MS and J.D.S contributed equally to the work.
Radical prostatectomy vs radiotherapy vs observation among older patients with clinically localized prostate cancer: a comparative effectiveness evaluation
Article first published online: 13 AUG 2013
© 2013 The Authors. BJU International © 2013 BJU International
Volume 113, Issue 2, pages 200–208, February 2014
How to Cite
Sun, M., Sammon, J. D., Becker, A., Roghmann, F., Tian, Z., Kim, S. P., Larouche, A., Abdollah, F., Hu, J. C., Karakiewicz, P. I. and Trinh, Q.-D. (2014), Radical prostatectomy vs radiotherapy vs observation among older patients with clinically localized prostate cancer: a comparative effectiveness evaluation. BJU International, 113: 200–208. doi: 10.1111/bju.12321
- Issue published online: 13 JAN 2014
- Article first published online: 13 AUG 2013
- Accepted manuscript online: 20 JUN 2013 03:53AM EST
- prostate cancer;
- To compare efficacy between radical prostatectomy (RP), radiotherapy and observation with respect to overall survival (OS) in patients with clinically localized prostate cancer (PCa).
- Using data (1988–2005) from the Surveillance, Epidemiology, and End Results–Medicare linked database, 67 087 men with localized PCa were identified.
- The prevalence of the initial treatment strategy was quantified according to patients' life expectancy ([LE] <10 vs ≥10 years) at initial diagnosis and according to tumour stage. To reduce the unmeasured bias associated with treatment, we performed an instrumental variable analysis.
- Stratified (by stage and LE) Cox regression and competing-risks regression analyses were generated for the prediction of OS and cancer-specific mortality, respectively.
- Among patients with <10 years of LE, most were treated with radiotherapy (49%) or observation (47%). Among patients with ≥10 years of LE, most received radiotherapy (49%), followed by RP (26%).
- In men with <10 years of LE, RP and radiotherapy were not different with respect to OS (hazard ratio [HR]: 0.81, 95% confidence interval [CI]: 0.45–1.48, P = 0.499). Conversely, in men with ≥10 years of LE, RP was associated with an improved OS compared with observation (HR: 0.59, 95% CI: 0.49–0.71, P < 0.001) and radiotherapy (HR: 0.66, 95% CI: 0.56–0.79, P < 0.001).
- Similar results were recorded in competing-risks regression analyses.
- In patients with an estimated LE ≥10 years at initial diagnosis, RP was associated with improved survival compared with radiotherapy and observation, regardless of disease stage.
In 2012, nearly 242 000 new cases of prostate cancer (PCa) were diagnosed in the USA, and there were ∼28 000 new deaths from the disease . With the widespread adoption of PSA testing, >70% of newly diagnosed cases are now classified as clinically localized disease [1, 2]. Based on risk factors and expected patient survival at diagnosis, the primary options for initial therapy for such patients include active surveillance, radiotherapy or radical prostatectomy (RP) ; however, the contemporary management of clinically localized PCa remains characterized by controversies and uncertainties [4-6].
Prospective randomized phase III data showed that RP confers no survival benefit compared with observation [7-9], whilst retrospective data have reported a higher survival amongst RP-treated men relative to radiotherapy [10, 11]. To date, however, no randomized trial has been performed to rigorously evaluate the relative merits of RP and radiotherapy with regard to overall survival (OS). Whilst most physicians are now fully aware of the burden of over-diagnosis and over-treatment of PCa on healthcare systems, many remain apprehensive of the detrimental effects of failing to treat a potentially lethal disease.
In an attempt to reconcile such pressing matters, we sought to examine and compare cancer-specific mortality (CSM) and OS amongst men with low-risk clinically localized PCa treated with RP, radiotherapy or observation with regard to life expectancy (LE). Our hypothesis was that, in patients with an adequate LE at diagnosis, the probability of cure is more favourable when treated with surgical intervention.
Patients and Methods
Men aged ≥65 years diagnosed with non-metastatic PCa between 1992 and 2005, with both Medicare Parts A and B claims available and who were not enrolled in a health maintenance organization throughout the duration of the study, were identified. We then excluded those diagnosed at autopsy or death certificate only, those whose original or current reason for Medicare entitlement was listed as a disability, or those who had a Medicare status code including disability. Subsequently, patients with T3–4 PCa, high/anaplastic/unknown grade, unknown stage, aged ≥80 years at diagnosis were also excluded.
