Saturday, January 10, 2015

Search filters

I have been invited to do the literature searching for a research project looking at care of older people in hospital.  The first part of the project is a literature review.  It's fascinating, and very instructive - I have spent a lot of time working to support others who are doing this, and it is quite eye opening to do it for myself!

The literature review is of systematic reviews (SRs) and meta analyses.  I first of all had to devise the strategy, and that was very interesting.  And then decide how to locate the SRs.  And so is that ...

I could:

1.  Use the standard Medline search limits.  There never used to be a limit for systematic reviews, and meta analysis was the nearest fit, although it is not the same thing.  Publication type limits in Medline are not without issue.  There may be references indexed as meta analyses which are not (but which say they are), and there may be things that are meta analyses, but which are not so indexed.  Now there is a SR limit, but in PubMed anyway it is a filter, not a straightforward limit.  Certainly a search in Medline for systematic finds zero.

2.  This leads us to the second possibility.  As a way to address the shortcomings of the standard limits, and as a way to get reproducible searches, there are search filters, not part of Medline, strategies which can be bolted on to your subject search.  But there are several SR filters.  Which one to choose?   Some have high sensitivity ("the number of relevant reports identified divided by the total number of relevant reports in existence") and others have high precision ("the number of relevant reports identified divided by the total number of reports identified" - this, and the definition of sensitivity are taken from the Cochrane Handbook). 

 The ISSG Search Filters Resource lists many filters, and evaluates many of them.  I wanted it to recommend one, but of course because it depends what you are aiming for in terms of sensitivity and precision, no one can say "this one is the best".

3.  The third option is not to use a filter or limit at all, but include or exclude things yourself.   If you do this, you know exactly what you have kept.  But, there is a practical consideration, which is that if you get 2000 results, do you have time to scan them all?  This is interesting, as I have certainly said to students I have been helping that they may end up with this sort of number of results!

So, what to do.  I think I will be running all the filters (and the "limit") separately, and merging the results.  I suspect, but don't know, that these separate result sets will overlap but that none will completely include any of the others.  We shall see.  I then plan to remove this merged set from the unlimited/unfiltered results, and then we can sample the remainder and see if any of the sample are an SR or a meta analysis.

I put out a request for experience and expertise on two Jiscmail lists, and the results (I shall summarise for the lists) have informed the above.

No comments: