Abstract
Background Reliable ascertainment of cancer recurrence is essential in order to evaluate therapeutic interventions to treat advanced disease. Specification of the timing of cancer recurrence using administrative claims from SEER-Medicare is challenging and has been limited to the use of algorithms based mainly on ICD9 diagnosis codes.
Aims We sought to identify scenarios or patterns of care that suggest cancer recurrence or progression using the detailed electronic medical record data contained in the HMO Cancer Research Network’s (CRN) Virtual Data Warehouse (VDW).
Methods Using data from the Virtual Tumor Registry (VTR) from two HMO CRN health plans where the tumor registry staff conduct annual chart abstraction to ascertain recurrence, we identified incident cases of stage 1–3a breast, colorectal, lung, and prostate cancer for the years 2000–2003. The VTR data was linked to VDW pharmacy, procedure, and diagnosis data from the subject’s incident cancer diagnosis date through the end of 2008. Descriptive analyses were performed. Scatter plots of events, including receipt of chemotherapy services, were created to determine “black out” periods associated with first course treatment and to identify other patterns of care consistent with recurrence or progression.
Results The proportion of incident cases with a noted recurrence during the observation period varied from 6.3% of prostate cancers to 19.1% of lung cancers. Fewer than 60% of cases (varied by cancer site) with a tumor registry-documented recurrence would have been identified as having a recurrence using an algorithm relying only ICD codes indicative of metastatic disease. Subjects with a noted recurrence had much higher use of advanced medical imaging (including CT, MRI and PET), laboratory procedures, and more pharmacy dispensings for narcotic analgesics. Scatter plots of chemotherapy services by month suggest that the interval of initial therapy among patients without recurrence varies by cancer type, so that algorithms for detecting recurrence based on these services need to be disease-specific.
Conclusions EMR data such as that available from the CRN VDW may be an optimal data source to be used to identify cancer recurrence. Next steps include the development of novel analytic strategies to estimate each patient’s probability of cancer recurrence.
