Can administrative data identify incident cases of colorectal cancer? A comparison of two health plans

Although commercial health insurance records are a potentially valuable source of data for cancer outcomes studies, the accuracy of administrative codes for identifying persons with cancer is unknown. The objective of this study was to determine the accuracy of administrative codes for identifying persons with newly diagnosed colorectal cancer. The Washington state SEER cancer registry was linked with enrollment files from two large commercial health plans: a fee-for-service plan and a staff-model HMO with combined enrollment of more than 1.7 million. The accuracy of ICD-9 codes for colorectal cancer for enrollees diagnosed between January 1, 1996 and December 31, 1999 was compared for each plan, using SEER records as the "gold standard." The positive predictive value of administrative codes was 33.4% at the fee-for-service plan and 30.1% at the HMO plan (difference p = 0.26). The overall sensitivity of administrative codes was 91.7% at the fee-for-service plan and 93.1% at the HMO plan (difference p = 0.09). Positive predictive values were higher for inpatient records, while sensitivity was higher for outpatient records and for those with more advanced tumor stages at diagnosis. In conclusion, administrative records for identifying new cases of colorectal cancer in two large health plans showed poor positive predictive value but high sensitivity. Accuracy did not vary substantially by insurance plan. Researchers without tumor registry or other high-quality sources to verify records should not rely on administrative data alone from commercial health plans to identify incident colorectal cancer cases.