Background: A substantial proportion of cancer-related mortality is attributable to recurrent, not de novo metastatic disease, yet we know relatively little about these patients. To fill this gap, investigators often use administrative codes for secondary malignant neoplasm or chemotherapy to identify recurrent cases in population-based datasets. However, these algorithms have not been validated in large, contemporary, routine care cohorts.
Objective: To evaluate the validity of secondary malignant neoplasm and chemotherapy codes as indicators of recurrence after definitive local therapy for stage I–III lung, colorectal, breast, and prostate cancer.
Research Design, Subjects, and Measures: We assessed the sensitivity, specificity, and positive predictive value (PPV) of these codes 14 and 60 months after diagnosis using 2 administrative datasets linked with gold-standard recurrence status information: CanCORS/Medicare (diagnoses 2003–2005) and HMO/Cancer Research Network (diagnoses 2000–2005).
Results: We identified 929 CanCORS/Medicare patients and 5298 HMO/CRN patients. Sensitivity, specificity, and PPV ranged widely depending on which codes were included and the type of cancer. For patients with lung, colorectal, and breast cancer, the combination of secondary malignant neoplasm and chemotherapy codes was the most sensitive (75%–85%); no code-set was highly sensitive and highly specific. For prostate cancer, no code-set offered even moderate sensitivity (≤19%).
Conclusions: Secondary malignant neoplasm and chemotherapy codes could not identify recurrent cancer without some risk of misclassification. Findings based on existing algorithms should be interpreted with caution. More work is needed to develop a valid algorithm that can be used to characterize outcomes and define patient cohorts for comparative effectiveness research studies.
*Department of Medical Oncology, Center for Outcomes & Policy Research, Dana-Farber Cancer Institute
†Harvard Medical School, Boston, MA
‡Institute for Health Research, Kaiser Permanente Colorado, Denver, CO
§Division of Biostatistics, The University of Toronto, Toronto, ON, Canada
∥The Dana-Farber Cancer Institute, Boston, MA
¶The National Cancer Institute, Bethesda, MD
#Center for Health Research, Kaiser Permanente Northwest, Portland, OR
Supplemental Digital Content is available for this article. Direct URL citations appear in the printed text and are provided in the HTML and PDF versions of this article on the journal's Website, www.lww-medicalcare.com.
M.J.H. and D.P.R. contributed equally to this work and are considered co-first authors.
Supported by a grant from the National Cancer Institute (RC2 CA148185-01 to J.C.W. and D.P.R.) and an NCI cooperative agreement (U19 CA79689, Cancer Research Network). The American Society of Clinical Oncology (Career Development Award) and Susan G. Komen for the Cure (Career Catalyst Award) provided salary support to M.J.H. The work of the CanCORS consortium was supported by grants from the National Cancer Institute (NCI) to the Statistical Coordinating Center (U01 CA093344) and to the Primary Data Collection and Research Centers (Dana-Farber Cancer Institute/Cancer Research Network U01 CA093332; Harvard Medical School/Northern California Cancer Center U01 CA093324; University of Iowa U01 CA01013; University of North Carolina U01 CA093326); and by a Department of Veterans Affairs grant to the Durham VA Medical Center VA HSRD CRS 02 164.
The authors declare no conflict of interest.
Reprints: Michael J. Hassett, MD, MPH, Department of Medical Oncology, Center for Outcomes & Policy Research, Dana-Farber Cancer Institute, 450 Brookline Avenue, Boston, MA 02215-5450. E-mail: email@example.com.