Surgery and radiation-based therapies are standard management options for men with clinically localized high-risk prostate cancer (PCa). Contemporary patterns of care are unknown. We hypothesize the use of surgery has steadily increased in more recent years.
Using the National Cancer Data Base for 2004–2013, all men diagnosed with high-risk localized PCa were identified using National Comprehensive Cancer Network criteria. Temporal trends in initial management were assessed. Multivariable logistic regression was used to evaluate demographic and clinical factors associated with undergoing radical prostatectomy (RP).
In total, 127 391 men were identified. Use of RP increased from 26% in 2004 to 42% in 2013 (adjusted risk ratio (RR) 1.51, 95% CI 1.42–1.60, P<0.001), while external beam radiation therapy (EBRT) decreased from 49% to 42% (P<0.001). African American men had lower odds of undergoing RP (unadjusted rate of 28%, adjusted RR 0.69, 95% CI 0.66–0.72, <0.001) compared to White men (37%). Age was inversely associated with likelihood of receiving RP. Having private insurance was significantly associated with the increased use of RP (vs Medicare, adjusted odds ratio 1.04, 95% CI 1.01–1.08, P=0.015). Biopsy Gleason scores 8–10 with and without any primary Gleason 5 pattern were associated with decreased odds of RP (vs Gleason score ⩽6, both P<0.001). Academic and comprehensive cancer centers were more likely to perform RP compared to community hospitals (both P<0.001).
The likelihood of receiving RP for high-risk PCa dramatically increased from 2004 to 2013. By 2013, the use of RP and EBRT were similar. African American men, elderly men and those without private insurance were less likely to receive RP.
Over 180 000 men were diagnosed with prostate cancer (PCa) in the United States in 2016.1 Approximately one-third with localized PCa are ‘high-risk’ based on National Comprehensive Cancer Network (NCCN) criteria.2 As the population ages and use of screening declines following the 2011 United States Preventive Services Task Force recommendation,3, 4 it is possible that high-risk cancers will comprise a greater proportion of the future PCa diagnoses.5
Practice guidelines recommend either radiation with androgen deprivation or radical prostatectomy (RP) as the initial treatment strategy for men with high-risk cancers.2, 6 For men with a reasonable estimated life expectancy, local therapy is favored over a systemic approach or observation.7, 8, 9
Recently, findings from multiple non-randomized comparative studies have suggested a survival benefit for men with high-risk PCa who received RP compared to external beam radiation therapy (EBRT).10, 11, 12, 13 We hypothesized the use of surgery has increased in recent years. As such, the objectives of our study were to: (1) evaluate temporal trends in the initial treatment of clinically localized high-risk PCa and (2) determine demographic and clinical factors associated with RP.
Materials and methods
Data source and patient population
Our cohort was derived from a hospital-based cancer registry, the National Cancer Data Base (NCDB) from 2004–2013. The NCDB captures data on 70% of cancer diagnoses in the United States from >1400 hospitals with cancer programs accredited by the American College of Surgeons’ Commission on Cancer (CoC) and American Cancer Society.14
All men diagnosed with localized PCa from 2004 to 2013 (n=1 065 906; Supplementary Figure S1) were identified. The NCDB collects data from reporting institutions based on CoC-accreditation status, which may change on a yearly basis. Only facilities that contributed at least one patient during each study year were included, hence, temporal changes in the management were not biased by institutional participation.
We excluded men with incomplete data for NCCN risk stratification (Gleason grade, clinical T stage and pre-treatment PSA) (n=278 089). Prior to 2010, the NCDB recorded Gleason score solely from the largest histologic specimen available. For example, if a patient received RP, this would supplant the previously documented prostate needle biopsy Gleason score. For consistency, from 2010–2013 we included Gleason score as a variable in a similar manner. We performed our sensitivity analysis with RP usage on patients diagnosed in 2010 through 2013 using biopsy Gleason score to determine high-risk status (n=46 518; Supplementary Figure S1) and biopsy Gleason score as an independent variable.
