This analysis utilized the CDC Breast and Prostate Cancer Patterns of Care Study (CDC-POC-BP) cohort which included re-abstraction (in 2007–2009) of medical records and external linkages with provider databases for men with histologically confirmed Prostate Cancer (PCa) diagnosed in 2004 in seven NPCR registries (Georgia, Louisiana, California, Minnesota, Kentucky, North Carolina and Wisconsin). Sampling methodology for random selection of cases by race/ethnicity, registry, and facility characteristics has been previously described^5,14,19,20. Treatment related variables for these cases were verified and supplemented by re-abstracting hospital records in addition to obtaining treatment information from physicians’ offices, ambulatory surgical centers, radiation treatment facilities, and long term care facilities for some cases.

We restricted our analysis to patients with localized PCa who received definitive initial treatment (RP or EBRT) within 6-months of pathology-confirmed diagnosis, accounting for 57% of the CDC-POC-BP cohort (those receiving RP followed by adjuvant or salvage EBRT [n=108], were excluded). EBRT included either 3D Conformal Radiation Therapy (CRT) or Intensity Modulated Radiation Therapy (IMRT) and external radiation with unspecified modality. RP included both open and minimally invasive surgical approaches (but the specific type of surgery received was not specified in the data abstracted). We focused only on men receiving either of these two most common forms of definitive therapy and excluded those treated with brachytherapy, either alone or in combination with EBRT (due to the different clinical indications for treatment between brachytherapy and RP or EBRT alone in this time period). We also excluded those initially receiving conservative therapy (active surveillance, expectant management, or primary androgen deprivation therapy). Finally, we excluded 33 patients who died within 6 months after diagnosis with no record of treatment, 239 patients that underwent cryo- or non-radical prostate surgery as well as 273 patients with missing information on AJCC Stage, PSA and/or Gleason Score. After these exclusions our analytic sample consisted of 4491 cases (2861 that were treated with RP and 1630 that were treated with EBRT).

Categories used for each covariate are shown in Table 1. Demographic covariates included age at diagnosis, Race/ethnicity, marital status and insurance status. Additionally, the 2000 U.S. Census tract-specific data based on the patient’s residence location at diagnosis was used to categorize level of urbanization and socio-economic status (based on poverty level and educational ascertainment^15,21,22). The registry location of the patient was categorized as A to G. Patient comorbidity severity was determined using the Adult Comorbidity Evaluation 27 (ACE-27), a validated chart-based comorbidity instrument^1–4. Recurrence risk groups were defined according to the National Comprehensive Cancer Network (NCCN) guidelines for PCa (ver. 1.2002), which utilizes a combination of clinical T stage (AJCC, American Joint Committee on Cancer Clinical Staging System, 6th Ed.), Gleason Score from trans-rectal ultrasound guided biopsy and prostate specific antigen (PSA) level. The definitions of the risk groups are: Low (T1–2a AND Gleason Score ≤ 6 AND PSA < 10 ng/mL), Intermediate (T2b-T2c OR Gleason Score 7 OR PSA 10–20 ng/mL) and High (≥ T3a OR Gleason Score 8–10 OR PSA > 20 ng/mL).

Provider (the physician whom diagnosed the malignancy) and facility characteristics were linked to patients using external information sources that were available for 2004. Provider year of graduation from medical school (categorized by decade starting from 1950 to 1990) and type of practice (solo vs. group) was obtained for the treatment year from the Medicare Physician Identification and Eligibility Registry (MPIER) File, maintained by the US Centers for Medicare and Medicaid Services as previously described^3,5–7. Facility ownership (for profit versus non-profit/government funded) and facility teaching status (based on the presence of post-graduate medical training) corresponding to the treatment year were determined from the American Hospital Directory (AHD), which includes data from the Graduate Medical Education Database (American Medical Association, Chicago, Illinois). The distance (miles) from the primary treatment (surgery or radiation) facility for each patient was determined by the great circle distance and was categorized by quartiles (<5, 5–9, 10–14 and 15+ miles). County level population estimates of males in 2004 from SEER and number of urologists obtained from the Area Resource File were used to calculate urologist density per 100,000 men as previously described^5,8–10. Based on quintiles of the residential distribution of patients, ratio-cut points were 0, >0–5.6, >5.6–6.8, >6.8–10.6 and >10.6–26.0 urologists/100,000 men.

We utilized two quality of care indicators from the RAND consensus method and the Physician Consortium for Performance Improvement (PCPI) that are endorsed by the American Urological Association (AUA) and could be assessed from variables available from the CDC-POC-BP cohort^2,11–13. These included 1) receiving a staging bone scan that is not recommended for patients with NCCN low-risk PCa and 2) receipt of concomitant androgen deprivation therapy (ADT) in those undergoing EBRT for NCCN high-risk PCa^2,14. Additionally, we also considered surgical quality of care in those treated by RP, including 1) receipt of pelvic lymph node dissection (PLND) in patients with NCCN intermediate or high risk PCa at the time of RP^2,3,6,15 and 2) achieving negative surgical margins in patients with pathologic T2 disease.

The data were analyzed using SAS software (Proc Genmod). Weights were calculated by the inverse of the sampling fractions used by each registry for each sampling stratum to represent the source population. Unadjusted associations between initial treatment (RP or EBRT) and socio-demographic, location, patient comorbidity, NCCN risk group, provider characteristics and patient distance to provider variables were assessed by chi-square tests. We utilized multivariable generalized estimating equation (GEE) models with a logit link to account for correlation in outcomes based on treatment facility. This analysis was performed on a subset of patients (n=3875) in whom facility level information was available. Physician identifiers were not available in our cohort to adjust for correlation at this level. We included variables that showed associations in unadjusted analyses (p < 0.1) and/or those deemed a priori to be clinically important. The final model to determine predictors of RP (versus EBRT) included age, race/ethnicity, SES, insurance, registry, ACE-27 comorbidity severity, NCCN risk group, provider’s medical school graduation date, practice type (solo vs. group), facility ownership, teaching status of facility, distance to treatment facility and regional urologist density. One registry (labeled as ‘E’) did not contribute data on provider characteristics and was excluded from the final model.

