OBJECTIVES: With increasing emphasis on using published evidence to quantify the value of drug treatment, we assessed the value of adding search terms identified from key articles to initial search strategies to locate studies in large subject areas. METHODS: Systematic review in five chronic disease settings (asthma, depression, diabetes, GI disorders, and migraine), including development of pre-defined criteria for study inclusion. Outcomes of interest were those relevant to the value of pharmaceutical treatment (economic, societal, and/or patient benefit). We constructed a MEDLINE search (via PubMed) with terms indicating drug treatment ("drug therapy" subheading) and outcomes ("Economics, pharmaceutical", "Costs and Cost Analysis," and "Quality of life" MeSH terms). We then reviewed indexing terms present in key articles found independently of the initial search, using an iterative process. Terms which led to identification of new, relevant citations without excessive increase in the total search yield were retained. Supplemental electronic (Current Contents, Cochrane, University of York NHS Centre) and manual searches were used to create a listing of accepted studies for comparison to the PubMed-only search results. RESULTS: Three thousand six hundred seventy-seven citations were downloaded and screened from all sources, resulting in 135 accepted papers across the five settings (asthma-20, depression-40, diabetes-16, GI disorders-41, migraine-18). Using initial search terms, we would have captured only two-thirds of the eligible studies in our PubMed search (range 55%-83% by setting). By undergoing the explorative process to identify additional terms, 90% of studies (122/135, range 80-100% by setting) were identified via PubMed. CONCLUSIONS: Careful iterative development of search terms should be a component of all systematic searches in PubMed. Supplemental electronic and manual searching remains necessary to obtain a comprehensive study set. PubMed indexing varies by clinical setting and journal subject, with somewhat weaker results for pharmacoeconomic/managed care journals (78%) than disease-specific (94%) or general medical (98%) journals.