The study protocol was approved by the local institutional review board at the University of Michigan.
Surveillance, Epidemiology, and End Results (SEER)-Medicare
The SEER program consists of a group of population-based tumor registries in selected geographic areas of the US . It covered 14% of the US population until 2001, and 26% thereafter. Medicare is a federal program that covers health services for 97% of persons aged 65 years and older. The information in the two programs has been linked. The SEER-Medicare database also contains the Hospital file, which includes information on hospital characteristics such as academic affiliation and is derived from the Provider of Service survey submitted by hospitals to Medicare. The SEER-Medicare database version used for this study contains Medicare claims through 2008 and cancer cases from SEER through 2007.
American Medical Association (AMA) Physician Masterfile
The AMA Physician Masterfile collects information on all physicians in the US, regardless of membership in the AMA . The information is collected from primary sources, such as medical schools, residency training programs, state licensing agencies, and the American Board of Medical Specialties. Physicians are also surveyed every 3 years regarding their current practice.
Men with incident prostate cancer from 2000 through 2002, and from 2004 through 2007, (men diagnosed in the year 2003 were excluded for reasons described below in the Study Variables section) who were at least 66 years old at diagnosis (n = 191,990) were initially selected. To ensure complete information, patients not enrolled in both Medicare Part A and Part B for 12 months before and 6 months following their cancer diagnosis (n = 19160), who died within 6 months of diagnosis (n = 3527), were members of an Health Maintenance Organization (n = 47833), or diagnosed by autopsy or on a death certificate (n = 2388) were excluded. The main interest for this study was the examination of ADT use for potentially inappropriate indications given the large changes in such use noted after the reimbursement cuts  and known variations in rates of such use as a function of urologist characteristics . We previously defined this group as men diagnosed with clinically localized (T1 or T2), low to moderate grade (Gleason 2 to 7) tumors, who did not receive treatment with radiation or radical prostatectomy. Use of ADT in that context would be potentially inappropriate given there is no clinical trial evidence supporting its efficacy as primary therapy , and even under theoretical considerations, it is difficult to show survival benefit from any intervention in such patients, due to the slow natural progression and competing risk of death from causes other than prostate cancer . This led to the exclusion of an additional n = 91913 patients (in whom ADT use would have been categorized as either discretionary or appropriate), leaving n = 27169 eligible patients for the main analysis.
Physicians providing care to patients within a year of diagnosis were initially identified through encrypted Unique Physician Identifier Numbers (UPINs) on Medicare physician claims, as previously published . Briefly, UPINs were linked to the AMA Physician Masterfile and only physicians with urology as their primary specialty code were selected. Patients who did not see at least one urologist in the year after diagnosis on at least two different days were excluded. If a patient saw 2 or more urologists, they were assigned to the urologist with ≥75% of urologist visits in the year after diagnosis. If no single urologist accounted for ≥75% of the visits, the patient was excluded.
Patient demographic and tumor characteristics were derived from the SEER records in the linked database and used to categorize patients by age, ethnicity, SEER region of residence at the time of diagnosis, year of diagnosis, clinical stage (T1 through T4), and grade (low - Gleason 2–4; moderate - Gleason 5–7; poor - Gleason 8–10). Stage was assigned using the SEER Extent of Disease-Clinical Extension classification system prior to 2004 and using the Collaborative Stage Clinical Extension system from 2004 onwards . Cancer grade was categorized only as a range as low (Gleason 2–4), moderate (Gleason 5–7) or high (Gleason 8–10) prior to 2003, but in 2003 only Gleason 7 was switched to the high grade category. From 2004 onwards, individual Gleason scoring has been available, allowing classification into similar groupings as for the data prior to 2003. To ensure comparability across the study period men with incident prostate cancer in 2003 were therefore excluded. The socio-economic characteristics of each patient were based on the percent of adults with less than 12 years of education and median income of the zip code of residence from the 2000 United States Census data. Comorbidity was measured using an adaptation of the Charlson Comorbidity Index for use with Medicare physician claims data, based on the period one year prior to diagnosis of prostate cancer [16,17].
Urologist board certification was available from the AMA based on information from the American Board of Medical Specialties . Patient panel size was defined as the number of patients with incident prostate cancer (regardless of ADT use) assigned to each urologist over the entire study period, and categorized as <14, 14–44, 45–74, and ≥75 patients. The categorization used was similar to an approach we utilized in a previous publication, chosen to ensure a reasonable distribution for the number of patients, with cut-offs roughly corresponding to the 2nd quartile, 3rd quartile and 90th percentile . Hospital academic affiliation was available from the SEER-Medicare Hospital file. Hospitals with a major academic affiliation were those that played an important role in the teaching program of a medical school. Hospitals with a minor academic affiliation had academic involvement limited to hosting of a residency program or occasional medical student rotations . Urologists were categorized as having an academic affiliation if all their inpatient Medicare claims submitted were from a hospital with major or minor academic affiliation. These urologists would be more likely to be salaried, through employment with a medical school, or hospital. All other urologists were categorized as having no academic affiliation. These urologists would be more likely to be in private practice or part of single specialty groups, and therefore derive most of their income from fee-for-service activities. In initial analyses of academic affiliation using three categories (major, minor or none) patterns of ADT use were essentially identical between urologists with major and minor academic affiliation. Therefore, to optimize statistical power the main analyses were presented using a binary classification: major or minor academic affiliation vs. none.
The outcome was receipt of androgen deprivation. Androgen deprivation was defined as the receipt of at least one dose of a GnRH agonist (identified through Medicare claims codes used to designate each dose given of injectable medications ) or orchiectomy (defined by the presence of the Current Procedural Terminology codes or International Classification of Diseases, 9th revision [ICD-9] procedure codes in the Medicare claims) in the first six months following diagnosis of cancer.
Differences across strata of urologist characteristics in the proportion of patients receiving ADT were tabulated. The effect of urologist characteristics on the outcome of use of androgen deprivation was evaluated using multilevel logistic regression models to account for clustering of patients within urologists. Models entering the urologist, patient and tumor characteristics listed above as independent variables were estimated. Odds ratios (OR) for the use of androgen deprivation for each urologist characteristic were calculated along with 95% confidence intervals (CI). As the main interest was to examine how the effect of urologist characteristics changed related to reimbursement cuts, the model results were stratified into two eras, 2000–2002 (prior to the reimbursement cut) and 2004–2007 (following the reimbursement cut). To specifically assess whether there were significant changes in the effect of urologist characteristics on use of ADT over the course of the two eras, tests of interaction were performed between each urologist characteristic and era, on the outcome of ADT use. In addition, the median urologist rates of androgen deprivation use as a function of academic affiliation were estimated from the models, and plotted by calendar year of diagnosis to show change in use over time. Analyses were performed with SAS version 9.3 (Cary, NC). All tests of statistical significance were two sided, with P values of less than .05 being considered significant.