Radiotherapy and RP treatment within 6 months of PCa diagnosis were identified through Medicare claim files [14, 15]. Observation was defined as absence of active treatment codes (e.g. codes for RP, external beam radiation, radiation implants, brachytherapy, hormonal therapy, orchiectomy). Additionally, for each patient, age at diagnosis, race (white, black, other), population density (rural, urban), and marital status (married, unmarried, unknown) was assigned using the SEER data. The Charlson comorbidity index (CCI) was derived from the Medicare claims 1 year before PCa diagnosis using a commonly used and validated algorithm . Tumour grade corresponded to well-differentiated (Gleason scores 2–4) or moderately differentiated (Gleason scores 5–7) tumours. Clinical extension information was obtained from the SEER database, and classified as ≤ T1c, T2a/b or T2c. LE was estimated using patient age at diagnosis derived from the Social Security Administration tables (http://www.ssa.gov/OACT/STATS/table4c6.html) as per guideline recommendations , and dichotomized as <10 vs ≥10 years for the purpose of the study.
The primary endpoint of interest was OS. The secondary endpoint was CSM, which was classified as patients who died from PCa (International Classification of Diseases [ICD] codes: ICD-9 185 or ICD-10 C61). Those who died from any other cause were grouped under other-cause mortality (OCM). Data on cause-specific mortality were available up to the end of the year 2007.
To reduce the residual confounding resulting from unmeasured patient and/or other pertinent characteristics, we relied on instrumental variable analysis . This was primarily designed to account and balance both measured and unmeasured variables between treatment groups. The instrument variable used was the local area treatment pattern for RP  and radiotherapy . The instrument was created by grouping patients from the SEER-Medicare database according to hospital referral regions, as developed by the Dartmouth Atlas of Health Care . This was calculated as the proportion of patients who received RP in each health service area, and similarly for radiotherapy, as previously described . Before using this instrument, we assessed its validity by confirming that the intensity of RP or radiotherapy use according to health services area (Table 2) was highly correlated with receipt of RP (F-statistic ≥106) and radiotherapy (F-statistic ≥649), but was not associated with survival in multivariable models (all P > 0.05).
In the second step, the comparative effectiveness of RP vs radiotherapy vs observation was tested through the conventional Cox proportional hazards regression for the prediction of OS, and through competing-risks regression models for the prediction of CSM, after adjusting for OCM . Separate models were generated, after stratification according to LE (<10 vs ≥10 years), and tumour stage (≤T1c, T2a/b, and T2c). All statistical testing was two-sided with a P value of 0.05 considered to indicate statistical significance. Analyses were performed using the R software environment for statistical computing and graphics (Austria, Vienna, version 2.15.2).
Overall, 67 087 men with clinically localized PCa treated with either RP (23%), radiotherapy (50%), or observation (27%) were identified (Table 1). Treatment patterns differed according to LE estimates (Fig. 1). Specifically, among patients with <10 years of LE at diagnosis (n = 7706), the majority were treated with radiotherapy (49%) or observation (47%), and only 3.4% of patients underwent RP. Among patients with ≥10 years of LE (n = 59 381), most received radiotherapy (49%), followed by RP (26%), then observation (24%, P < 0.001).
