Abstract
PURPOSE: Researchers are using diagnosis codes from health claims to identify metastatic disease in cancer patients. The validity of this approach has not been established. METHODS: We used the linked 2005-2007 Surveillance, Epidemiology and End Results (SEER)-Medicare data to assess the validity of metastasis codes at diagnosis from claims compared with stage reported by SEER cancer registries. The cohort included 80,052 incident breast, lung, and colorectal cancer patients aged 65 years and older. Using gold-standard SEER data, we evaluated sensitivity, specificity, positive predictive value, and negative predictive value of claims-based stage, survival by stage classification, and patient factors associated with stage misclassification using multivariable regression. RESULTS: For patients with a registry report of distant metastatic cancer, the sensitivity, specificity, and positive predictive value of claims never simultaneously exceeded 80% for any cancer: lung (42.7%, 94.8%, and 88.1%), breast (51.0%, 98.3%, and 65.8%), and colorectal (72.8%, 93.8%, and 68.5%). Misclassification of stage from Medicare claims was significantly associated with inaccurate estimates of stage-specific survival (P < .001). In adjusted analysis, patients who were older, black, or living in low-income areas were more likely to have their stage misclassified in claims. CONCLUSIONS: Diagnosis codes in Medicare claims have limited validity for inferring cancer stage and metastatic disease.