Materials and Methods
Data Registries and Study Population
The Cancer of the Prostate Strategic Urologic Research Endeavor (CaPSURE) is a longitudinal, observational database of men with biopsy-proven prostate cancer recruited from a total of 40 urologic practice sites. The majority of sites are community based; four university-affiliated Veterans Affairs (VA) medical centers are included. CaPSURE collects ~1000 clinical and patient-reported variables and men are followed until death or study withdrawal. Additional details regarding project methodology have been reported previously. Of 13 821 men registered in the database, 9333 were diagnosed between 1998 and 2006 and were included in time-trend analysis. Of these, 5156 had detailed clinicodemographic data available for multivariate analysis.
The Medicare claims data set includes information from the Medicare Provider Analysis and Review (MEDPAR) and the National Claims History records, and includes enrollment, claims and hospitalization data based on the International Classification of Diseases (ICD-9, 9th revision) and procedure codes using the Current Procedural Terminology coding system (4th edition, CPT-4). Medicare does not include clinical information, such as stage or grade, required for tumor risk stratification. Between 1998 and 2006, 77 216 men were diagnosed with prostate cancer via report of both ICD-9 code 185 and CPT codes for either a prostate biopsy or a TURP within 180 days of the ICD-9 code report using a 5% Medicare sample. Men under age 66 were excluded, as were those not covered continuously by Medicare Part A&B (those with Medicare managed care coverage were also excluded). Of the remaining 55 395, 1073 managed primarily with orchiectomy were excluded for presumed metastatic disease, leaving 54 322 in the final cohort.
The i3 database (Ingenix, Salt Lake City, UT, USA) contains medical claims providing use and cost data on privately insured individuals of 75 large employers. Between 2002–2006, 63 150 men were diagnosed with prostate cancer as confirmed by the ICD-9 and CPT codes described above. As with the Medicare cohort, those managed primarily with orchiectomy or chemotherapy were excluded, as were those with discontinuous insurance coverage between the interval from 6 months before to 18 months after primary treatment. A total of 16 161 men met these criteria and were included in the analyses.
Data Analysis
Demographic data available for all three cohorts included age, race, place of residence and primary treatment. Additional clinical information for the CaPSURE cohort included PSA, Gleason grade, clinical stage, insurance type, education level and household income. For CaPSURE, these clinical variables were used to stratify patients into clinical risk groups using the well-validated Cancer of the Prostate Risk Assessment (CAPRA) score.
Imaging tests in CaPSURE are reported directly by treating clinicians. In the Medicare and i3 databases, CPT codes were searched to identify BSs (CPT 78306), CT scans (CPT 72192, 72193, 72194) and MRI scans (CPT 72195, 72196, 72197). Imaging utilization for all three tests in all three groups was determined if a test was performed during the staging interval, defined as the period between diagnosis of prostate cancer and date of initial treatment. For men who were not found to have an associated treatment, the staging interval was defined as 180 days following diagnosis. Misclassification of patients undergoing watchful waiting/active surveillance in this category of patients with no treatment was minimized by ensuring there was no evidence of further follow-up. Among men who underwent treatment with any form of radiation therapy, pre-treatment dosimetry planning imaging was distinguished from staging imaging via different CPT codes and excluded.
The rates of BS, CT and MRI were determined in each cohort of men, and time trends were analyzed by linear regression. For men in CaPSURE, demographic and clinical predictors of test utilization were explored using a multivariable regression model. All analyses were performed using Stata statistical software, version 10.1 (StataCorp LP, College Station, Texas, USA).