Use of androgen deprivation therapy as salvage treatment after primary therapy for clinically localized prostate cancer

Abstract

Purpose

The optimal use of androgen deprivation therapy as salvage treatment (sADT) for men after initial prostatectomy or radiotherapy for clinically localized prostate cancer is undefined. We describe patterns of sADT use and investigate clinical and sociodemographic characteristics of insured men who received sADT versus surveillance in managed care settings.

Methods

Using comprehensive electronic health records and cancer registry data from three integrated health plans, we identified all men with newly diagnosed clinically localized prostate cancer between 1995 and 2009 who received either prostatectomy (n = 16,445) or radiotherapy (n = 19,531) as their primary therapy. We defined sADT based on the timing of ADT following primary therapy and stage of cancer. We fit Cox proportional hazard models to identify sociodemographic characteristics and clinical factors associated with sADT.

Results

With a median follow-up of 6 years (range 2–15 years), 13 % of men who underwent primary prostatectomy or radiotherapy received sADT. After adjusting for selected covariates, sADT was more likely to be used in men who were older (e.g., HR 1.70, 95 % CI 1.48–1.96 or HR 1.33, 95 % CI 1.17–1.52 for age 70+ relative to age 35–59 for primary prostatectomy or radiotherapy, respectively), were African-American, had a short PSA doubling time, had a higher pre-treatment risk of progression, had more comorbidities, and received adjuvant ADT for initial disease.

Conclusions

In men with localized prostate cancer in community practice initially treated with prostatectomy or radiotherapy, sADT after primary treatment was more frequent for men at greater risk of death from prostate cancer, consistent with practice guidelines.

Introduction

Prostate cancer is the most common non-skin cancer in men in the USA, with an estimated 220,800 new cases diagnosed in 2015 [1], and approximately 2.5 million prostate cancer survivors are currently alive [2]. Androgen deprivation therapy (ADT) has been the standard treatment for metastatic prostate cancer for decades [3]. It has also been used as primary therapy for localized disease and has been increasingly used as adjuvant or salvage therapy among men diagnosed with localized disease after prior surgery or radiotherapy [4]. Clinical evidence suggests that ADT increases survival for the use of: (1) long-term neo-adjuvant or adjuvant ADT in combination with radiotherapy for high-risk disease [5]; (2) short-term ADT in combination with radiotherapy for intermediate-risk disease [6]; and (3) adjuvant therapy after surgery for node-positive disease [7]. Following either surgery or radiotherapy, approximately 20 % of men experience biochemical recurrence within 10 years, as indicated by a rising serum prostate-specific antigen (PSA) level [8], and ADT is frequently prescribed as salvage treatment for men with a rising PSA following surgery or radiotherapy for clinically localized disease [9]. Nevertheless, uncertainty exists about the optimal use and timing of salvage ADT (sADT) for men experiencing only a biochemical recurrence following primary therapies.

Among men who received prostatectomy (surgery) and experienced biochemical recurrence, salvage radiotherapy with ADT appears to be effective in prolonging progression-free survival, especially for high-risk patients [10]. A single-center study indicated that sADT alone after primary prostatectomy or radiotherapy has similar survival benefits as primary ADT [11]. Decisions about sADT must weigh the potential for survival benefit with the risks of treatment with ADT. For example, the use of ADT as a primary monotherapy is associated with adverse events including cardiovascular disease and diabetes mellitus [12, 13]. Side effects of ADT also include osteoporosis/fractures, obesity, sarcopenia, and psychosocial issues [14, 15].

The median time to distant metastasis is 8 years for men with a biochemical recurrence after prostatectomy and with an additional 5 years for median time to death [16, 17]. Given the prolonged natural history of localized prostate cancer, the high median age of diagnosis for prostate cancer, and the large percentage of men who are diagnosed with indolent disease, sADT use for serologic relapse might result in greater morbidity without possibly prolonging cause-specific survival [18]. Little is known about sADT use in community practice settings (including both academic and non-academic settings). We investigated the factors associated with sADT use after primary therapy with prostatectomy or radiotherapy in a large community-based cohort of men.