Clinical T stage was available for the entire study period. The NCDB uses the most updated American Joint Committee on Cancer (AJCC) Cancer Staging Manual edition during each year of diagnosis. Thus, prior to 2009, the 6th and 7th edition was used in the following years. There were no changes between the two that affected our clinical staging.15
We defined high-risk PCa using the 2016 NCCN risk stratification criteria: pre-treatment PSA >20 ng ml–1, Gleason 8–10 or cT3-4.2 We excluded men with low- or intermediate-risk PCa (n=626 758), PSA >50 ng ml–1 (n=25 607), missing treatment data (n=3 986) or missing demographic information (n=4 075; Supplementary Figure S1). Men with a PSA >50 ng ml–1 were excluded due to a high clinical suspicion of metastatic disease and their exclusion from previous RP clinical trials.7, 16
Covariates included facility type, facility geographic location, patient comorbidities (Deyo–Charlson comorbidity index (CCI)17) and insurance type. Facility type is based on CoC accreditation and is determined by the annual number of cancer patients a facility encounters and graduate medical education participation.18
Income was categorized into quartiles according to median household income for each patient’s zip code. Education attainment was categorized as the proportion of adults in a patient’s zip code who did not obtain at least a high school education. Income and education were derived from the 2012 American Community Survey data spanning 2008–2012 and divided into quartiles. Race and ethnicity were used to create a composite variable categorized as non-Hispanic White, non-Hispanic African American, Hispanic and other/unknown.
Distance between a patient and their treating facility was estimated based on the distance between the zip code of the patient’s residence and the facility and was categorized as <60, 60–120 and >120 miles. Each patient’s county of residence was defined based on a 2013 population as metropolitan (>249 999) or urban/rural (<250 000). Age and the log of pre-treatment PSA were analyzed as continuous variables.
Our primary outcome was initial treatment: RP, radiation or primary ADT. Primary RP was defined as RP performed within 1 year of diagnosis with or without androgen deprivation therapy (ADT). Radiation-based therapies were divided into four categories: EBRT without ADT, EBRT with ADT, EBRT and brachytherapy with or without ADT, and brachytherapy with or without ADT. If RP and a radiation-based therapy were both used within 1 year of diagnosis, the initial treatment was defined as the treatment performed first. Primary ADT was defined as ADT initiated within 1 year of diagnosis and no RP or radiation-based therapy. No treatment was defined as no treatment was performed within 1 year of diagnosis.
Categorical variables were compared using Pearson’s χ2 tests including changes in initial treatment between the years 2004 and 2013. The Mann–Whitney U-test was used to compare non-parametric continuous data. Modified Poisson regressions19 were used to evaluate the association of covariates with RP use among men with high-risk PCa. To account for variability between treatment facilities (n=1087), we adjusted for clustering of patients within treatment facilities.20 Adjusted risk ratio (RR) was calculated in this manner. Using Kaplan–Meier analyses, log rank tests and Cox proportional hazards, we assessed the effect of treatment on overall survival. For all analyses, tests were two-tailed, <0.05 was considered significant and Stata 13.0 (College Station, TX) was used. Institutional review board deemed our study exempt from approval due to use of de-identified, retrospective patient data.
Temporal trends in initial management of high-risk localized prostate cancer
Our final cohort included 127 391 men with high-risk localized PCa (Supplementary Figure S1). Use of RP increased from 26% in 2004 to 42% in 2013, use of radiation decreased from 55% to 43%, while ADT monotherapy decreased from 8% to 6%. The proportion of patients who did not receive any initial treatment decreased from 11% to 9% (all P<0.001; Figure 1 and Supplementary Table S1). The median (interquartile range) of percentage of patients receiving RP at each facility was 27% (13–44%).
Among all patients, use of EBRT with any combination of additional treatment was 44% overall (Supplementary Table S2). EBRT usage decreased from 49% (11% EBRT alone, 29% with ADT and 9% with brachytherapy) in 2004 to 41% (8% EBRT alone, 29% with ADT and 4% with brachytherapy) in 2013 (P<0.001; Figure 2). Brachytherapy use fell from 16% in 2004 (9% with EBRT and 7% without EBRT) to 6% (4% with EBRT and 2% without EBRT) in 2013 (P<0.001).
Demographic and clinical factors associated with radical prostatectomy use
The odds of receiving RP increased greatly from 2004 to 2013 (26–42%: RR 1.51, 95% CI 1.42–1.60 P<0.001; Tables 1 and 2). Each category of PSA >4.0 ng ml–1 was associated with decreased likelihood of RP. In total, 42% of men with PSA 0.1–4.0 ng ml–1 received RP whereas only 29% of men with PSA>20 ng ml–1 received RP (Continuous as log of PSA RR 0.84, 95% CI 0.83–0.85, P<0.001).
