National long-term survival estimates after radical prostatectomy for prostate cancer

Abstract

Objective:

To examine survival and disease control outcomes, including metastasis-related survival outcomes, in a large contemporary cohort of patients undergoing radical prostatectomy for localized prostate cancer.

Methods:

We conducted a retrospective study of men with localized prostate cancer treated with radical prostatectomy from 2005–2015 with follow up through 2019 in the Veterans Health Administration. We defined biochemical recurrence (BCR) as a PSA ≥0.2 ng/mL. We used a validated natural language processing encoded dataset to identify incident metastatic prostate cancer. We estimated overall survival from time of surgery, time of BCR, and time of first metastasis using the Kaplan-Meier method. We then estimated time from surgery to BCR, BCR to metastatic disease, and prostate-cancer-specific survival from various time points using cumulative incidence considering competing risk of death.

Results:

Of 21,992 men undergoing radical prostatectomy, we identified 5,951 (27%) who developed BCR. Of men with BCR, 677 (11%) developed metastases. We estimated the 10-year cumulative incidence of BCR and metastases after BCR were 28% and 20%, respectively. Median overall survival after BCR was 14 years, with 10-year survival of 70%. From the time of metastasis, median overall survival approached 7 years, with 10-year overall survival of 34%. Prostate cancer-specific survival for the entire cohort at 10 years was 94%.

Conclusions:

In this large contemporary national cohort, survival for men with biochemically recurrent prostate cancer is longer than historical cohorts. When counseling patients and designing clinical studies, these updated estimates may serve as more reliable reflections of current outcomes.

Introduction

While radical prostatectomy provides prostate cancer control for most men, over one-third develop a detectable prostate-specific antigen (PSA) following surgery, i.e., biochemical recurrence, with few progressing to metastatic disease.^1–3 Optimal treatment plans and design of clinical trials for biochemical recurrence after radical prostatectomy depend on accurate estimations of the natural history of disease progression to metastases and death, as well as response to and timing of treatments including observation and androgen deprivation therapy.

However, many existing studies of prostate cancer recurrence after surgery are either outdated, limited in scope (e.g., single institution, clinical trial eligible population), or lack comprehensive data (e.g., metastasis-free survival). For example, a landmark study of biochemically-recurrent prostate cancer describing a single institutional experience of 304 patients with biochemical recurrence after surgery in the 1980s-1990s is commonly referenced as a benchmark for prostate cancer survival following recurrence, but the characteristics of this small cohort do not reflect contemporary practice.⁴ More recent studies primarily use data from randomized clinical trials, including seminal clinical trials comparing radical prostatectomy to observation (PIVOT⁵) or active surveillance (ProtecT⁶). While more recent and comprehensive than historical cohort studies, these results reflect selected clinical trial cohorts that may not be representative of the general populations of men undergoing surgery.^7,8 While these trials have long-term follow-up available, the characteristics of patients may not reflect current practice. On the other hand, though large-scale databases such as the National Cancer Database have been used to reflect more recent and representative patient populations, they often lack data on development of metastatic disease and cause of death, limiting the ability to estimate progression-free, conditional, and disease-specific survival.⁹ Further, the trade-offs of granular data from single institutional experiences may compromise generalizability to national populations.¹⁰ Additionally, accurate patient counseling can rely on conditional estimates that may not be available in existing analyses. For example, the ProtecT trial directly informs decision making regarding active surveillance, prostatectomy, and radiation therapy, but does not provide estimates of mortality or metastasis for men with recurrent disease after prostatectomy (i.e., biochemical recurrence). Even if these types of estimates could be completed, the absolute number of biochemically recurrent cases within randomized trials is low (i.e., 58 surgical patients in ProtecT with any clinical progression¹¹, 89 surgical patients in PIVOT with asymptomatic progression (i.e., biochemical recurrence)¹²), limiting inference given low overall rates for the post-surgical group with biochemical recurrence at risk for events such as metastasis or death.