Materials and methods

Data source

We conducted a retrospective observational cohort study of men with newly diagnosed localized prostate cancer who were members of one of three integrated health systems within the HMO research network (HMORN)—Kaiser Permanente Northern California, Kaiser Permanente Southern California, and the Henry Ford Health System in Detroit, Michigan. These three health systems possess integrated electronic health records that incorporate comprehensive information from all inpatient and outpatient clinical encounters, laboratory test values (including PSA values), pharmacy data, and linkages to local SEER (Surveillance, Epidemiology, and End Results) cancer registry data.

Cohort selection

We identified all men aged ≥35 years diagnosed with localized prostate cancer (T1–T2, N0, M0) between 1995 and 2009 (n = 51,511; Fig. 1). We included men who received primary therapy with prostatectomy or radiotherapy within 12 months after their initial prostate cancer diagnosis. Radiotherapy includes external beam radiotherapy and brachytherapy. We excluded men who received primary ADT, orchiectomy, or chemotherapy as part of initial therapy since these therapies indicate the presence of metastatic disease, leaving n = 38,830. Then, we excluded men with less than 2 years of follow-up after diagnosis (n = 2854) to ensure that sADT exposure was captured. The final cohort included 35,976 men with localized prostate cancer treated with primary prostatectomy (n = 16,445) or radiotherapy (n = 19,531). These patients were observed through December 31, 2010 (median follow-up = 6.0 years, range 2–15 years).

Fig. 1.

Study cohort. sADT salvage androgen deprivation therapy, RT radiotherapy, and RP radical prostatectomy

We defined ADT use as either a gonadotropin-releasing hormone (GnRH) analog or GnRH antagonists. Due to possible incomplete PSA data in some health plans, adjuvant ADT and sADT were defined in the entire cohort based on the timing relative to primary therapy. We defined adjuvant ADT as ADT given within 6 months following primary prostatectomy for men with pathologically regional summary stage or within 6 months following the last date of primary radiotherapy. We defined sADT as either: (1) ADT after primary prostatectomy for men with pathologically localized summary stage; (2) initiation of ADT after a period of 6 months or more following primary prostatectomy for men with pathologically regional summary stage, or after 6 months or more following the last dose of primary radiotherapy; or (3) ADT after 12 or more months following the last date of adjuvant ADT. We required at least 12 months of an ADT-free period in this latter definition to reduce the possibility of misclassifying adjuvant ADT as sADT.

Baseline characteristics

We obtained from the tumor registries and electronic health records the following variables: age at diagnosis, racial/ethnic group (non-Hispanic white, African-American, Hispanic, other), year of diagnosis, and diagnosis of prior or subsequent primary cancers other than prostate cancer. Clinical baseline characteristics included serum PSA, Gleason score, and clinical stage. We defined baseline PSA value (in ng/mL) using electronic laboratory values from the date closest to the date of cancer diagnosis. We obtained the two-value summed Gleason score from the first biopsy leading to the prostate cancer diagnosis. The American Urological Association (AUA) risk group at baseline was assessed based on PSA, Gleason score, and clinical T stage [19], with low risk defined as PSA ≤10, Gleason score ≤6, and stages T1c–T2a; intermediate risk defined as PSA 11–20 or Gleason score = 7 or stage T2b; and high risk defined as PSA >20 or Gleason score ≥8 or stages T2c– T3a. We computed the Elixhauser comorbidity index by assessing the presence of 30 individual health conditions diagnosed between 2 years before prostate cancer diagnosis and 3 months after diagnosis [20]. We counted conditions with an inpatient diagnosis or at least two outpatient diagnoses 30 days apart to minimize counting of rule-out diagnoses.

Factors following primary therapy

We selected serum PSA values after primary therapy until receipt of sADT, or to the end of follow-up for men without sADT. To explore PSA rise as a proxy for recurrence/progression, we estimated both the highest PSA observed and the PSA doubling time based on available PSA data. Details about their classifications are included in “Appendix.” If men had primary prostatectomy, we assessed their subsequent receipt of salvage radiotherapy if this was given. We also included receipt of adjuvant ADT within analysis.

