The impact of selective searching on the results of systematic reviews

Article type
Authors
Hartling L1, Featherstone R2, Nuspl M2, Shave K2, Vandermeer B2
1Child Health Field; University of Alberta, Canada
2Alberta Research Center for Health Evidence; University of Alberta, Canada
Abstract
Background: One of the hallmarks of a well-conducted systematic review (SR) is a comprehensive search. However, empiric evidence is lacking on the impact of database selection on the results of SRs.
Objectives: To examine the impact on the results of SRs based on selective searching.
Methods: The sample was all child-relevant SRs with at least one meta-analysis from three Cochrane Review Groups (Acute Respiratory Infections (ARI), Infectious Diseases, Developmental Psychosocial and Learning Disorders; n = 129). We searched 13 databases to determine the proportion of relevant studies that were indexed in each source. We re-ran the primary meta-analysis with the studies identified in MEDLINE only and in MEDLINE plus each of the other databases. We calculated a ratio of the new point estimate to the original point estimate and a ratio of confidence interval widths. We determined how often results changed in statistical significance.
Results: These results are based on the 57 SRs from ARI; complete results for all SRs will be available in October 2015. The average number of relevant studies was highest for MEDLINE (85%), EMBASE (80%), and BIOSIS (65%). From a total of 398 studies, 65 were not found in MEDLINE; 20 of the 65 were found in EMBASE and 13 in BIOSIS; other databases identified three or less. The mean ratio of point estimates for MEDLINE to the reference standard for dichotomous outcomes (n = 37) was 1.06. The smallest ratio was 1.03 for MEDLINE+BIOSIS. The mean ratio of confidence interval widths of MEDLINE to the reference standard was 1.12; the smallest ratio was 1.09 for MEDLINE+EMBASE. The mean ratio of standardized mean differences (n = 19) was 0.01 for MEDLINE alone, with the mean ratio of confidence interval 1.17 (smallest for MEDLINE+EMBASE, 1.15). For MEDLINE alone, five meta-analyses changed in significance. The fewest number of meta-analyses changing in significance (n = 1) was for MEDLINE+EMBASE.
Conclusions: This study provides quantitative data regarding the impact on SR results of restricting searches to select databases. This information may be useful to increase efficiencies in the conduct of SRs.