Additionally, 4 separate GEE models with each of the quality of care measures as the outcome variable were constructed to determine the association between patient and provider/facility characteristics with these outcome measures. Variable selection for each of these models was based on the same technique for assessing predictors of definitive treatment modality. Race/ethnicity was retained in all models as there is potential evidence that race impacts multiple measures of quality of surgical care, and specifically the utilization of pelvic lymph node dissection (PLND)^6–8,10,16. As we were specifically interested in the association of provider and facility level variables on quality of care measures, all these variables were included in all models.

A total of 2861 (66.2% based on weighed percentages) men underwent RP while 1630 (33.8%)were treated by EBRT. The breakdown of EBRT by treatment modality included IMRT, 3D-CRT and unspecified external beam treatment in 548 (33.6%), 102 (6.3%) and 980 (60.1%) patients, respectively. The characteristics of patients receiving RP or EBRT as initial treatment are summarized in Table 1. In comparison to those receiving EBRT, younger patients were more likely to receive RP than older patients (e.g. 88.9% of those <60 had a RP compared to 11.3% of those 75+). Black patients and those with lower SES were less likely to receive RP compared to those in other racial/ethnic groups or in higher SES groups. There was evidence of significant variation in the utilization of RP between registries, ranging from 56.3% to 76.0%. Patients with no medical comorbidities were more likely to receiving RP compared to those with comorbidities of any severity (e.g. 76.0% of those with no comorbidities had a RP compared to 53.5% of those with severe comorbidities). Likewise, men with in the low NCCN risk group were more likely to receive RP compared to those in the high risk group (71.8% vs. 47.5%). There were statistically significant differences in provider and facility characteristics for patients receiving RP vs. EBRT (Table 2). Specifically, those who received care from a for-profit facility (vs. non-profit) as well as those seen at a hospital that had a teaching designation (vs. non-teaching) were more likely to receive a RP.

GEE regression analysis (used to account for the effect of multiple patients linked to a facility) was utilized to assess the association of patient and provider/facility characteristics with the receipt of RP versus EBRT (Table 3, n=2953 used in model, after exclusion of cases with missing values). Demographic factors were related to treatment choice. Older age (adjusted OR: 0.07, 95% CI: 0.05–0.11 for 70–74 vs < 60), as well as Black race versus White (OR: 0.47, 95% CI: 0.30–0.73), and being single vs. married (0.72, 95% CI 0.54–0.98) were associated with a decreased risk of receiving RP. Clinical factors including greater severity of comorbidities (vs. none) and high NCCN risk group (vs. low) were also associated with lower risk of having a RP. High versus low SES was associated with an increased risk of receiving RP as was residing in an urban setting vs rural. Among the provider/facility variables, group versus solo practice (OR: 0.47, 95% CI: 0.25–0.91) was associated with a decreased risk of RP. Although not statistically significant after accounting for clustering by facility, there was a decreased risk for receiving a RP at a teaching facility (vs. non-teaching), and an increased risk of RP if seen at a for profit facility (vs. non-profit/government facility).

A total of 1455 (of 1575) NCCN low-risk prostate cancer patients had information available on receipt of bone scan. There were 444 patients (30.5%) who received a bone scan that was not recommended and the GEE regression model that identified predictors of this are shown in Table 4. Compared to Whites, Blacks (OR: 1.80, 95% CI: 1.00–3.25) and API/AI/AN’s (OR: 2.76, 95% CI: 1.16–6.56) had an increased risk of receiving a bone scan that was not recommended. Medicaid insured patients (OR: 2.78, 95% CI: 1.24–6.26) also had an increased risk of a bone scan that was not recommended. While none of the facility variables were significantly associated with the use of bone scans, there was an increased risk of receiving an bone scan that was not recommended in areas with a higher density of urologists.

Surgical quality of care was assessed by identifying NCCN intermediate and high risk patients who appropriately received pelvic lymph node dissection at the time of RP (Supplementary Table 1, n = 1289 used in model). A total of 77.3% of patients underwent appropriate PLND at the time of RP. Men aged 65–59 (vs. <60) and those in the highest NCCN risk group (vs. intermediate) were more likely to receive PLND. Contrary to expectations, more recent year of graduation of the Urologist was associated with a decreased risk receiving appropriate PLND (Adj. OR: 0.36, 95% CI: 0.15–0.88 for 1990+ vs. 1950–1969) and RP performed at a teaching facility was associated with a decreased risk of receiving PLND (OR: 0.55, 95% CI: 0.34–0.90).

Surgical quality of care was also assessed by determining predictors of achieving a negative surgical margin in patients with organ confined (pathological T2) PCa (Supplementary Table 2, n=620 used in model). High versus low pre-operative NCCN risk was associated with a decreased risk of achieving a negative margin (OR: 0.32, 95% CI: 0.13–0.75). Further, higher urologist density was associated with increased risk of achieving a negative margin (OR: 3.30, 95% CI: 1.58–6.90 for 10.6–26 Urologist/100,000 Men vs. none).

Quality of care for radiation therapy (EBRT) was assessed by the concomitant prescription of ADT during treatment of NCCN-defined high risk PCa. The regression model was limited to 224 patients (after including those for whom outpatient records had been reviewed) and but did not converge.