|No. of patients (%)||33 613 (50.1)||15 532 (23.2)||17 942 (26.7)||–|
|Mean (median)||73 (73)||70 (69)||73 (74)|
|Race, n (%)||<0.001|
|White||29 476 (87.7)||13 666 (88.0)||14 918 (83.1)|
|Black||2 613 (7.8)||938 (6.0)||1 986 (11.1)|
|Other||1 524 (4.5)||928 (6.0)||1 038 (5.8)|
|CCI, n (%)||<0.001|
|0||14 771 (43.9)||12 289 (79.1)||12 202 (68.0)|
|1||10 033 (29.8)||2 229 (14.4)||3 394 (18.9)|
|2||4 886 (14.5)||443 (2.9)||823 (4.6)|
|≥3||3 923 (11.7)||571 (3.7)||1 523 (8.5)|
|Population density, n (%)||0.001|
|Urban||29 001 (86.3)||13 442 (86.5)||15 293 (85.2)|
|Rural||4 612 (13.7)||2 090 (13.5)||2 649 (14.8)|
|Marital status, n (%)||<0.001|
|Married||25 410 (75.6)||12 906 (83.1)||12 043 (67.1)|
|Unmarried||5 736 (17.1)||2 112 (13.6)||3 785 (21.1)|
|Unknown||2 467 (7.3)||514 (3.3)||2 114 (11.8)|
|Tumour stage, n (%)||<0.001|
|≤T1c||14 211 (42.3)||5 314 (34.2)||9 335 (52.0)|
|T2a/b||15 970 (47.5)||8 060 (51.9)||7 574 (42.2)|
|T2c||3 432 (10.2)||2 158 (13.9)||1 033 (5.8)|
|Tumour grade, n (%)||<0.001|
|Gleason score 2–4||2 522 (7.5)||1 023 (6.6)||3 667 (20.4)|
|Gleason score 5–7||31 091 (92.5)||14 509 (93.4)||14 275 (79.6)|
|Estimated LE*, n (%)||<0.001|
|<10 years||3 794 (11.3)||260 (1.7)||3 652 (20.4)|
|≥10 years||29 819 (88.7)||15 272 (98.3)||14 290 (79.6)|
In men with an estimated LE of <10 years at diagnosis (Fig. 2A–D), the 10-year OS rates were significantly better amongst RP- and radiotherapy-treated patients relative to their counterparts under observation, regardless of tumour stage (all log-rank P ≤ 0.004). No significant differences were recorded between radiotherapy and RP (all log-rank P ≥ 0.06). In men with an estimated LE of ≥10 years at diagnosis (Fig. 2E–H), the 10-year OS rates were significantly better amongst RP-treated patients relative to their radiotherapy and observation counterparts, irrespective of tumour stage (all log-rank P < 0.001). Similar survival trends were observed in cumulative-incidence CSM rates, after adjusting for OCM (Fig. 3A–H).
Two-Stage Residual Cox Regression Model
In all men with an estimated LE of <10 years (Fig. 4), no significant difference was recorded between RP and radiotherapy (hazard ratio [HR]: 0.81, 95% CI: 0.45–1.48, P = 0.499). In all men with an estimated LE of ≥10 years (Fig. 5), RP was associated with better OS compared with observation (HR: 0.59, 95% CI: 0.49–0.71, P < 0.001) and radiotherapy (HR: 0.66, 95% CI: 0.56–0.79, P < 0.001). The protective effect of RP on OS increased with advancing stage compared with observation (HR: 0.56 to 0.41, all P ≤ 0.009) and radiotherapy (HR: 0.72 to 0.54, all P ≥ 0.03).
Two-Stage Residual Competing-Risks Regression Model
In secondary analyses, we focused on CSM after adjusting for all other covariates, including OCM (Table 3). In men with <10 years of LE, RP was associated with a lower risk of CSM overall compared with observation (HR: 0.03, 95% CI: 0.02–0.42, P = 0.009) and radiotherapy (HR: 0.10, 95% CI: 0.03–0.75, P = 0.03).
In men with ≥10 years of LE, RP vs observation was associated with a lower risk of CSM (HR: 0.36, 95% CI: 0.19–0.69, P = 0.002), regardless of tumour stage, with the exception of patients with T2a/b PCa. Relative to radiotherapy, RP remained significantly associated with improved CSM overall (HR: 0.40, 95% CI: 0.24–0.68, P = 0.007), and across tumour stage, with the exception of patients with T2a/b PCa (Table 2).
