Developing optimal search strategies for detecting clinically sound treatment studies in EMBASE

Abstract

Objective: The ability to accurately identify articles about therapy in large bibliographic databases such as EMBASE is important for researchers and clinicians. Our study aimed to develop optimal search strategies for detecting sound treatment studies in EMBASE in the year 2000.

Methods: Hand searches of journals were compared with retrievals from EMBASE for candidate search strategies. Six trained research assistants reviewed fifty-five journals indexed in EMBASE and rated articles using purpose and quality indicators. Candidate search strategies were developed for identifying treatment articles and then tested, and the retrievals were compared with the hand-search data. The operating characteristics of the strategies were calculated.

Results: Three thousand eight hundred fifty articles were original studies on treatment, of which 1,256 (32.6%) were methodologically sound. Combining search terms revealed a top performing strategy (random:.tw. OR clinical trial:.mp. OR exp health care quality) with sensitivity of 98.9% and specificity of 72.0%. Maximizing specificity, a top performing strategy (double-blind:.mp. OR placebo:.tw. OR blind: .tw.) achieved a value over 96.0%, but with compromised sensitivity at 51.7%. A 3-term strategy achieved the best optimization of sensitivity and specificity (random:.tw. OR placebo:.mp. OR double-blind:.tw.), with both these values over 92.0%.

Conclusion: Search strategies can achieve high performance for retrieving sound treatment studies in EMBASE.

Figure 1

Formula for calculating the sensitivity, specificity, precision, and accuracy of EMBASE searches for detecting clinically sound treatment studies. Sensitivity = a/(a+c); specificity = d/(b+d); precision = a/(a+b); accuracy = (a+d)/(a+b+c+d). All articles classified during the manual review of the literature = (a+b+c+d)