Building up a disease-specific trials register in hemato-oncology: experiences from a project of the Cochrane Haematological Malignancies Group (CHMG)

Tags: Poster
Trelle S, Kober T, Higgins G, Engert A

Objectives: To describe our experience with building up a disease specific trials register for Hodgkins disease, and to evaluate the usefulness of different databases and the appropriateness of a broad and sensitive search strategy.

Methods: We searched MEDLINE, EMBASE, CENTRAL, the Institute of Scientific Information Web of Science (ISI WoS), PsychIndex, AMed, CancerLit for randomized controlled trials (RCT) in Hodgkins disease (HD). Previously described filters for identifying RCTs were used for MEDLINE and EMBASE. For the ISI WoS we used the term random* and no methodological filter was used for the other databases. These methodological filters were combined with a sensitive disease specific strategy (among others the text word hodgkin* was used). References were downloaded to Reference Manager 9.0 (ISI ResearchSoft) using the duplicates function. Retrieved references were checked and coded by one of two authors (ST, TK) for their potential eligibility. Interrater agreement was assessed regularly using randomly selected references from the raw database which were evaluated by both. Disagreements were discussed in detail. Full text articles were obtained for all potential reports and assessed (ST, TK). All reports found to be RCTs after full text assessment were eligible for inclusion in the database. Results were compared with the CHMG trials register and a search using the Clinical Queries function of PubMed (specific search strategy for therapy). Database searches were performed in April 2003. Analysis is restricted to reports published until 2002.

Preliminary results: We retrieved 20371 references (MEDLINE: 3549; EMBASE: 6364; CENTRAL: 1431; ISI WoS: 4767; PsychIndex: 290; AMed: 94; CancerLit: 3876). Reference Manager identified 6998 duplicates leaving 13373 references for assessment. Of these, 300-400 were found to be RCTs in Hodgkins disease. This figure is comparable to the number of trials in Hodgkins disease already listed in the CHMG trials register. The PubMed search retrieved 366 references. A detailed comparison will be presented as well as a detailed evaluation of the search strategy and the usefulness of the different databases used. Interrater agreement for the initial selection of potential reports was good with a median kappa of 0.61 but highly variable over time (range: 0.34 0.95).

Conclusions: Building up a disease specific trials register in hemato-oncology using a sensitive search strategy is time consuming and sometimes cumbersome for the reviewers. The CENTRAL database of the Cochrane Collaboration seems to be fairly comprehensive today. Applying a less sensitive search strategy may yield a comparable amount of retrieved RCTs and would be by far less tedious . If the few trials not identified by using the less sensitive searches would introduce bias in meta-analyses or clinical decision making remains unclear.