|Characteristics||High RP use||Low RP use||P||High radiotherapy use||Low radiotherapy use||P|
|No. of patients||32 845||34 242||–||33 746||29 612||–|
|Mean (median)||72 (72)||72 (72)||72 (72)||72 (72)|
|Race, n (%)||<0.001||<0.001|
|White||29 398 (89.5)||28 662 (83.7)||28 448 (84.3)||29 612 (88.8)|
|Black||1 381 (4.2)||4 156 (12.1)||4 088 (12.1)||1 449 (4.3)|
|Other||2 066 (6.3)||1 424 (4.2)||1 210 (3.6)||2 280 (7.7)|
|CCI, n (%)||<0.001||<0.001|
|0||21 208 (64.6)||18 054 (57.2)||17 650 (52.3)||21 612 (64.8)|
|1||6 804 (20.7)||8 852 (25.9)||8 802 (26.1)||6 854 (20.6)|
|2||2 418 (7.4)||3 734 (10.9)||3 686 (10.9)||2 466 (7.4)|
|≥3||2 415 (7.4)||3 602 (10.5)||3 608 (10.7)||2 409 (7.2)|
|Population density, n (%)||0.005||<0.001|
|Urban||28 142 (85.7)||29 594 (86.4)||29 286 (86.8)||28 450 (85.3)|
|Rural||4 703 (14.3)||4 648 (13.6)||4 460 (13.2)||4 891 (14.7)|
|Marital status, n (%)||<0.001||<0.001|
|Married||25 762 (78.4)||24 597 (71.8)||24 455 (72.5)||25 904 (77.7)|
|Unmarried||5 233 (15.9)||6 400 (18.7)||6 165 (18.3)||5 468 (16.4)|
|Unknown||1 850 (5.6)||3 245 (9.5)||3 126 (9.3)||1 969 (5.9)|
|Tumour stage, n (%)||<0.001||<0.001|
|≤T1c||13 266 (40.4)||15 594 (45.5)||15 291 (45.3)||13 569 (40.7)|
|T2a/b||16 004 (48.7)||15 600 (45.6)||15 422 (45.7)||16 182 (48.5)|
|T2c||3 575 (10.9)||3 048 (8.9)||3 033 (9.0)||3 590 (10.8)|
|Tumour grade, n (%)||<0.001||<0.001|
|Gleason score 2–4||3 862 (11.8)||3 350 (9.8)||3 263 (9.7)||3 949 (11.8)|
|Gleason score 5–7||28 983 (88.2)||30 892 (90.2)||30 483 (90.3)||29 392 (88.2)|
|Estimated LE*, n (%)||<0.001||0.003|
|<10 years||3 609 (11.0)||4 097 (12.0)||4 001 (11.9)||3 705 (11.1)|
|≥10 years||29 236 (89.0)||30 145 (88.0)||29 745 (88.1)||29 636 (88.9)|
|RP vs radiotherapy||RP vs observation|
|HR (95% CI)||P||HR (95% CI)||P|
|<10 years of LE*|
|Overall||0.05 (0.03–0.75)||0.03||0.03 (0.02–0.42)||0.01|
|≤T1c||3.87 (0.13–117.87)||0.4||2.99 (0.10–86.80)||0.5|
|T2a/b||0.14 (0.01–1.59)||0.1||0.07 (0.02–0.83)||0.04|
|≥10 years of LE*|
|Overall||0.40 (0.24–0.68)||0.01||0.36 (0.19–0.69)||0.002|
|≤T1c||0.36 (0.15–0.87)||0.02||0.37 (0.14–0.97)||0.04|
|T2a/b||0.68 (0.30–1.54)||0.4||0.60 (0.21–1.73)||0.4|
|T2c||0.14 (0.03–0.60)||0.01||0.03 (0.003–0.20)||<0.001|
In spite of a 44% reduction in CSM between 1993 and 2009 , controversies surrounding the management of this disease persist [4-6]. On the one hand, randomized trials failed to detect any meaningful difference in OS between RP and observation [7-9], on the other hand, physicians remain unconvinced that observation may be considered a reliable option in patients with adverse disease characteristics, or in individuals with a long LE, given the uncertainty of how the disease could progress over time. In a parallel setting, radiotherapy represents the most widely used treatment method in contemporary men diagnosed with clinically localized PCa , yet, no clinical trial has been performed to rigorously compare the efficacy of RP with that of radiotherapy, nor it is likely that such a trial will be feasible in the future because of the large number of patients needed to be enrolled and the long-term follow-up required. In an attempt to examine these unresolved issues, patients, physicians and policy-makers must rely on observational data. For this reason, we sought to perform the first comparative effectiveness assessment between RP, radiotherapy and observation through a two-stage residual approach to instrumental variable analysis to attenuate bias within a large cohort of men with clinically localized PCa.