For these reasons, we studied a contemporary national cohort of men undergoing radical prostatectomy for localized prostate cancer. We used a combination of cancer registry, electronic laboratory, pharmacy, radiology, and pathology data, as well as a validated natural language processing algorithm to identify development of prostate cancer metastases, allowing us to provide updated estimates of the natural history of prostate cancer after radical prostatectomy.¹³ Herein, we present contemporary survival estimates in what is to our knowledge the largest series of radical prostatectomy cases with comprehensive biochemical recurrence, metastasis, and survival data. These findings add to existing knowledge and may influence the design of clinical trials, particularly in an era of evolving imaging techniques (i.e., PSMA PET) and salvage radiation, hormone, and systemic therapy contexts, especially given the favorable disease-specific survival outcomes.

Methods

We accessed electronic health records for men undergoing radical prostatectomy within the Veterans Health Administration (VHA) from 2005–2015. We extracted demographic, clinical, laboratory, and surgical data from the VHA databases and uniquely enriched the dataset with date of prostate cancer metastases using a recently validated and published natural language processing algorithm with reported sensitivity and specificity of 0.92 and 0.98, respectively.¹³ We combined these data with death records including cause of death from the National Death Index. We excluded patients who received radiation or hormonal therapy within a year of surgery, had metastatic disease prior to surgery, or had clinical T4 disease. To allow for estimates of cause-specific survival, we additionally excluded patients with missing cause of death (n=50).

We defined biochemical recurrence (BCR) as a patient with an initially undetectable post-operative PSA that rose to at least 0.2 ng/mL. Incident metastasis and date of detection of metastatic disease was extracted from a dataset encoded by a previously published and validated algorithm.¹³ Cause of death was accessed from the National Death Index and categorized into prostate cancer or other-cause death. Censoring dates for all endpoints were set as December 31, 2019, if there was no date of death recorded, as National Death Index data were only available to this date.

The primary outcomes were cumulative BCR incidence, metastasis, prostate-cancer-specific death, and overall death. We estimated overall survival using the Kaplan-Meier method, cumulative incidence for prostate cancer death with other-cause death as a competing risk, and cumulative incidence of other outcomes with any-cause death as a competing risk. In the overall dataset, survival time was measured from time of radical prostatectomy to the event, death, or censoring date (December 31, 2019). Metastasis-free survival was defined as time to incident metastatic disease or death.

We estimated conditional survival by performing additional analyses in data subsets. Within the BCR cohort, survival time was measured from the date of BCR to the event, death, or censoring date. The metastatic cohort was defined as patients who developed BCR, then metastases. Within the metastatic cohort, survival time was measured from time of first development of metastatic disease to the event, death, or censoring date.

We additionally performed time-to-event analyses in groups stratified by AUA clinical risk. Low-risk patients had clinical T1c or T2a disease with PSA < 10 ng/mL at time of diagnosis and Gleason 6 on biopsy. Intermediate-risk patients had clinical T2b-c disease, or PSA 10–20 ng/mL at time of diagnosis, or Gleason 7 on biopsy. High-risk patients had clinical T3 disease, or PSA > 20 ng/mL at time of diagnosis, or Gleason 8–10 on biopsy.

We performed sensitivity analyses to estimate survival stratified by age and to account for variable definitions of cause of death (Appendix).

This study was approved by the Veterans Affairs and University of Michigan Institutional Review Board (IRB).

Results

Of 24,993 men undergoing radical prostatectomy from 2005–2015, 21,992 met inclusion criteria (supplementary figure). As shown in Table 1, men were on average in their early 60s at diagnosis, most had cT1c disease (71%), with PSA 4–10 ng/mL at diagnosis (69%), and had Gleason 7 disease on biopsy (54%). At time of surgery, 18% of men had positive surgical margins, and 0.6% had positive lymph nodes. In the BCR and metastatic disease subgroups, cancer characteristics were more aggressive.

Table 1.

