How useful is searching Biological Abstracts (BIOSIS) for reports of randomised trials? A comparison with MEDLINE and EMBASE in renal disease

Ruth Mitchell [1]
Steve McDonald [2]
Jonathan Craig [1]
1. Cochrane Renal Group, Centre for Kidney Research, Childrenís Hospital at Westmead, Sydney, Australia
2. Australasian Cochrane Centre, Monash University, Melbourne, Victoria, Australia

Background: Biological Abstracts (BIOSIS) is a major life sciences database, containing over 5 million records (1980-present) from 5000 journals, meeting proceedings, books and other publications. Although there is a sizeable overlap with MEDLINE, many Cochrane groups recommend reviewers search BIOSIS as an additional source of trials.

Objectives:To evaluate the usefulness of BIOSIS as a source of randomised trials of interventions for renal disease not already indexed in either MEDLINE or EMBASE.

Methods: We devised a sensitive free-text search strategy for identifying randomised or possible randomised trials in renal disease which could be run against both BIOSIS and MEDLINE. Using Ovidís multiple database search facility we ran this search across all fields in both databases for the years 1980 onwards (the version of EMBASE available could not be included in the multifile search). We removed duplicates online and initially downloaded the unique BIOSIS records for two sample periods (1980-81 and 1998-99). We read the titles and abstracts of these records to identify all randomised or possible randomised trials irrespective of subject area. For each trial report identified, we double-checked to see if the record was in MEDLINE and then cross-checked against EMBASE and The Cochrane Controlled Trials Register (CCTR).

Result:The search strategy yielded 37,990 records (19,762 unique MEDLINE; 3939 unique BIOSIS; 14,289 duplicates). We downloaded unique BIOSIS records for the two sample periods (303 for 1980-81; 431 for 1998-99). We identified 150 randomised trials from scanning these records, however, 126 (71%) were in fact in MEDLINE but had not been detected by Ovidís deduplication system (mostly due to differences in the way that author, title or source information was entered). A further 28 (18%) were indexed in EMBASE, leaving only 16 (11%) truly unique BIOSIS records. None of these were found in the CCTR. We also identified 80 possible randomised trials, and are currently assessing them. The free-text nature of the search resulted in some trials from non-Renal Group topic areas e.g. cardiac disease and renal cancers.

Conclusion: Since at least 78% of the BIOSIS records from the original multifile search were also in MEDLINE, searches of BIOSIS are most efficient when online deduplication is available. However, the records already examined have shown several ways in which duplicate records can miss detection: 1) Titles have been translated differently for the two databases from original languages into English. 2) Titles of Chinese language journals are indexed differently e.g. Medline tends to use the old Wade-Giles version, BIOSIS the Pinyin version, Embase mainly uses English language translations. 3) Misspelling or alternative spellings of authors names. 4) BIOSIS cites some journal names differently, especially when they have supplements. The overlap could be therefore be as high as 90% when records are assessed individually. Despite the modest retrieval of truly unique records from BIOSIS, the number of trials identified from the two sample periods indicates that BIOSIS should be considered by Cochrane groups. We intend to evaluate the remaining unique Biosis records for 1982‚97.

Presented at the 9th International Cochrane Colloquium, Lyon, October 2001

Correspondence
Ruth Mitchell
Cochrane Renal Group
Centre for Kidney Research
The Childrenís Hospital at Westmead
Locked Bag 4001
Westmead NSW 2145 Sydney Australia
Tel: + 61 2 9845 3049
Fax: +61 2 9845 3038
Email:Ruth Mitchell