The retrieval of randomized clinical trials in liver diseases from the medical literature: manual versus MEDLARS searches.

Recently, the usefulness of MEDLARS computer searches in biomedical research was questioned; the conclusion was drawn that to completely capture a specific population of articles, the MEDLARS system was inadequate. This conclusion was based on a comparison of MEDLARS and manual searches for articles on random clinical trials in liver disease for the period 1966-1982. The present study re-evaluated the validity of this conclusion. In the initial revised search strategy, the recall of valid articles was not significantly improved, as compared with the original MEDLARS search strategy, but precision of this revised search was significantly increased. The number of valid articles recalled was significantly increased by broadening the MEDLARS search strategy criteria, but not without a simultaneous and significant decrease in precision. A limiting capture rate of about 79% of the population of articles on random clinical trials was reached when all logical expansions of the search strategy were exhausted. Strategies for increasing the efficiency of MEDLARS searches in general are discussed citing limitations specific to the population of random clinical trial articles.