Statistical analysis

Patient characteristics were stratified by primary therapy (prostatectomy or radiotherapy) and use of sADT. Bivariate analyses were performed using Chi-square tests for categorical variables and t tests for continuous variables. Cochran–Armitage trend test was utilized for sADT trend analysis. We fit Cox proportional hazard models to quantify time to receipt of sADT as a function of patient characteristics [both baseline and post-primary therapy factors (including diagnosis year for potential confounding bias due to technology innovations during the study period)]. The follow-up started with date of primary surgery (or end date of primary radiotherapy) and ended with either the use of sADT or censoring due to death or end of follow-up. We first constructed a model including the post-primary therapy PSA doubling time [21], which is more often utilized to make clinical judgments [22]. Due to the strong effects from post-primary therapy factors on sADT, as a sensitivity analysis, we constructed another model without postprimary therapy factors and with the risk group variable replaced by baseline PSA, Gleason score, and clinical stage to investigate effects of only baseline characteristics. We describe methods for handling missing data in “Appendix.”

Results

The mean (±SD) age of the study cohort was 60.7 (±7.2) and 67.1 (±7.4) with median 61 and 68 years for men who received primary prostatectomy and radiotherapy, respectively. Overall, 12.8 % of patients received sADT after a median 6.0 years of follow-up. In the prostatectomy group, 2107 men (12.8 %) received sADT, while in the radiotherapy group, 2511 men (12.9 %) received sADT.

The use of sADT increases by age (Fig. 2). African-American men had the highest rate of sADT use. The use of sADT increased by AUA risk group among men with primary radiotherapy (p < 0.001 for trend analysis), whereas no obvious trend was found among men treated with prostatectomy. We found a statistically significant difference (p < 0.01) in sADT use according to PSA doubling time, with higher proportions of men having sADT for the prostatectomy group than the radiotherapy group among those with shorter doubling times. Men who received sADT were older; had a higher initial Gleason score, higher baseline PSA, higher post-primary therapy PSA, shorter PSA doubling time; and were more likely to have had bone scans and adjuvant ADT before sADT than men who did not receive sADT, in both primary therapy groups (Table 1). The use of adjuvant ADT was more common among men treated with initial radiotherapy than with surgery.

Fig. 2.

Salvage androgen deprivation therapy (sADT) utilization by key variables, stratified by primary prostatectomy or radiotherapy. *p < 0.01 from Chi-square tests between two proportions

Table 1. Bivariate analysis of factors associated with salvage ADT up to 15 years after primary prostatectomy or radiotherapy among men with newly diagnosed localized prostate cancer.

Variable	Categories	Prostatectomy (%)			Radiotherapy (%)
		No sADT	sADT	p value	No sADT	sADT	p value
N		14,338	2107		17,020	2511
Baseline characteristics
Age at diagnosis (years)	35–49	6.8	5.6	<0.001	1.3	0.6	<0.001
	50–59	37.5	31.7		15.9	11.4
	60–64	25.1	25.9		18.5	15.3
	65–69	22.2	22.6		25.1	26.5
	70–74	6.5	9.6		24.1	27.9
	75+	1.8	4.7		15.1	18.2
Race	White	60.8	57.2	<0.001	67.6	64.9	0.002
	Black	16.9	21.7		14.1	16.6
	Hispanic	14.6	14.8		10.4	11.2
	All others	7.7	6.3		8.0	7.4
Gleason score	<6	60.1	48.7	<0.001	60.2	41.5	<0.001
	7	34.5	36.6		30.6	38.4
	8	3.8	8.6		6.2	12.0
	9–10	1.6	6.0		3.0	8.1
PSA (ng/mL)	≤4	28.2	24.9	<0.001	13.0	9.4	<0.001
	4 < PSA ≤ 10	57.3	48.9		60.6	46.6
	10 < PSA ≤ 20	11.1	16.2		18.5	24.4
	PSA > 20	3.4	10.1		8.0	19.6
T stage	≤T2A	48.3	57.9	<0.001	68.0	58.6	<0.001
	T2B	7.9	13.8		8.3	18.6
	T2C+	43.8	28.3		23.7	22.8
Risk group	High	41.3	38.7	0.001	27.6	40.3	<0.001
	Intermediate	31.5	35.5		37.8	41.3
	Low	27.2	25.7		34.6	18.4
Tumor sequence	Single primary	89.2	85.0	<0.001	85.0	81.7	<0.001
	1st of multiple	5.6	9.6		8.3	11.7
	2nd or more	5.2	5.4		6.6	6.6
Elixhauser score	0	41.4	41.9	0.887	31.8	34.5	0.020
	1	29.7	29.2		28.9	28.7
	2	16.2	15.8		18.3	17.9
	3+	12.7	13.1		21.0	18.9
Characteristics after primary therapy
PSA category	At least 1 PSA ≥10.0 ng/mL	0.8	9.0	<0.001	2.9	27.3	<0.001
	0.2 (or 2.0+ nadir for RT)–10.0 ng/mL	12.3	32.5		7.3	27.6
	All PSA <0.2 ng/mL (or 2.0 +nadir for RT)	64.3	14.3		65.9	27.8
	No PSA data in relevant interval	22.7	44.3		23.8	17.2
PSA doubling time category (in months)	0–<9.0	1.4	18.1	<0.001	6.0	28.4	<0.001
	9.0–<15.0	1.5	7.2		6.6	16.2
	15.0–<36.0	3.8	7.3		18.9	20.5
	36.0+ (no rise)	66.3	15.8		43.0	15.4
	Unknown	27.0	51.6		25.4	19.5
Salvage radiotherapy	Yes	6.7	19.8	<0.001	–	–	–
Adjuvant ADT	Yes	0.7	2.3	<0.001	26.0	35.1	<0.001