Characteristics of people undergoing radical prostatectomy

Variable	All radical prostatectomy patients (n=21,992)	Patients with biochemical recurrence (n=5,951)	Patients with biochemical recurrence and metastasis (n=677)
Clinical T stage (n,%)
cT1c	15,634 (71.1)	3,984 (66.9)	414 (61.2)
cT2	1,055 (4.8)	328 (5.5)	41 (6.1)
cT2a	2,408 (10.9)	656 (11.0)	70 (10.3)
cT2b	746 (3.4)	270 (4.5)	41 (6.1)
cT2c	1,997 (9.1)	636 (10.7)	95 (14.0)
cT3	45 (0.2)	19 (0.3)	5 (0.7)
cT3a	107 (0.5)	58 (1.0)	11 (1.6)
Pathologic T stage (n,%)
pT0	18 (0.1)	3 (0.1)	0
pT2	15,621 (71.0)	3,382 (56.8)	322 (47.6)
pT3	3,564 (16.2)	1,451 (24.4)	263 (38.8)
pT4	31 (0.1)	15 (0.3)	3 (0.4)
pTx	2,758 (12.5)	1,100 (18.5)	89 (13.1)
Pathologic N stage (n,%)
Node negative	16,302 (74.1)	4,245 (71.3)	504 (74.4)
Node positive	125 (0.6)	86 (1.4)	48 (7.1)
Unknown nodal status	5,565 (25.3)	1,620 (27.2)	125 (18.5)
Age (years, mean (standard deviation))	61.7 (6.1)	62.2 (6.5)	61.9 (6.3)
PSA at diagnosis (n,%)
<= 4 ng/mL	3,323 (15.1)	743 (12.6)	74 (10.9)
4–10 ng/mL	15,209 (69.2)	3,855 (65.4)	409 (60.4)
10–20	2,396 (10.9)	952 (16.1)	141 (20.8)
>= 20 ng/mL	916 (4.2)	345 (5.9)	43 (6.4)
Unknown	148 (0.7)	56 (0.9)	10 (1.5)
Biopsy Gleason Category (n,%)
<= 6	7249 (33.0)	1,505 (25.3)	82 (12.1)
7	11,952 (54.3)	3,303 (55.5)	353 (52.1)
8–10	2,684 (12.2)	1,104 (18.6)	237 (35.0)
Unknown	107 (0.5)	39 (0.7)	5 (0.7)
Surgical Gleason grade group (n,%)
<GG1	26 (0.1)	6 (0.1)	0
GG1	1,819 (8.3)	219 (3.7)	14 (2.1)
GG2	5,350 (24.3)	1,003 (16.9)	83 (12.3)
GG3	1,862 (8.5)	566 (9.5)	74 (10.9)
GG4	573 (2.6)	204 (3.4)	32 (4.7)
GG5	459 (2.1)	223 (3.7)	58 (8.6)
Unknown	11,903 (54.1)	3,730 (62.7)	416 (61.4)

PSA, prostate-specific antigen.

Of all 21,992 men undergoing radical prostatectomy for localized disease, 3,267 men died, with estimated overall survival at 5 years of 95.8% and at 10 years of 84.8%. A total of 817 men developed metastases, with estimated cumulative incidence of metastases of 1.8% and 4.4% at 5 and 10 years, respectively, and metastasis-free survival estimates of 94.4% and 82.2% at 5 and 10 years, respectively. Prostate cancer-specific survival was 99.7% and 98.7% at 5 and 10 years, respectively, with 285 men dying from prostate cancer.

A total of 5,951 men developed BCR. The estimated cumulative incidence of BCR was 20.5% at 5 years and 28.4% at 10 years (Figure 1). Of men who developed BCR, 1,140 died. Median overall survival was estimated as 14.3 years from the time of BCR, with 5-year survival of 88.3% and 10-year survival of 69.7%. Prostate-cancer specific survival from time of BCR was 98.0% at 5 years and 94.4% at 10 years, with 224 deaths from prostate cancer.

Figure 1:

Cumulative incidence of biochemical recurrence, considering the competing risk of death

Black: biochemical recurrence

Grey: death

Of the 5,951 men with BCR, 677 developed metastases. The median time to metastatic disease was not reached. Estimated cumulative incidence of metastases was 10.0% at 5 years and 19.6% at 10 years after surgery, with metastasis-free survival estimates of 81.2% at 5 years and 60.0% at 10 years (Figure 2). Of men who developed metastases, 235 died. Median overall survival from time of development of metastatic disease was 6.6 years, with 5-year estimated survival of 55.6% and 10-year estimated survival of 34.2% (Supplemental Figure 2).