Our findings were as follows. Primarily, the survival benefit of RP compared with both radiotherapy and observation was consistently apparent given an estimated LE of ≥10 years at initial diagnosis. Specifically, the 10-year OS rate for RP was 79.9% overall, and ranged between 76.5 and 81.2%, vs ranges of 57.5–65.7% for radiotherapy and 51.3–57.5% for observation, across tumour stage categories. Similarly, CSM rates were distinctly in favour of RP (10-year CSM rates: 1.1–2.1%) compared with radiotherapy (3.5–5.8%) and observation (3.6–10.6%) in cumulative incidence analyses.
In multivariable Cox regression analyses, RP-treated men were 34 and 41% less likely to die relative to their radiotherapy and observation counterparts, respectively. After stratifying according to tumour stage, the survival benefit of RP vs radiotherapy progressively improved, with a 28, 40 and 46% lower risk of any death for ≤ T1c, T2a/b, and T2c stages, respectively. This relationship was also present when RP was compared with observation, with a 44, 33 and 59% lower risk for ≤ T1c, T2a/b, and T2c stages, respectively. Similar trends were subsequently recorded in competing-risks regression models, where men treated with RP were 60 and 64% less likely to die from PCa relative to radiotherapy and observation, respectively, in all men with a LE of ≥10 years.
Equally important were our findings that pertained to men with <10 years of LE. In this part, our analyses showed that, as expected, patients under observation had the least favourable OS rates (31–35%) after 10 years of follow-up compared with RP and radiotherapy (both 42–49%), as well as the highest CSM rates at 10 years (4.7–20.4%) compared with RP and radiotherapy (both 0–5.7%). In comparison, RP and radiotherapy-treated men fared similarly with respect to survival, both in unadjusted and adjusted analyses.
Several important clinical implications may be drawn from our findings. First, our results contradict the final results originating from randomized trials comparing RP and observation [7-9]. To understand the differences between these trials and our findings, it is important to discuss the differences in the populations studied. In the present study, it is likely that a significant proportion of patients treated with observation among Medicare beneficiaries were denied RP for medical reasons unrelated to the extent – and likelihood of cure – of their PCa, which may explain the high and variable CSM rates in this group regardless of LE estimates. This observation highlights the importance of using an instrumental variable approach to adjust for such unmeasured confounders. Nonetheless, it is unlikely that it will capture all biases that have not been accounted for. Moreover, there were probably several differences in the adherence and implementation of surveillance protocols in men on observation from the SEER–Medicare database compared with those of randomized trials. Nevertheless, several authors have argued against the design of previous randomized trials. For example, it has been argued that the most recent PIVOT study had inadequate statistical power, accrual difficulties, inadequate treatment adherence, treatment contamination and was closed prematurely [4-6]. That said, tremendous respect should be accorded to its authors for its initiative and randomized design. If the findings from the current study differ from that of randomized reports, it is not meant to dissuade physicians from considering observation, but to warrant caution in the liberal recommendation of observation for all clinically localized PCa, as the failure to identify and treat a lethal disease would ultimately harm the patient.
The present study may also be used to better describe the intricacies of radiotherapy use in comparison with RP. In this instance, for patients with a LE >10 years, our findings favour the use of RP because of its superior efficacy, both in cancer control and OS. Even when quality-of-life outcomes are prioritized, a previous landmark study showed few or no significant difference between the two groups . Furthermore, with the ongoing regionalization of care  and refinements of the RP technique , quality-of-life outcomes after surgery might have even further improved. To date, no randomized trial has directly assessed efficacy between the two treatment methods. Nonetheless, radiotherapy was used in nearly half of the patients in the current cohort, regardless of LE, despite retrospective data showing worse survival and cancer-control outcomes compared with RP [10, 11]. In that regard, it should be stated that contemporary techniques of radiotherapy have undergone considerable improvements and changes, which may also positively influence survival outcomes ; however, without a prospective randomized trial, there are few data to support the use of radiotherapy over RP in men with a long LE and/or more advanced disease stage.
