti,ab. 474

7 caeserian.ti,ab. 12

8 ceserian.ti. 0 (I think when I did this yesterday there was one)

There may well be others.   To help find them all, I tried wildcards.  In Ovid: 

? stands for one character or none, and can be used in the middle of a word or at the end.

# stands for one character in the middle or at the end.

* stands for any number of characters at the end, as does $, and both can be qualified with a number, so *3 or $3 would find the word stem followed by up to three characters.

c#?s#r#an.ti,ab. finds 67579.

or/1-8 finds 67506, which means that term with wildcards finds an additional 73.   Some I can't explain, some are personal names like Cesarman and others are the Spanish word conservan.  

But I can see ceasarean, cessarian and casarean.  

We could truncate our wildcarded term, c#?s#r#an*.ti,ab., which finds 68360.

I think too much detail for most of my sessions, but worth a mention to anyone doing a systematic review.   

Oh yes, and then on the same day courtesy of Improbable Research, @improbresearch, brian injury.  A few other examples of this and brian damage, in Medline...  This one is just a typo, I am sure, not a word like caesarean whose spelling is difficult.

What's in a name?

You have waited a long time for this post, I notice!   I am sure you have had plenty to do while waiting.

In their article Chronic renal confusion, Chi-yuan Hsu and Glenn Chertow looked at the wide variety of terms used to describe the same level of kidney function, and the wide range of renal function described by the use of a term like "chronic renal failure".   They examined a sample of literature to reach their conclusions (1).

They argue that some terms should be dropped - a term like "chronic renal failure" has negative connotations for patients, and a term like "pre-dialysis" might discourage practitioners from attempting other sorts of treatment.   They argue for the use of "chronic renal insufficiency" instead. 

I am not sure people have been listening in the intervening 15 years!    Maybe it is too big an "ask".  There are still lots of terms for the same thing, and people do use terms interchangeably (their example is "chronic renal failure" and "end stage renal disease").

Their article shows two things:

1.  The wide variety of terms used to describe the same thing demonstrates why you need to include synonyms in a search.  This is particularly important if you are doing a systematic review, or other sort of systematic search.  You must try to use all the terms that authors might have used, however odd that usage might be.

2. It is important to review your search results, to remove those where the authors' definition of that condition does not match yours.   There is not a lot you can do in the search strategy to deal with this.

I think point 2 is the responsibility of the requestor of the search, or the reviewer of the results, and not the information specialist/librarian.

Point 1 is definitely the responsibility of the searcher, so often, the librarian, with help from the requestor or reviewers.

As an example, I devised a search strategy for chronic kidney disease, which is on the UHL Clinical Librarian Blog.

(1) Hsu CY, Chertow GM. Chronic renal confusion: insufficiency, failure, dysfunction, or disease. Am J Kidney Dis. 2000;36(2):415-8

Reusing a search strategy

Other people's search strategies can be a useful source of search terms, and synonyms.   I wrote in an earlier post that you should not just reuse a strategy.  Using an existing strategy as a source of ideas is ok, but you need to understand how it works, and make sure that there is nothing missing. 

But, of course, you might be updating a review.  In which case, you would want to rerun the existing search.  I think these are the issues to look for, assuming that you are not changing the research question or topic of the review.

1.  Is there anything that should have been done differently?  Is there anything missing?  Is it ok to amend the strategy, just because you now see a better way to do it?    At the moment, I think this is a matter for your supervisor, tutor, or someone with more knowledge of systematic review methodology than me!

2.  Are there any new subject headings to include?   MeSH is updated annually, and perhaps since the search was first run, there is a new heading.  I think in these cases, it is ok to amend.   For example, there is a MeSH term Patient Handoff/.  It was first used in 2013.  If the original search predates that, it will not include the term, so you ought to add it.   What should you do to ensure you still find older references?  How are they indexed?   Checking MeSH (for example, in PubMed's MeSH database) may tell you. 

3.  Was the original search run in the same interface?   Perhaps it was run in Medline via Ebsco, and you have access to Medline via Ovid.  If this is the case, you need to translate the strategy and make sure that proximity operators and truncation are correct for the new interface.   I wrote an earlier post about this sort of thing.  It is also possible that the capabilities of the interface have changed since the search was first run, and this is a "translation" as well.

