Health Services & Outcomes Research Methodology 2: 67±76, 2001 # 2001 Kluwer Academic Publishers. Manufactured in The Netherlands. Using Hospital Tumor Registries to Identify Research Subjects MICHAEL SCHOENBAUM*, MEREDITH L. KILGORE, BEVERLYA. WEIDMER, SANDRA H. BERRY, JOSE J. ESCARCE, DANA P. GOLDMAN RAND, 1700 Main Street, Santa Monica, CA 90407 JOY H. LEWIS University of California, 405 Hilgard Ave., Los Angeles, CA 90095 NIKHILWAGLE Harvard University, 25 Shattuck Street, Boston, MA 02115 Received June 20, 2000; revised February 20, 2001; accepted February 20, 2001 Abstract. Hospitaltumorregistriesmaypermitef®cientidenti®cationofrelativelylargenumbersofpatientsfor clinical and health services research. This study investigates the feasibility and cost of using hospital tumor registry data for identifying cancer patients with particular clinical characteristics, corresponding to the protocol entry criteria offour randomly sampled Phase III clinical treatment trials for cancer. We screened tumor registry data on 717 patients diagnosed with cancer in 1997 and 1998 who had been identi®ed and abstracted by the registry of a large academic medical center by April, 1999; and we abstracted the medical records of the 122 patients who passed the registry screen. For each clinical pro®le, the registry screen eliminated a substantial fraction of patients who did not meet the relevant criteria. Of the patients identi®ed from the tumor registry as potential matches, 41% matched the relevant clinical pro®le based on detailed medical records review. The cost and effort of the registry-based method were substantially lower than would have been necessary if we had reviewed medical records without the registry screen, suggesting that tumor registry data can be a relatively ef®cient tool for identifying research subjects. Keywords: cancer, tumor registry, research methods, research costs 1. Introduction Clinical researchers often face the task of identifying patients with speci®c character- istics in order to study treatment costs, patient satisfaction with care or quality of care, or to ®nd patients for clinical trials. Small numbers of patients at individual institutions can be identi®ed in a variety of ways, including conversations with providers or by reviewing medical records. However, such methods are likely to be impractical for identifying relatively larger numbers of patients with speci®c characteristics and, in the case of ad hoc methods such as identi®cation by providers, may produce biased samples of patients. *Corresponding author: Michael Schoenbaum, RAND, 1200 South Hayes St., Arlington, VA 22202. Fax: 703-413-8111, E-mail: mikels@rand.org