Patients with Gleason score 3+4 (43%) and 4+3 (39%) had higher likelihood of RP compared to men with Gleason score ⩽6 (29%; RR 1.35, 95% CI 1.29–1.41, P<0.001 and RR 1.29, 95% CI 1.22–1.36, P<0.001). Clinical stages cT2 and cT3 were associated with decreased likelihood of RP compared to cT1 (both P<0.001).
Other covariates associated with increased likelihood of RP included private insurance (vs Medicare, RR 1.04, 95% CI 1.01–1.08, P=0.015) and receiving treatment at a comprehensive or academic facility (vs community, RR 1.34, 95% CI 1.21–1.50, P<0.001 and RR 1.55, 95% CI 1.38–1.74, P<0.001, respectively). Overall, 88% of RP’s occurred at comprehensive or academic facilities (Table 1). In addition, living >60 miles from the treatment center greatly increased the odds of RP (60–120 miles, RR 1.35, 95% CI 1.28–1.43, P<0.001; >120 miles, RR 1.29 95% CI, 1.13–1.48 P<0.001).
Covariates associated with decreased odds of RP included African American race compared to White race (28 and 37%, respectively; RR 0.69 95%, CI 0.66–0.72, P<0.001; Table 2), older age, and Medicaid or no insurance compared to having Medicare.
To assess the connection between African American race and type of treatment facility, we ran our regression again limiting our analyses to patients treated at academic, and comprehensive cancer centers. We found the RR were similar to the results of our main regression (RR 0.69, 95% CI 0.65–0.74, P<0.001 and 0.68, 95% CI 0.63–0.73, P<0.001, respectively).
Compared to patients who were excluded based on missing risk stratification data (n=278 089), patients with information on Gleason score, PSA and clinical stage (n=787 817) were comprised of more White patients (73.0 vs 71.5%, P<0.001), were older (Median (interquartile range): 65 (59–72) vs 65 (59–71), P<0.001) and were comprised of more men with a Charlson comorbidity index of zero (85.2 vs 83.4%, P<0.001).
For the years 2010–2013, 50 288 men in our final cohort had information available on biopsy Gleason score. Of these, 46 518 (93%) were still considered high-risk by NCCN criteria when using biopsy Gleason score to risk stratify (Supplementary Figure S1). Further, we replicated our regression analysis of initial treatment on these men using biopsy Gleason score as an independent variable. This analysis (Supplementary Table S3) demonstrated few qualitative differences to the initial analyses (Table 2). However, higher clinical stage was no longer associated with decreased RP usage and Gleason scores 3+4 and 4+3 were no longer associated with higher RP usage (all P>0.05; Supplementary Table S2). Patients with biopsy Gleason 8–10 with and without any primary Gleason 5 pattern had significantly lower odds of RP usage (vs Gleason score ⩽6 RR 0.81, 95% CI 0.76–0.85, P<0.001 and RR 0.81, 95% CI 0.75–0.87, P<0.001, respectively). As an additional analysis, we assessed patients diagnosed with biopsy Gleason 8–10 disease from 2010–2013 (n=33 847). There were no qualitative differences between the multivariable regression restricted to this cohort and that in Supplementary Table S3. The odds of RP increased from 2010 to 2013 (RR 1.29, 95% CI 1.19–1.41, P<0.001).
Survival based on treatment
Median (interquartile range) for follow-up was 4.4 (2.6–6.5) years. Based on Kaplan–Meier analysis, the median overall survival for men who received no treatment was 8.3 years, EBRT with ADT was 9.9 years and ADT alone was 5.7 years (log rank P<0.001, Supplementary Figure S2). Median survival was not reached for men who received RP. The 10-year overall survival for no treatment was 41%, EBRT with ADT was 49%, ADT alone was 24% and RP was 77%. Using Cox proportional hazards to adjust for multiple covariates, we assessed the hazards ratio for each treatment. Compared to RP, no treatment (HR 2.98, 95% CI 2.83–3.15, P<0.001), EBRT with ADT (HR 1.60, 95% CI 1.53–1.68, P<0.001) and ADT alone (HR 3.36, 95% CI 3.17–3.56, P<0.001) were each associated with worse overall survival (Supplementary Table S4).
