Translating filters between different interfaces: not as easy as you might think

Session: 

Oral session: Searching and information retrieval (2)

Date: 

Monday 17 September 2018 - 11:50 to 12:10

Location: 

All authors in correct order:

Glanville J1, Foxlee R2, Wisniewski S2, Noel-Storr A3, Edwards M1, Dooley G4
1 York Health Economics Consortium, United Kingdom
2 Cochrane Editorial Unit, United Kingdom
3 Cochrane Dementia and Cognitive Improvement Group, United Kingdom
4 Metaxis, United Kingdom
Presenting author and contact person

Presenting author:

Julie Glanville

Contact person:

Abstract text
Background:
Information specialists frequently translate search filters from one interface to another. It is unlikely that the impact and performance of such translations is investigated in depth because of the time needed to conduct a thorough assessment and information specialists may well assume that databases are largely the same across different publishers and that therefore interface adjustments are the only issue for successful translation.

Objectives:
To investigate whether we would find the same set of records after translating the Cochrane Embase search filter from one interface (Ovid) to another (Embase.com).

Methods:
We compared the records produced by searching with the original Cochrane Embase filter (for Ovid) with the records produced by searching using a carefully translated filter (for Embase.com). We compared the results of the searches line by line to see if each line had found the same records in each interface.

Results:
Even thought the majority of the same records were retrieved from the two filters in the different interfaces, we also identified a range of unexpected differences. Some records had differences in publication years and subject headings between the two versions of the database, and some records were in one database but not in the other. We also encountered interface differences relating to truncation, proximity operators, and use of the original title or original abstract field. We contacted both Ovid and Embase.com to highlight unexpected technical issues.

Conclusions:
It is challenging and time-consuming to test search performance in different interfaces, in what might be anticipated to be the same database, but it does seem to be essential to test, so that appropriate adjustments can be made. The translation can be impacted by version differences and also by interface differences in unexpected ways. Translating a filter may not be straightforward and any search filters intended for regular use in new interfaces should be carefully tested.

Patient or healthcare consumer involvement:
No patients or healthcare consumers were involved. In this situation the researchers were the consumers of the database services that were being evaluated.

Relevance to patients and consumers: 

This research reports on a search translation intended to improve the efficiency of trial identification for CENTRAL. The more efficient identification of trials should make review production more efficient and aid the identification of as much relevant data as possible so that reviews are as complete as possible. These efforts help to make evidence available more quickly and efficiently to patients and consumers. Patients and consumers did not participate in this research or this statement.