'Clustering' documents automatically to support scoping reviews of research: a case study.

BACKGROUND

Scoping reviews of research help determine the feasibility and the resource requirements of conducting a systematic review, and the potential to generate a description of the literature quickly is attractive.

AIMS

To test the utility and applicability of an automated clustering tool to describe and group research studies to improve the efficiency of scoping reviews.

METHODS

A retrospective study of two completed scoping reviews was conducted. This compared the groups and descriptive categories obtained by automatically clustering titles and abstracts with those that had originally been derived using traditional researcher-driven techniques.

RESULTS

The clustering tool rapidly categorised research into themes, which were useful in some instances, but not in others. This provided a dynamic means to view each dataset. Interpretation was challenging where there were potentially multiple meanings of terms. Where relevant clusters were unambiguous, there was a high precision of relevant studies, although recall varied widely.

CONCLUSIONS

Policy-relevant scoping reviews are often undertaken rapidly, and this could potentially be enhanced by automation depending on the nature of the dataset and information sought. However, it is not a replacement for researcher-developed classification. The possibilities of further applications and potential for use in other types of review are discussed.