The present study has some limitations. First, even after controlling for unmeasured confounders, both the Cox and competing-risks models varied slightly compared with the lack of adjustment for unmeasured confounders. This may be the case because there is little treatment selection bias in this retrospective cohort or because the instrumental variable analysis, no matter how rigorous its methodology, is unable to truly capture external factors that can confound the endpoints examined. Such limitations are inherent in all retrospective population-based studies. The reliance of stringent statistical methods in such reports are essential to reduce to a minimum the presence of such biases, keeping in mind that such methods cannot be considered equivalent to those of a randomized trial. Nonetheless, the study represents the first head-to-head comparison of RP, radiotherapy and observation to date. Further studies are encouraged to corroborate or refute the current findings. Second, the definition of patients in the observation group was based on the absence of active treatment codes. Consequently, we could not differentiate patients who actually underwent active surveillance, as opposed to observation. The former has been shown to represent a safe primary management strategy in patients with low-risk disease [6, 27]. Moreover, our analyses could not account for improvements in radiotherapy delivery, which improved over time, as well as the dosage of radiotherapy; therefore, it may be possible that a more refined definition of treatment types may radically change the findings. Third, some have argued that untreated patients with PCa are seemingly more likely to die from their cancer, while patients with PCa who receive treatment are more likely to die from other causes . In that context, Penson et al.  previously reported highly correlated cause of death information between death certificates and the SEER. Fourth, the current database did not contain sufficient information on PSA values, nor on the number of positive cores for the majority of individuals, which prevented us from stratifying patients according to such characteristics. Fifth, quality-of-life measures were not available. Such assessment may prove to be even more important than cancer-control outcomes in contemporary years . Sixth, LE estimates were calculated based on a period life table using the Social Security area population, which provides an average number of years of life remaining given a certain age. For men with PCa, the estimation of LE is critical for informed decision-making and treatment. In clinical practice, however, there are many additional variables that physicians may rely on to estimate a patient's LE, such as comorbid conditions and functional impairments . The current analysis relied solely on chronological age for prediction of LE. Certainly, this represents an averaged approximation for groups of men, and may not necessarily be very accurate for individuals. That said, estimates of LE were based on a conceptual framework, which, despite the considerable over- or underestimation of LE given a patient's health at diagnosis, can be considered a proxy for better individualized medical decisions. Finally, while the current cohort can be considered a large sample size, after stratification according to disease stage and LE, the number of patients in the group with a LE <10 years after diagnosis was low, as can be seen with the large CIs. Caution is therefore warranted in interpreting data from this subset of patients, with respect to the effect of treatment type on survival.
In conclusion, for patients with an estimated LE >10 years at initial diagnosis, RP was associated with improved OS and cancer-control outcomes compared with radiotherapy and observation, regardless of disease stage; however, such survival benefits are not consistently observable in patients with an estimated LE of <10 years.
Conflict of Interest
- 3National Comprehensive Cancer Network. Prostate Cancer. 2012 Dec 13; 1–69. Available from: http://www.nccn.org
- 11Comparison of mortality outcomes after radical prostatectomy versus radiotherapy in patients with localized prostate cancer: a population-based analysis. Int J Urol 2012; 19: 836–844; author reply 844–5. doi: 10.1111/j.1442-2042.2012.03052.x, , et al.
- 12Overview of the SEER-Medicare data: content, research applications, and generalizability to the United States elderly population. Med Care 2002; 40 (8 Suppl): IV-3–20118, , , , .
- 20A proportional hazards model for the subdistribution of a competing risk. J Am Stat Assoc 1999; 94: 496–509, .
- 21SEER Cancer Statistics Review, 1975–2009 (Vintage 2009 Populations). 2011-11-xx)[2012-04-18]. Bethesda, MD: National Cancer Institute, 2012. Available at: http://seer.cancer.gov/csr. Accessed February 2013, , et al.
Surveillance, Epidemiology, and End Results; LE, life expectancy