Several recent studies have documented encouraging long-term cancer-specific survival in men with high-risk PCa who receive RP.21, 22 Using the NCDB, we demonstrated surgery for men with high-risk cancer increased significantly from 2004 to 2013, and by 2013 RP rates were similar to those of EBRT.
The possible reasons for an increase in the use of RP are many. In the years leading up to our study period, population-based studies suggested improved cancer-specific survival benefit with RP compared to radiation.23, 24 Although these studies were limited by their retrospective design and unmeasured variables, they were likely influential on patients and physicians.25 Early in the study period, more population-based data continued to suggest a survival benefit for RP over radiation for men with high-risk disease, although all studies had varying degrees of methodological limitations.26, 27, 28 Our analysis of overall survival in the NCDB also demonstrated a survival benefit with RP, although our retrospective analysis was subject to selection bias. The decision of optimal management depends on the maturation of prospective randomized data.29
An analysis of a registry of 45 urology practices in the United States showed a similar increase in RP for men with high-risk PCa from ~25% during 2005–2009 to ~50% during 2010–2013.30 This study likely demonstrated higher rates of RP due to its registry being composed of several types of urology practices while the NCDB is composed of CoC hospitals. Greater than 80% of our cohort were treated at either an academic or comprehensive cancer center and we found patients treated at these facilities were more likely to receive RP compared to men treated at community hospitals.
Declines were seen in rates of brachytherapy with and without EBRT. This finding is consistent with a prior NCDB report31 which included men with all risk categories of prostate cancer, and could be due to changes in referral patterns related to reimbursement, advances with alternative therapies or lack of adequate training experience for brachytherapy. However, early trial results show an advantage in disease control with brachytherapy boost compared to EBRT alone for intermediate- and high-risk PCa.32
Use of primary ADT remained relatively low over the study period. This is encouraging given evidence demonstrating the superior efficacy of RT with ADT and lack of survival benefit for primary ADT, especially among high-risk patients in previous studies.33, 34 Although our survival analysis is limited to overall survival and may be subject to selection bias, data from the NCDB confirmed worse survival associated with primary ADT.
African American men with high-risk PCa were less likely to receive RP and instead had higher rates of non-treatment and radiation therapy. This was observed even after adjusting for multiple demographic and clinical factors. In addition, even though use of RP among African American men has increased over time, the gap of RP usage between African American men and all patients has also increased. Although these results has been previously demonstrated in the Surveillance, Epidemiology and End Results (SEER) database,35 there has been very little evidence to explain why. Interestingly, this treatment disparity was seen even when we limited our analysis to patients treated at academic and comprehensive cancer treatment center suggesting access to high volume centers does not explain the difference in RP receipt. It is possible these disparities in primary treatment may be secondary to racial differences in socioeconomics.36 While we were able to adjust for regional levels of education and income in our multivariable analysis, we could not account for individual levels of socioeconomic status which may further account for the racial disparities in treatment selection.
Previous data from the SEER has shown men with Medicaid or no insurance received definitive treatment less often compared to men with private insurance, especially among high-risk patients.37 In the NCDB, even after adjusting for age and other covariates, having private insurance was associated with higher odds of RP while having Medicaid or no insurance was associated with lower odds of RP compared to Medicare. These findings may be attributable to previous associations drawn between aggressive cancer treatment and the extent of insurance coverage,38 even within a Medicare population.39
The strengths of our study include the large and diverse patient population collected in the NCDB. Unlike other data sets such as the SEER database, NCDB provides valuable information on the use of ADT and the order in which treatments were received to accurately discern initial treatments. The NCDB also captures information on comorbidities and more complete information on patient insurance compared to SEER.
Our study is not without its limitations. The NCDB is a hospital-based cancer registry and not population-based. The data from the NCDB should be interpreted as generalizable to hospitals similar to those included in its registry. Nearly three quarters of patients in the current study were treated at a comprehensive or academic facility, thus treatment data from community and office-based health centers may be under-represented. Finally, a large number of patients were excluded due to missing data regarding risk stratification. However, although comparison of patients with and without missing risk data revealed statistically significant differences in age, race composition and comorbidities, the differences were relatively modest.
In the United States between 2004 and 2013, the likelihood of receiving RP increased for high-risk PCa. In 2013, the use of RP and EBRT, were similar. African American men, older men and those without private insurance were less likely to receive RP.
Supplementary Information accompanies the paper on the Prostate Cancer and Prostatic Diseases website (http://www.nature.com/pcan)