Linking related publications#142

Sometimes studies report results across multiple publications (e.g., one publication for primary outcomes and another for quality of life outcomes). The ability to link such publications is very helpful in organizing the studies. Other SR software such as Distiller and Covidence have implemented this type of feature to varying degrees, but one aspect I have not seen yet is using automation to help identify related publications. I’m not sure the best way to implement it, but I imagine it would be something similar to duplicate detection, except it would be looking for substantial overlap in authors/titles rather than exact matches.

Funny timing- I was just chewing on this earlier today in updating some of our tools to PRISMA 2020 (https://nested-knowledge.nolt.io/4). It’s simpler to assume (impose) that reports and studies are in 1-1 correspondence in a review (our current functionality), but it’s not realistic, and we’ll be addressing this by building some method of marking multiple reports of a single study.

As to identifying related publications, NCT #s indexed on PubMed (or vice versa) could be a start; I’m certain these have spotty indexing, but it’s at least a start. Overlap in investigators / titles could also be interesting - it’s predicated on the putative matching report being present in your search. Theoretically, if the search strategy is good and the 2+ reports are relevant to the review, that should happen most of the time!

Similar request found in https://nested-knowledge.nolt.io/308

+1 to this request! Would be very helpful!

@Dnikitin we’ve finished the design for this feature, so it should be rolling out fairly soon!

Implemented with release 1.44.0. Thanks for this request! RE automations, we started from the easiest place - importing linked publications from ClinicalTrials.gov & auto-marking them as related reports (it also covers records already in your nest). We have been chatting with the ECRI team about doing broader searching for related reports, but it’s a large volume of data & still an unsolved problem in my book (Liu 2022 seems to have a nearly viable system in terms of recall@1, although it’s unclear how well it would generalize to real datasets).

Some notes on the feature:

• Access via the “Related Reports” tab
• Search for records using author year or title format
• You can select one “primary” report per group of related reports
  • Currently, this is only metadata for AutoLit. If you’re only extracting one dataset per study (not per report), you may find marking the primary important for keeping track of which report to gather under.
  • We have plans to represent the primary/secondary status in Synthesis, in the future.
• Records marked as related reports of the same study will be reflected in your PRISMA diagram. Specifically, in the inclusions box that differentiates reports & studies.

In action:

Wow! This is going to be a huge time saver when it comes to organizing our included references! Am I understanding correctly that it is currently linking by NCT and requires the abstract to spell out the exact id (ex. NCT04311424) for this to work?

Does this work only on unscreened references? I’ve gone through 100+ refs in both Study Inspector of final inclusion refs and those that were screened but excluded and I’ve had no hits so far in Related Reports tab.

Mining on NCT ID requires that the record have its NCT ID (a bibliographic field in NK) populated. This field will be populated when you either:

1. (automatic) Import a ClinicalTrials.gov record
2. (automatic) Import a PubMed record that has a linked NCT ID
3. (manual) enter it under bibliographic data

Furthermore, we don’t automatically mine on NCT ID. You, the user, must click the “Bibliomine NCT” button.

It sounds like:

1. You’d like the NCT ID to be automatically extracted from abstracts of all imported records (e.g. including RIS imports), when present
2. You’d like the option to automatically bibliomine on NCT ID for a bunch of records at once.

I really like (1), and it’s technically quite feasible. If I’m hitting the right notes, could you file new nolts?

Oh, and to be clear - you can (manually) mark related reports completely independent of NCT IDs. NCT IDs just provide mechanism that makes marking related reports easy :)