Trip Database Blog

Liberating the literature

New search, new server

We’ve just upgraded to a new and more powerful server system – lots of extra memory and speed.

As part of the new system we’ve also improved two other things:

  • Upgraded the version of our search engine technology to the latest version. This has improved many things including my own bugbear – duplicate terms in the title boosting the result. In the old version if a document had the search term(s) mentioned twice in the title it would get a large boost. It didn’t happen often but it was annoying! But it appears that’s been fixed.
  • Improved our indexing of PubMed.  Previously we’d been grabbing all the content from PubMed via eUtilities and we only realised recently that this included all the additional meta-data including things like related articles and cited articles data.  This created false positive results.  Our new indexing will now only grab content from the abstract.  This means results counts are both reduced and more accurate (less noise).

We’ve tested it extensively but if you spot anything strange please let us know.


NOTE: I can see the new server but it might take 24 hours for the internet’s address book to point at the new version. If the ‘Evidence Maps’ link at the top of the page is highlighted yellow, you’re still pointing at the old version!

Include all of PubMed in Trip?

Currently Trip includes a sub-set of PubMed. This includes all RCTs and systematic reviews and the content of around 500 ‘core’ journals. In total this represents around 10-15% of all the PubMed content. Putting it another way we have around 3-4 million PubMed records out of around 28 million.

But, to be clear not all records are equal.  Some will be to do with veterinary medicine (e.g. Serum amyloid A and plasma fibrinogen concentrations in horses following emergency exploratory celiotomy)  and others might be very specialised technical articles (e.g. Phase-matched virtual coil reconstruction for highly accelerated diffusion echo-planar imaging). So, we’ll be potentially including extra noise. We can mitigate against that – to a point.

In summary, searching all of PubMed is potentially very useful but there may be downsides. So with that in mind what do you think?

What’s happening?

Currently we are migrating to a new super-duper-shiney new server. This is a fairly large task – hence us being a bit quite recently. We’re hoping to have finished that in the next couple of weeks.

At the same time we’re working on the Trip community feature/service. This is experimental, so no timeline except for my usual impatience to get it delivered!

I’ve just uploaded the latest monthly – manual – batch of content. Most content to Trip gets added automatically but every month (as I have done each month for the last twenty years) I go to a number of sites and manually grab any new content.  At times this manual approach can feel like a bit of a chore, but generally I marvel at the breadth of new content being delivered.  This month I’ve found myself refreshing/boosting our links to Patient Decision Aids.

Finally, our annual accounts have been finalised and our subscription income has reached a key milestone – Trip feels more secure than ever. But I have no plans to allow this to make me feel comfortable. The moment we stop trying to improve is the moment we start to fail. Onwards and upwards.

RCTs in Trip – much improved thanks to RobotReviewer

RCTs are a crucial element of EBM and we’ve had a ‘filter’ for them for years. This allows a user to search and click the ‘Controlled trials’ filter and the results will only show clinical trials.  It’s one of our most used filters.

In the earliest iteration we used our own search filter to identify trials in PubMed and added them to Trip. This was ok, but lots of false negatives (missed trials) and false positives (identified papers as trials which weren’t).  A few years ago we started using the RCTs identified by the wonderful RobotReviewer team.  This used machine learning and made the results dramatically better.

Over the weekend we’ve used the latest version of the RobotReviewer code to make the identification of trials the most accurate yet!  Previously we had identified 479,197 trials but now it’s up to 532,479.  We had thousands of trials labelled as trials which no longer are and thousands we had previously missed. There were also a load of new trials that weren’t in Trip.

This is a brilliant piece of work and we’re indebted to RobotReviewer – thank you.

Summary: We have an easy to search collection of over half a million clinical trials!

Loads of extra full-text and a new results display to celebrate!

