All the information retrieved from the search strategy was saved and imported into OmniViz™ for analysis. Thus, a new database with nine components (publication type, author, title, affiliation, reference, abstract, medical subject headings or MeSH terms, PubMed unique identifier or PMID, and publication date) was created. A visual representation of information, or Galaxy View, was generated using the default text analysis algorithms. To generate a more coherent view of the database, the cluster number was limited to 20 clusters, and both title and abstract words were selected to create the clusters.

The Galaxy View is a proximity map, such that closely related records and clusters are placed near each other, while thematically distinct clusters are located far apart. The record clustering is determined by the numeric vector calculated for each record, based on word occurrence, distribution, and associations located in a selected component (our analysis parameter was both title and abstract words).

In the Galaxy View, each dot represents a record and each record represents a publication retrieved from MEDLINE. Within a cluster, each record can be identified and selected to view, in the information panel, a detailed information about the cluster label and the record. The cluster label is comprised by top three discriminating major terms. Major terms are the words or phrases located in the selected component that are most relevant to discriminating one record from another and they are the principal parameters used in clustering the records. Nearby clusters often share major terms, indicating that they have some themes in common. Detailed view of the selected clusters or records (e.g., publication type, author, title, abstract) can also be seen.

In the next phase, the clusters were queried to identify the major term, and determine the identity and number of clinical trials and systematic reviews. The search strategies used to identify the clinical trials and systematic reviews were based on the work of Haynes RB et al. [17] and Montori VM et al. [23]. The number of identified records and clusters was determined and tabulated using Excel for windows (Microsoft, USA) spreadsheets. A final analysis was done in a descriptive fashion.