Administrative health care databases are increasingly used as a source of data for economic studies in cancer. In order to adjust for disease severity, several investigators have utilized ICD-9 codes indicating metastases as a proxy for cancer staging. OBJECTIVE: To determine the validity of using ICD-9-CM codes indicating metastases as a proxy to classify lung cancer patients by stage of disease. METHODS: This retrospective database analysis used diagnosis codes to classify subjects to either localized or advanced stage disease and then compared this classification to the tumor registry staging, which was considered as the "gold standard". Study subjects included all lung cancer patients treated at an academic institution during 1996-97 who were also members of a large insurance company. Data was derived from inpatient cancer-related claims linked with the institution's tumor registry data. Advanced stage disease (stages II to IV) was defined by claims indicating lymph node involvement or metastases (ICD-9 codes 196-199.1). The tumor registry staging of the disease for these patients were clustered into two groupings, stages 0-I (localized) and stages II-IV (advanced). RESULTS: Tumor registry entries were identified for 85.7% of patients. The crude concordance between the claims and tumor registry classifications was 74.2% (Kappa coefficient=0.4848). The positive predictive value of identifying localized disease utilizing ICD-9 coding was 57.6%,while the predictive value of a negative test was 91%. The sensitivity and specificity for dichotomized disease stage was 86.4% and 68.2% respectively. CONCLUSIONS: For a population of lung cancer patients in an academic institution, the use of ICD-9 coding was associated with modest predictability for disease staging. The use of ICD-9 coding as a proxy for disease staging in economic evaluations should be executed with caution.