Making PubMed “Easy”?

Jon Brassey writes:

I may have missed something, but none of these alternate interfaces allow easy searching of PubMed. Some are wonders of programming, some allow some very neat tricks but none make searching of PubMed easy.

That’s a fair criticism, I suppose. I think that although PubMed has come a very long way in developing tools that make searching the NLM’s databases easier for medical librarians, clinicians and consumers, it still takes some knowledge and skill to perform a really useful search of the primary literature.

Jon continues:

I suppose my biggest issue with PubMed is that doing a search of statins returns 18,491 results. Unpicking that a bit:

* Most research shows search engine users finish looking after 3 pages of results.
* From our own experience with TRIP we also know that most users only use single search terms (e.g. asthma, hypertension).

So what I’m saying is that statins is a realistic search term and that suggests that 18,431 (18491-60) results are superfluous.

Therefore, the two challenges to me are:

* Return fewer results in the first place
* Allow users to easily qualify their searches.

Let’s bring Jon’s challenges to Healia’s PubMed/MEDLINE Search. It isn’t my favorite, but if Jon uses it to search for statins, he’ll see that, at the time of this writing, only 7,731 results are returnedI’m guessing that the reason why PubMed returns 18,431 results and Healia’s PubMed/MEDLINE search returns only 7,731 is that Healia’s search is only looking for the string statins.

PubMed, on the other hand, translates statins into “hydroxymethylglutaryl-coa reductase inhibitors”[MeSH Terms] OR “hydroxymethylglutaryl-coa reductase inhibitors”[Pharmacological Action] OR statins[Text Word].

After all, if we search PubMed for “statins” as a string, we get only 7,990 results. and that there are a number of tools for “qualifying” the search right there on the search results page.

You can adjust for date:

You can filter for review articles or for English language articles only:

You can filter by patient demographics:

Healia even recognizes that we’re searching about a class of drugs and gives us tabs so we can filter by Dosage, Usage, or Side Effects:

So let us assume that Jon’s hypothetical user wants to see English-language review articles about the dosage of statins from only the last 5 years. That’s only twelve results. I think this passes Jon’s test.

With that out of the way, I need to add:

Criticizing PubMed for returning too many search results for a query as inadequate as “statins” seems unreasonable to me. A major part of what makes PubMed an amazing tool is its complexity. I believe and hope that new tools will continue to be developed that make the data useful to various kinds of users in various new ways, but I don’t expect that getting exactly what one wants from the primary literature will ever truly be “easy.”

8 thoughts on “Making PubMed “Easy”?

  1. Thanks for commenting on my post, a few points:

    1) Healia is a pretty good interface but aren’t you concerned that it misses out over half the potential results as it doesn’t auto-map to MeSH?

    2) I find the MeSH terms for demographics, especially age, are pretty poor.

    3) Take your example of using the tabs, try using side-effects . In Helia you get 31 reviews, in English, from the last 5 years

    I’ve repeated your search for side effects of statins. Instead of using Helia’s method, I’ll use my own.

    #1 = statins
    #2 = myalgia OR myositis OR myopathy OR rhabdomyolysis
    #3 = #1 AND #2
    Note: #2 is made up of the major side-effects, I used those for demonstration purposes. It didn’t include things such as altered LFTs etc.

    If you limit that to the last 5 years and English you get 112 reviews. Are those ‘missing’ 81 reviews important?

    In a way I should be pleased; less results is what I requested!

    While searching for ‘statins’ is inadequate, it’s also realistic. In the analysis of TRIP the majority (by a long way) were single search terms. I wouldn’t have thought the users of Medline are likely to be significantly more ‘savvy’ than TRIP users.

    Don’t get me wrong I think that PubMed is wonderful, I use it every day. But the complexity you praise is also its downfall.

    I agree that, in time, new interfaces will come out to enhance and improve access to the literature contained in PubMed. An ‘easy’ way (other than asking a librarian/info specialist) may never come to fruition, but it’d be a shame not to aspire to that.

  2. How does Healia work? Does it use the ever-existng PubMed filters? I guess so in the case of publication type and population.
    How does it do the dosage/adverse effects search? Using MeSH-subheadings “/adverse effects…”?

  3. Jon, I think you’ve pretty much made my point for me. It is impossible to produce specific, relevant results without formulating a query full of specifics. Medical librarians and other expert searchers aren’t going to use Healia’s PubMed/Medline search because it is inferior for their purposes.

    I believe that with software, power and ease-of-use frequently have an inverse relationship. PubMed is powerful, but has a learning curve. Healia PubMed/Medline is much easier to use, but nowhere near as powerful.

    1) Healia is a pretty good interface but aren’t you concerned that it misses out over half the potential results as it doesn’t auto-map to MeSH?

    Yes, that’s why I specifically mentioned it. See the footnote.

    While searching for ‘statins’ is inadequate, it’s also realistic. In the analysis of TRIP the majority (by a long way) were single search terms. I wouldn’t have thought the users of Medline are likely to be significantly more ‘savvy’ than TRIP users.

    Lastly, I think it is a mistake to attempt to estimate how people use PubMed based on how they use TRIP. They have different (although overlapping) user bases, they contain different sorts of information, and TRIP doesn’t use a controlled vocabulary. But even if it is, as you write, “realistic”, a search for “statins” in PubMed is mapped to MeSH terms and the user will review the first couple of results pages to see all the most recent articles that match the mapped query. If a user is so unsophisticated as to search just for statins, that search is likely to suffice. Someone who knows that they need more specific information will likely ask for it by querying for something like statins adverse effects.

  4. I disagree to a significant extent with the last part of the last paragraph. However, evidence is pretty scant, so we’ll have to agree to disagree. I seem to remember some research showing that non-expert searchers located around 30% of relevant/pertient documents in PubMed. Information experts tended to find nearer 70%.

    One of the big ideas in search is intentions. So when someone searches for something can we know what they’re interested in? To an extent (and I’m not suggesting we’ve done it too well) we’ve had a go with our specialist search engines. For instance if you’re an oncologist and search for statins you’re likely to want different information than, say, a cardiologist. So an oncologist could use this link and get articles pertient to their profession.

  5. Hi Jon-

    Disagreement is *good*! 🙂

    Could you share more about what your ideal “easy” PubMed search would look like? Would it be sort of a search wizard that would take the query “statins”, offer synonyms and MeSH mappings, then prompt the user for more specifics? (Like asking if the user wants information about dosage, adverse effects, efficacy, etc.) Would it ask the user what condition or discipline their query ins in relation to?

    How do you make a PubMed/Medline interface “easier” without sacrificing power and specificity?

  6. Thanks for the feedback on Healia’s Pubmed/Medline search tool. Our initial focus was to make Pubmed searching easier given that our focus is on the consumer. However, we recognize that it can be improved to meet the needs of higher end users and professionals. We are hoping to release enhancements in the near future. The audience/demographic filters are based on our content assessment algorithms.