Figure 2:

Cumulative incidence of metastasis after biochemical recurrence, considering the competing risk of death

Black: death

Grey: metastasis

When stratified by AUA clinical risk group, BCR was much more frequent in high-risk than low-risk patients (42.2% vs. 19.5%; Table 2). The 10-year incidence of metastasis was also much higher in high-risk than low-risk patients (10.9% vs. 1.6%), and once metastases developed the median time to death was shorter in high risk than low risk patients (5.5 years vs. median unreached). Estimated 10-year overall survival ranged from 87.1% for low risk to 81.5% for high-risk patients.

Table 2:

Outcomes after radical prostatectomy stratified by AUA clinical risk group

Endpoint	Whole Surgical Cohort (n=21,992)	Low Risk (n=5,361)*	Intermediate Risk (n=12,536)*	High Risk (n=3,562)*
Overall Survival
5-year	95.8%	96.6%	95.9%	94.8%
10-year	84.8%	87.1%	84.8%	81.5%
Prostate-cancer Specific Survival
5-year	99.7%	99.8%	99.8%	99.2%
10-year	98.7%	99.4%	98.9%	96.8%
Cumulative Incidence of Biochemical Recurrence
5-year	20.5%	14.3%	19.9%	32.0%
10-year	28.4%	19.5%	28.7%	42.2%
Cumulative Incidence of Metastases
5-year	1.8%	0.7%	1.4%	4.7%
10-year	4.4%	1.6%	4.0%	10.9%
Time from metastasis to death
Number of metastatic cases	817	86	409	308
Median years from metastasis to death	7.7	Not reached	8.9	5.5
5-year survival from time of metastasis	57.4%	57.8%**	62.3%	52.7%
10-year survival from time of metastasis	40.4%	57.8%**	46.1%	29.3%

^*Only cases without missing staging data were included in sub-groups

^**Number at risk of death with metastases from low-risk group was 14 at 5 years and 2 at 10 years

Discussion

In this large national contemporary cohort of men undergoing radical prostatectomy for localized prostate cancer, we estimated long-term outcomes for BCR, metastasis, prostate cancer-specific and overall 10-year survival. We found the cumulative incidence of BCR after radical prostatectomy was 28%. Of men who developed BCR, we estimated roughly 1 in 5 men would develop metastatic disease within 10 years. After developing metastatic disease, we estimated median overall survival of more than 6 years.

Our results support the long-term results from the major clinical trials for prostate cancer that included surgical arms. In comparison to the long-term results from the ProtecT trial^6,11, our study demonstrated a slightly higher incidence of metastatic disease, with similar overall prostate cancer-specific mortality. The 10- and 15-year incidence of metastatic disease for the patients treated with prostatectomy on ProtecT were 2.4% and 4.7%, respectively, compared with a 10-year rate of 4.4% observed in the entire surgical cohort followed in our study. The BCR rate was also lower in the ProtecT surgical arm (10.5% at 15 years) relative to our cohort (28% at 10 years). However, patients enrolled in ProtecT much more commonly had lower risk disease than our cohort (only 23% of patients enrolled on ProtecT had Gleason ≥7 disease, compared to 67% in our study), likely accounting for these differences. Rates of metastasis in the PIVOT surgical arm were similar to our study (4.7% at 12 years vs. 4.4% at 10 years in our study), as were BCR rates (24.4% at 12 years in PIVOT versus 28% at 10 years in our study).^5,12,14 However, there was a higher rate of death from any cause in PIVOT (40.9% at 12 years vs. 15% at 10 years in our cohort). This may be due to younger age in our cohort relative to PIVOT (mean age 67 years in PIVOT vs. 62 years in our cohort).

Notably, our series is much larger than existing studies, particularly with respect to men postsurgery with BCR. Our cohort is roughly 40 times larger than the largest surgery arm from a randomized trial (PIVOT), with nearly 6,000 men postsurgery experiencing BCR, compared to the 58–89 men with progression or BCR from the surgery arms of existing clinical trials.^11,14

Our estimates of BCR are consistent with prior reports. We estimated 28% BCR at 10 years, comparable to prior reported rates of 18–50%.² As expected, rates of BCR were higher for AUA/NCCN intermediate- and high-risk groups relative to the low-risk group. Taken together, we found comparable rates of BCR after radical prostatectomy, but improved prostate cancer-specific mortality.