Median follow-up of 6 years

sADT salvage androgen deprivation therapy, PSA prostate-specific antigen, RT radiotherapy

Results from the multivariable Cox models assessing time to sADT are presented in Table 2. In the prostatectomy group, we found that older men were more likely to receive sADT [e.g., hazard ratio (HR) 1.70, 95 % CI 1.48–1.96 for men diagnosed at ages 70+ than those diagnosed at ages 35–59]. African-Americans were more likely to receive sADT than whites (HR 1.23, 95 % CI 1.10–1.37). Men with intermediate (HR 0.77, 95 % CI 0.70–0.86)- or low-risk disease (HR 0.74, 95 % CI 0.65–0.83) were less likely to receive sADT than high-risk men. A shorter PSA doubling time was strongly associated with sADT (e.g., HR 28.4, 95 % CI 24.3–33.3 when PSA doubling time is 0–<9.0 months) relative to men without any evidence of a rising PSA. Men with prostate tumor as their first of multiple primaries were more likely to receive sADT than men with prostate tumor as their only primary cancer (HR 1.41, 95 % CI 1.21–1.64). Comorbidity, salvage radiotherapy, and adjuvant ADT use were also independently associated with the use of sADT after adjusting for other factors. Among men receiving primary radiotherapy, the findings were similar except for tumor sequence. When only baseline characteristics were considered as a sensitivity analysis, men with greater Gleason score, higher baseline serum PSA, or higher T stage were more likely to receive sADT, after adjusting for other characteristics (Table 3).

Table 2. Cox model of time to salvage ADT use after primary prostatectomy or radiotherapy among men with newly diagnosed localized prostate cancer, including baseline and follow-up predictors.

