| Abstract | Randomized clinical trials make up only a small fraction of published articles concerning care of the newborn infant and an even smaller fraction of articles about all human subjects. The busy pediatrician who wants to keep abreast of the medical literature requires strategies to detect such relevant studies promptly and reliably. Computer searching of MEDLINE is an attractive, potentially powerful but not sufficiently validated means of achieving this goal. Therefore, the sensitivity, specificity and positive predictive value of two MEDLINE search strategies designed to detect randomized clinical trials for prevention and treatment of newborn diseases among all original articles about human subjects appearing in ten pediatric and general medical journals during 1985 were determined. The yield of both MEDLINE searches was compared to that of a manual search. Fifty-three randomized clinical trials of newborn care were identified by hand search among 233 articles concerning care of the newborn and 2,988 original articles about human subjects. The sensitivities of the MEDLINE searches were 53% and 34%, respectively, and the postive predictive values were 82% and 69%, respectively. Specificity of both computer searches was virtually 100%. Twenty-one randomized clinical trials were not identified by either MEDLINE search strategy, 17 of them for failure of the indexer to assign any methodologic terms at all or failure to assign sufficiently stringent methodologic terms. Consequently, sensitivities were higher, 77% and 68% respectively, when no methodologic terms were used during repeat searches. However, positive predictive values decreased concomitantly to 20% or less. It is concluded that searching MEDLINE for randomized clinical trials is not yet sufficiently sensitive to be recommended as the sole strategy of keeping up with this literature. |