Full-text access is really desirable and it’s something we’re always trying to improve on. Up until now Pro users have been able to access full-text only if it exists on PubMed Central. This meant that 38% of all journal articles linked out to full-text.  We’re now working with an external partner who has some very clever technology and now – in some areas – the link out to full-text is over 80%.

We’ve redesigned the results to help people see these full-text options and so here they are:

Old results

New results

So, what are the changes:

  • If you’re a Pro user the link out defaults to the full-text. In the image above this is clearly marked with (full text) after the document title. There is also the option to link to the PubMed abstract.
  • If you’re a Pro user and we haven’t got a full-text link out we signify this with (PubMed) after the document title.
  • We’ve improved the spacing of the results.
  • The links to tweeting, starring etc have been made more obvious and aesthetically better.
  • If you’re a free user you can clearly see if a full-text is available.


The redesigned results is a great improvement while the massive increase in link outs to full-text is phenomenal.

Clinical Q&As – keeping them relevant

Trip started as a result of my work in running a responsive clinical question answering service for the National Health Service – many years ago.  I have had the pleasure of running a number of such services (alas, not one for a few years) and have kept these questions and answers within Trip.  Unfortunately, these have been neglected: they are often two years out of date and they look terrible:

Click here to see it on the web in all its glory!

This is a valuable resource, knowledge needs from real clinicians not the typical top-down approach to knowledge.  We’ve got well over 5,000 Q&As

As part of this years developments in Trip I’m keen to see if we can help save/update these Q&As.  So, how might we do this? Here are some thoughts:

  • Improve the design – obviously!  But as well as how it looks it might also be functional elements. Currently it’s the Q and then the A – with no structure.  Could we add things like ‘bottom line’?
  • Produce a priority list of Q&As to be updated.  This could be based on a number of elements e.g. (1) How popular is the Q&A (2) How old is it (3) We’re hoping to produce a system to find new, related, articles – so if we find some recent evidence that gets a higher weighting than those with none.
  • Open it up to the community – wiki-style – to edit and update. We’re going to be doing lots of work around the Trip community in the near future and this will lend itself nicely to a wiki-approach.

I’m really excited by this and if it works well we could, one day, allow users to pose questions and get the crowd to answer.


What’s important for our Rapid Review system? RESULTS

We had a lot of interest in our post What’s important for our Rapid Review system? with – at the time of writing – 297 votes.  The breakdown being:

I was surprised by a few results. I wasn’t expecting ‘teams’ to appear so highly and I was expecting ‘citations’ to be higher up.  Further evidence of the benefits of engagement. These will now help us refine our offering.

The general idea being to develop a ‘minimal viable product’ (MVP) to allow testers to try it out and give further insight. Once that’s done we finalise things and release.  I’d like to think we an get the MVP ready by the end of February (March at the latest) and the finished product by the middle of the year…

Blog stats

Not sure why but I thought I’d look at the stats for this blog over the last few years and I’m quite pleased as (a) larger numbers than I was expecting (b) it’s increasing year-on-year!

2016 – 8,172 visitors and 14,316 page views

2017 – 11,577 and 20,641 (a 44% increase in page views from 2016)

2018 – 19,598 and 33,710 (a 63% increase in page views from 2017 and 135% from 2016)


What’s important for our Rapid Review system?

This is the year we build our rapid review system.  In short it’s a step-by-step ‘wizard’ that supports users (either via automation or community support) to rapidly generate evidence reviews. The steps are as follows:

  • Question setting/clarification, including the extraction of search terms.
  • Search and document selection, including various techniques to unearth articles that may have been missed.
  • Data extraction to produce an evidence table.
  • Narrative review including a conclusion or clinical bottom line (to be decided).
  • Review finalisation. All the above parts pulled together to produce an easy to read rapid review.
  • Publication and peer review.

To help us understand where to particularly focus can you select which elements below are really important – please select the 5 most important aspects. If you think we’ve missed something (or have any general comments) please let us know via the ‘Other’ box:

Thank you

Blog at

Up ↑