Abstract
Background: Collecting chemotherapy administrations is important to any study of cancer treatment. However, abstracting data from medical records can be time consuming while chemotherapy data from tumor registries may be lacking sufficient detail. The goal of this study was to describe the accuracy of automated chemotherapy data from administrative health plan databases compared with data from tumor registries.
Methods: From eight Cancer Research Network (CRN) tumor registries, we identified 1218 women diagnosed with primary ovarian cancer between January 2004 and June 2006. Using the Virtual Data Warehouse (VDW), we linked cases to all known automated chemotherapy administrations through June 2006 by pooling administrative health plan data from procedure, pharmacy, and diagnosis automated databases. We then calculated the sensitivity of these three data sources for capturing any chemotherapy administrations compared to chemotherapy as coded by tumor registries at each site.
Results: The health plan data contained records of at least one chemotherapy administration in 51.3% of cases (range across sites 4.3%–80.0%) while the tumor registry data contained such records in 68.8% of cases (range 57.3%–95.2%). The overall sensitivity of administrative data compared to tumor registry data was 69.3% (range 2.9%–95.2%) and varied by administrative data source: 52.8% (range 0.0%–90.3%) using procedures, 46.5% (range 0%–80.7%) using pharmacy records, and 29.1% (range 0.0%–65.9%) using diagnosis codes. Sensitivity decreased over time from 81.4% among cases diagnosed January–June 2004, to 47.7% among diagnoses between January–June 2006.
Conclusions: Tumor registries appeared to capture the chemotherapy administrations more often than automated health plan administrative data. We noted wide variability in the sensitivity of administrative data by site. Capture of administrative data may be enhanced through site-specific coding of chemotherapy administration and additional understanding at each site of clinical and billing coding practices.
- Received September 11, 2008.