Translating search strategies

This post has been updated - look at the updated one instead.

Identifying search terms

For any literature search, you will need to identify your search terms.  What words or phrases are you going to search for?   What thesaurus terms are you going to choose?

If you are doing a detailed search for evidence for a systematic review, or doing a systematic search for some other reason, identification of search terms will include identification of synonyms, that is, alternative terms for the concept you are interested in.

Here are some possible ways to identify thesaurus and free text terms, and synonyms. 


Scoping search

Use a PICO grid or similar to identify the search concepts, and search for those.   For example:

Search question: antibiotics to treat otitis media in children under 5.

P - under fives with otitis media
I - antibiotics
C - no stated comparision
O - no stated outcomes

So, search for otitis media and antibiotics, and see what terms are used in titles and abstracts of the articles you find.

Existing strategies

An existing systematic review should include the search strategy used.  Don't feel you have to copy it (of course, if you are doing this exercise for a dissertation, assignment or thesis, you should not and must not copy it!), but it may give you useful terminology.

Background research

Use clues from any background reading you have done about the topic, textbooks, reputable websites, NICE Evidence Search for any clinical topic.

Drug information sources

The British National Formulary is included in NICE Evidence Search, but also try Electronic Medicines Compendium, the European Medicines Agency, and Drugs.com.  Include generic names (long acting insulin) and brand names (glargine), include classes of drugs (antibiotics) and specific types (penicillin).  eMC is a collection of drug information sheets and patient information leaflets.

“Used for” in thesauri 

In a search of Medline, Embase or Cinahl, you must identify relevant thesaurus terms.   Your scoping search will help you - in Ovid or the NHS interface, when you type in a search term, thesaurus terms are suggested to you.  But use the "scope note" and check the "used for" entries.  Articles using those terms in title or abstract will be indexed using the thesaurus term you are looking at.   Therefore, those "used for" entries are useful synonyms to include in your search.

Other people

If you are working on a research project as part of a team, then team members will have suggestions for search terms.   So will patients with that condition, and the team may be able to ask them.  If you are working on an assignment, dissertation or thesis, you may be on your own (although if you have a supervisor, ask if they can advise you).   It is worth talking to others, and checking that you understand the search keywords in the same way as them, and that you know what they mean. 

Known items

You looked at titles and abstracts of items that appeared in the results of your scoping search.  But have a look at the titles and abstracts, and full text, of items that you know about already, and items you find in your actual search.   And look at the titles of references in the bibliographies of relevant items.

MeSH on Demand

This is a National Library of Medicine service, which identifies MeSH terms in text that you enter.  I have not yet tried this, but it seems to have possibilities.  I have written a separate post about this.

More about abstracts that make it no further

Or do they?  This carries on from an earlier post.

I searched Conference Proceedings Citation Index (Web of Science) for the phrase "antibiotic resistance" in titles, published in 2007.  Many of the results were not abstracts, but papers presented at conferences, something useful to remember.   Refining to meeting abstracts left 34 items.

I searched Science Citation Index and Conference Proceedings Citation Index, for all years, to see if I could find a full paper presenting the same results.  Of course, in CPCI there are no abstracts of abstracts, so without checking the abstract, all I had to compare was the authors and title.   I searched for the first author, using their first initial and truncation, and a word or phrase that seemed significant out of the title.  I was assuming, of course, that the first author was involved in any full paper, which might not be true.  It might be better to use all the authors, combined with OR.  Some of them would be the same.   I am also assuming that the full paper would have a similar title to the conference abstract.

For three of the abstracts, there were no other results.

For one, there was one possible match, involving all four authors (but not listed in the same order), but the full paper had a more specific title, although one that could fit the abstract.  The abstract was titled "Antibiotic resistance and penicillin tolerance in clinical isolates of group B streptococcl", and the paper (published two years later) "Antibiotic resistance and penicillin tolerance in ano-vaginal group B streptococci".  The two were not in the same journal.

