Did you know that, on average, 7-25% of citations found in an initial literature search are duplicates? We've even been on systematic reviews where up to 40% of citations are duplicates! This isn't a mistake; it's a by-product of conducting a comprehensive systematic search. You see, when you search multiple databases for relevant literature, you're bound to discover instances where:

  • the same citation appears in multiple databases but is indexed differently in each one, or
  • the same study was published in multiple journals or in multiple formats.

We call the first type of duplicate a database duplicate and the second type of duplicate a study duplicate.

Identifying and addressing duplicate citations is a crucial step in maintaining the quality of your review. If you fail to properly address duplicate citations, you risk biased results because you have, unwittingly, inflated the “apparent number of studies available for analysis” (Hammer et al, 2023).

As a conscientious evidence synthesist, it's your duty to identify duplicates and clearly document your approach to handling them. But that's no small feat, especially when your initial search yields thousands of citations. That's why our team is happy to announce our new, automated citation deduplication feature! Read on to learn more.

How does deduplication work in MetaReviewer?

Our citation deduplication feature is designed to help you identify and address database duplicates. We developed an algorithm that identifies potential database duplicates by applying a series of matching rules, starting with the most reliable methods and progressing to more flexible ones.

First, the algorithm generates duplicate sets for citations that have an exact match on unique identifiers such as the DOI. Such identifiers provide the highest confidence that two records represent the same publication.

Next, the algorithm compare citations based on a normalized version of the title and publication year. Citations with an identical (normalized) title and year are pulled into a duplicate set.

Finally, for citations that cannot be exactly matched, the algorithm performs a fuzzy title comparison. To keep the process efficient, the algorithm only compares citations that share the same publication year, the same first author's last name, or a similar title prefix.

Throughout this process, we use a union-find data structure to combine matched citations into duplicate groups. The union-find structure ensures that if Citation A matches Citation B, and Citation B matches Citation C, all three citations are grouped together into a single duplicate set, even if A and C were never directly compared.

The algorithm runs any time you import a citation file--with your import options now expanded to include RIS and BIB files--into MetaReviewer. You can learn more about importing citations in our User Guide.

Keeping humans in the loop

When it comes to automation, our goal is to make reviews easier without sacrificing data quality. And, while algorithms can be quite efficient, they are imperfect and make mistakes! So, instead of automatically flagging and removing all potential duplicates, MetaReviewer elevates potential duplicate sets for project teams to review and address.

After importing your citations, if the algorithm has identified potential duplicates, you'll see a table that contains a citation of record (i.e., the one you'll keep) and its potential duplicates:

You can scroll horizontally and vertically to compare the citation of record to each potential duplicate across key fields like title, publication year, authors, journal, and abstract.

If the algorithm has correctly identified a duplicate record, you can select fields from the duplicate record that have information you want to transfer the citation of record. You can then either delete the duplicate or mark it as a duplicate. If you mark the citation as a duplicate, it will remain in your project but will be flagged. This might be preferable for hoarders or those who want to double-check counts for their PRISMA chart later on down the line.

If the algorithm elevates a false duplicate, you can simply select the "not a dupe" button and move onto your next duplicate set.

What about study duplicates?

Earlier, we mentioned that you are likely to find cases where the same study was published in multiple journals or in multiple formats. Or, you might find that the authors published a report for each year of findings for the study. Depending on the similarity of the citations, our algorithm might elevate these for your review.

We advise against deleting these duplicates since these are citations that should be linked to the same study and reviewed during full-text screening and coding. Instead, we recommend either 1) flagging as "not dupes" and documenting their linkage on their individual study page, or 2) marking them as duplicates, so you don't forget to attend to them throughout your review.

In summary, seeing double due to duplicate citations is common in comprehensive systematic reviews. But thanks to MetaReviewer's new citation management functionality, managing duplicates doesn't have to be a manual, time-intensive task. By combining automated detection with human review of duplicates, MetaReviewer makes citation management faster, easier, and more transparent, helping your team maintain an accurate evidence base throughout the review process.