These estimates suggest a better prognosis for contemporary patients than in previous series. For example, a potentially outdated landmark study of the natural history of prostate cancer reported half of men with BCR developing metastases within 8 years, much higher than our estimated 1 in 5 men developing metastases within 10 years.⁴ Once metastases developed, we estimated a median survival of 6.6 years compared to the previously reported median survival of 5 years.⁴ With respect to prostate cancer-specific mortality, our estimates are also higher than prior series, with 10-year prostate cancer-specific survival of 94.4% in our series compared to 73% in the prior landmark study.¹⁵ Other series estimate 10-year prostate cancer-specific survival of 77–99%, varying by risk factors.^1,16–18

It is possible that advances in treatment for advanced disease explain the observed improved survival over the last 20 years. Over this time period, multiple systemic treatment options became available, beginning with docetaxel for metastatic castration resistant prostate cancer in 2004 and followed more recently by treatments in the castration-resistant and hormone-sensitive contexts such as abiraterone, apalutamide, and enzalutamide.¹⁹ Given the success of these agents in randomized trials, it is likely that at least part of the observed improved survival in recent years is due to more advanced treatments for recurrent and advanced disease. However, further work is needed to understand how and when to apply these treatments to optimally improve patient survival while minimizing toxicity of treatments.

There are limitations to our study. First, it is possible that a VA cohort may be different from non-VA patients. However, BCR rates in our cohort were similar to a CaPSURE study (28% vs 30%, respectively), comprising patients followed in multiple practice types.²⁰ Prostate cancer-specific survival was higher in our study than CaPSURE (94% vs 84% at 10 years), though many patients included in the CaPSURE study were treated before the approval and dissemination of advanced therapies. Next, we did not adjust for baseline or pathologic disease characteristics or receipt of subsequent treatments after prostatectomy, except excluding treatments received in the perioperative setting. Our objective was to update estimates of prognosis in a contemporary cohort, reflecting expected prognosis for “all-comers” after radical prostatectomy. Increasing complexity in the selection, timing, and sequence of treatments after BCR coupled with advanced imaging techniques necessitates further study for accurate survival estimates. However, we did estimate survival outcomes stratified by age as a sensitivity analysis (Appendix), with an expected decrease in overall survival in older subgroups more typical of the Medicare population. We were unable to determine pathologic Gleason Grade Groups for Gleason 7 disease on biopsy as primary and secondary Gleason score were unavailable, similar to the PIVOT trial. While we were able to expand on clinical trial data by reporting on post-surgical Gleason grade group, these data were unknown for about half of all patients in our cohort. This may limit understanding of the specific risk level of this cohort, similar to a limitation of the PIVOT trial. Additionally, we relied on data reported from individual VHA centers, cause of death from the National Death Index, and metastatic disease data from the previously published algorithm, all of which may contain some inaccuracies.¹³ However, the previously reported accuracy of the algorithm had both high sensitivity and specificity, and we performed sensitivity analyses to estimate rates of prostate cancer-specific mortality that may be misattributed in cause of death by the National Death Index, with changes in estimates of less than 2 percentage points (Appendix). Lastly, given the timeframe of our study our outcomes estimations are limited to 10 years, as opposed to the 15–29 year data available for clinical trials of prostatectomy patients. However, these trials reflected highly selected populations eligible for surveillance or observation, while our study included all patients undergoing prostatectomy in a contemporary timeframe, many of whom had worse disease characteristics as noted above. Additionally, the number of patients undergoing prostatectomy and with BCR after prostatectomy was roughly 30-fold higher in our study compared to prior trials and series adding robustness to our outcome estimates.

Despite these limitations, our study is a robust update to the expected course of prostate cancer after radical prostatectomy, BCR, and metastatic disease. When counseling men undergoing radical prostatectomy, and in designing clinical studies and trials, the longer survival course relative to prior studies should be taken into consideration.

Conclusions

In a nationwide contemporary cohort of people undergoing radical prostatectomy, we estimated rates of BCR, metastatic disease, and survival. Survival at all time points has increased relative to prior studies, and should be considered in the design of clinical studies and trials and in counseling men undergoing radical prostatectomy.

Supplementary Material

1

Supplementary Figure: Cohort Selection