Interestingly, although a bit of an aside, that abstract title does really say "streptococcl" with an L.  If you search for titles containing that word, you find one other, and WoS suggests that you meant "streptococcAl", with a link to 9000+ results.   I shall report both streptococcl titles!  You would find the paper with other search terms, as correct spellings would appear in the abstract and keywords, but possibly a different story with the meeting abstract.

For one other of the five I searched, I am fairly confident I found a match, different journal, published in 2013, same two authors in the same order, and an abstract that indicates that the data was collected in 2005-2006.

So tiny sample, and problems with methodology, but 1/5 of the abstracts made it to a full paper (and therefore a presentation of the full data), and another 1/5 might have done but we would need to check both items in full to be sure.   The other 3 (methodological problems excepted) went no further. 

Another thought about contacting the authors - I was wondering what exactly the students had said, or asked.   Would it make a difference how specific they were when asking for data, or what language they used?

Abstracts that make it no further

I recently ran a literature searching tutorial for MSc students undertaking a systematic review as a piece of course work.  I'd seen many of them individually, as had colleagues, and so offered the tutorial to make sure we had not missed anyone.  Of course, those who turned up had already been to see us individually!   The keen ones take up all the offers and the challenge is to reach those who are not so keen, don't feel they can ask for help, or think they are ok.  

I had it mind to talk about general things, but began by asking if they had any specific questions.  This was one: if you find a conference abstract, and you want to use the data in your analysis, and the data in the abstract is not detailed enough, where do you go to get the detailed data?  

I said that you would have to look for an article about the same study.   But how easy is this article to find?   Presumably it has the same authors and title, but maybe not.   If the abstract is of a trial, and the trial is recorded in a registry, the registry entry may well record the article.   

The students also had an example of the same trial being written up in two articles.  Ben Goldacre's work records that many trials never make it into any sort of article.   Both these things are, it seems to me, rather naughty.  Perhaps it is more difficult to find the associated article than I think.  

I had a quick look in Web of Science Core Collection for conference abstracts from 2007 about antimicrobial resistance.   I chose three (which I know is a tiny and unrepresentative sample).   I then looked in PubMed (a proper look would have course been in WoSCC as well, which I shall endeavour to do before too long) for the first two authors, or authors and a title word if that found too many.   One found something very similar which on closer inspection was from 2005!   The other two did not seem to appear.   Perhaps they were never written up in full articles.  If not, why not?   Some subject areas in non biomedical science assign great importance to abstracts and maybe things get no further.  Is antimicrobial resistance an area like that within biomedical science and was there no need to write it up?    If so, then this means it really is important to search for abstracts and not just articles.  Or were the studies not quite up to it?

Of course, a fuller and more systematic investigation is needed!  There is a bit more elsewhere in another post.

If you can't find an article that has the detail you need, and there is not enough detail in the conference abstract, what then?   Is this one of those situations where you contact the authors for more information, as the textbooks about systematic reviewing say?    Interestingly, the students had tried this (not sure if it was for data from a study that had only appeared as an abstract, or not) and had received the reply that the author(s) was/were too busy to help.   Perhaps if you are not a student, but a postdoctoral researcher, you have more luck.

Another quick look in WoSCC turned up two papers by the same team, Yunal Dundar and colleagues from the University of Liverpool, from 2006, one about searching for abstracts and the other comparing data from abstracts and subsequent full articles .  (The links are via DOI and you will need a subscription to read them). 

Both are in the context of health technology assessments and I need to read both before I meet people in the same situation as my MSc students!

Medline, PubMed, or both?

A question raised in a recent teaching session with MSc students undertaking systematic reviews.  Should I search Medline and PubMed?

Actually, people don't usually ask the question, but instead tell you that they have searched both, or that someone has recommended they do so, or that they have read a review where the reviewers have searched both.

I always say I am not convinced you need to search both.  But am I right?

I know PubMed includes things that Medline does not:

1. Medline indexes some journals (like Science and Nature) selectively, and will not include non biomedical / life science articles.  PubMed, however, does include them.  

I don't think this would matter to a systematic reviewer in healthcare.  

2. Issues of a newly accepted journal that predate the date when it was accepted for indexing in Medline will be included in PubMed.  This might matter, but I have never been convinced that it matters enough to search both databases.

