6521 Background: Researchers are increasingly using diagnosis codes from administrative claims for cancer patients to identify metastatic disease at initial diagnosis or recurrence. However, the validity of metastasis codes on claims has not been established. We used the linked SEER -Medicare data to assess the completeness and validity of metastasis codes from Medicare claims for three common U.S. cancers. Methods: The study included 80,052 breast, lung, and colorectal cancer patients diagnosed with localized, regional, or distant disease in the SEER data between January 1, 2005 and December 31, 2007. From Medicare claims, patients were classified as having regional or distant disease at diagnosis if they had one hospital claim or two physician claims with metastasis codes within 3 months of diagnosis. Patients without claims with metastases codes were classified as having local disease. Using SEER data as the gold standard, we calculated sensitivity, specificity, positive and negative predictive values of metastasis codes on Medicare claims. We conducted multivariate logistic regression analysis to evaluate patient factors associated with stage misclassification for each cancer site. Results: For patients with distant disease per SEER data, the sensitivity and PPV of the claims to identify distant disease was: breast (50.6%, 67.3%), colorectal (72.2%, 68.8%) and lung cancer (42.1%, 88.6%). None of the measures for stage simultaneously exceeded 80% for sensitivity, specificity, and PPV for any of the cancer sites. In adjusted analysis, older, lower-income, and African American patients were more likely to have stage at diagnosis misclassified from Medicare claims. Conclusions: Use of diagnosis codes alone in Medicare claims will misclassify stage at diagnosis for cancer patients, particularly for patients with metastatic disease. Our findings also suggest that using diagnosis codes for metastasis to define recurrence in Medicare claims will be limited.
PS1-39: Diagnosis Codes for Cancer Metastasis on Medicare Claims Have Limited Accuracy and Completeness