3.  References sent from the publisher, but not yet indexed for Medline.  This of course, might matter, but we have access to what used to be called "PreMedline", and what is now called "Medline In Process", which is these references.  So, if your search includes freetext, you ought to find it without using PubMed.

But what about the fact that PubMed and Medline have interfaces that work in different ways?   Is it possible to find material in one that you did not find in the other, just because of the way the search works?   Surely if you search PubMed using the MeSH browser, and select the same MeSH terms as you did in Medline, you would find the same material.   I have to say, that when I have been doing detailed searching for research projects (and this is not something I do a lot of, I do spend much more time teaching others to do it), I have never used PubMed, and always used Ovid Medline.

Here is a search that I did in Ovid Medline and PubMed.  The syntax is Ovid's, and I used the MeSH browser and Advanced Search in PubMed to replicate it.  The unexploded MeSH heading was not used in the results set, but I just wanted to check that PubMed had exploded the heading in set 1.

Searches run 2230 BST 18.4.15
		Ovid Medline	PubMed
1	Exp Diabetes Mellitus, Type 2/	90273	88785
2	Diabetes Mellitus, Type 2/	90098	88612
3	Type 2 diabetes.ti,ab.	72369	72674
4	Insulin glargine.ti,ab.	1221	1234
	(1 or 3) and 4	767	775
	(1 or 3)	116064	115948

 

1542 references in total.  Using RefWorks to remove close duplicates leaves 795. Because author names are not formatted the same, and journal titles in one database are abbreviated and in the other not, finding exact duplicates did not identify many.  However, the first pair of references (that is, a pair that RefWorks thought were duplicates) were not, which makes me wonder if I deleted items that were not duplicates.  

So, there are a few references in PubMed that are not in Ovid Medline, and vice versa.  I need to look in more detail at this to see why.

Removing large numbers of duplicates in RefWorks (I tell my students they can use RefWorks to manage duplicates when working on a systematic review) was interesting!  I changed the default number of references per page to 500, or I would have had 50+ pages to deduplicate.  I also looked at using EndNote Online to do this but it only seems to remove duplicates across the whole library, not a single folder, and as I had been using it for a real project, I decided not to do that.

 

Searching PROSPERO

Prospero is an international database of systematic reviews.  Reviewers can register review protocols in it, and link to publications once they appear.   Prospero is hosted by the Centre for Reviews and Dissemination at the University of York.  There is information about registration and the data that is required here.

Searching Prospero before you start a review could stop you duplicating work that is already in progress.  

As part of an enquiry at work, I was searching Prospero for reviews to include in a review of reviews (an "umbrella review").   Searching Prospero would locate reviews in progress, and would identify published reviews (alongside those identified using Medline and other databases). 

I have not found any search help specific to Prospero.  I am fairly sure the search help at http://www.crd.york.ac.uk/crdweb/GuideToSearching.asp applies to the other CRD databases.  

The References and Resources section contains a lot of useful information, including reviews of progress, and there are some interesting articles about Prospero under "Registering a systematic review", but the main emphasis of all the material seems to be about registration, not searching.  So, I thought it might be useful to "publish" my notes about searching Prospero.

Complex search strategies are difficult.  You can't see a search history, and so can only submit one query at a time.  If you use one search box, you can't use Boolean operators, although you can use * to truncate.  You can't enter phrases in inverted commas, but I wonder if phrase searching is the default.  Using more than one search box does give you a way to search using Boolean in one field (change the drop down box at the top) or across fields, but you have to have just AND or OR.  

You can't export results, only view them, so reviewing results for items to include in your research would need to be done on screen, and I am not sure how you would then put those items into reference management software for retention.

The Participants/Population field records details about the group(s) being studied in the review.  My enquiry involved finding material relating to older people.  Searching for aged did not work, as this found phrases like "patients aged 10 or above".  Older worked better, although that still found phrases like "30 years and older".  Elderly and geriatric worked as well.  There is no standardisation in this field, and no equivalent of Medline's age limits.

You can display all published records, and doing so shows that at the moment there are 5739 records, so the dataset is a lot smaller than something like Medline, and this combined with the fact that searching Prospero would be to augment more systematic searches done elsewhere, means that broader searches are still practical.