	Prostatectomy (N = 16,445) HR (95 % CI)	Radiotherapy (N = 19,531) HR (95 % CI)
Age at diagnosis (years)
35–59 (ref)	1.00	1.00
60–64	1.18 (1.05, 1.31)**	1.13 (0.97, 1.31)
65–69	1.07 (0.96, 1.21)	1.28 (1.11, 1.47)**
70+	1.70 (1.48, 1.96)**	1.33 (1.17, 1.52)**
Race
White (ref)	1.00	1.00
Black	1.23 (1.10, 1.37)**	1.14 (1.02, 1.28)*
Hispanic	0.99 (0.87, 1.13)	1.12 (0.99, 1.28)
All others	0.99 (0.83, 1.19)	1.02 (0.87, 1.19)
Risk group
High (ref)	1.00	1.00
Intermediate	0.77 (0.70, 0.86)**	0.68 (0.63, 0.75)**
Low	0.74 (0.65, 0.83)**	0.49 (0.43, 0.55)**
Tumor sequence
Single primary (ref)	1.00	1.00
1st of multiple	1.41 (1.21, 1.64)**	1.11 (0.98, 1.26)
2nd or more	0.95 (0.78, 1.16)	1.07 (0.91, 1.26)
Elixhauser score
0 (ref)	1.00	1.00
1	1.05 (0.95, 1.17)	1.01 (0.91, 1.12)
2	1.07 (0.94, 1.22)	1.13 (1.01, 1.27)*
3+	1.23 (1.07, 1.42)**	1.18 (1.05, 1.33)**
PSA doubling time category (months)
0.0–<9.0	28.40 (24.25, 33.27)**	15.82 (13.88, 18.02)**
9.0–<15.0	12.50 (10.25, 15.25)**	7.41 (6.42, 8.55)**
15.0–<36.0	5.59 (4.59, 6.81)**	2.64 (2.31, 3.02)**
36.0+ (no rise) (ref)	1.00	1.00
Salvage radiotherapy	1.43 (1.27, 1.61)**	–
Adjuvant ADT	1.68 (1.26, 2.25)**	1.17 (1.07, 1.29)**

Models adjusted for diagnosis year, health plan, and unknown PSA doubling time

HR hazard ratio, CI confidence interval, ADT androgen deprivation therapy, PSA prostate-specific antigen

^*p < 0.05;

^**p < 0.01

Table 3. Cox model of time to salvage ADT use after primary prostatectomy or radiotherapy among men with newly diagnosed localized prostate cancer, including only baseline predictors.

	Prostatectomy (N = 16,445) HR (95 % CI)	Radiotherapy (N = 19,531) HR (95 % CI)
Age at diagnosis (years)
35–59 (ref)	1.00	1.00
60–64	1.14 (1.02, 1.27)*	1.06 (0.91, 1.24)
65–69	1.03 (0.91, 1.16)	1.17 (1.01, 1.34)*
70+	1.76 (1.53, 2.03)**	1.23 (1.08, 1.40)**
Race
White (ref)	1.00	1.00
Black	1.24 (1.11, 1.38)**	1.09 (0.97, 1.22)
Hispanic	1.06 (0.94, 1.21)	1.10 (0.96, 1.25)
All others	0.92 (0.77, 1.11)	0.97 (0.83, 1.13)
Gleason score (baseline)
≤6 (ref)	1.00	1.00
7	1.50 (1.37, 1.66)**	1.67 (1.53, 1.83)**
8	2.44 (2.07, 2.87)**	2.35 (2.05, 2.68)**
9–10	3.74 (3.09, 4.53)**	3.39 (2.90, 3.97)**
PSA (baseline, ng/mL)
≤4 (ref)	1.00	1.00
4 < PSA ≤ 10	0.92 (0.82, 1.03)	1.03 (0.89, 1.19)
10 < PSA ≤ 20	1.32 (1.15, 1.52)**	1.44 (1.24, 1.69)**
PSA > 20	2.46 (2.08, 2.91)**	2.41 (2.05, 2.83)**
T stage
≤T2A (ref)	1.00	1.00
T2B	1.29 (1.13, 1.47)**	1.33 (1.19, 1.49)**
T2C+	0.90 (0.80, 1.01)	1.16 (1.05, 1.29)**
Tumor sequence
Single primary (ref)	1.00	1.00
1st of multiple	1.33 (1.14, 1.54)**	1.13 (0.99, 1.28)
2nd or more	1.03 (0.85, 1.25)	1.03 (0.88, 1.21)
Elixhauser score
0 (ref)	1.00	1.00
1	1.09 (0.98, 1.21)	1.07 (0.96, 1.18)
2	1.17 (1.02, 1.32)*	1.22 (1.09, 1.37)**
3+	1.31 (1.13, 1.51)**	1.31 (1.17, 1.48)**

Models adjusted for diagnosis year and health plan

HR hazard ratio, CI confidence interval, ADT androgen deprivation therapy, PSA prostate-specific antigen

^*p < 0.05;

^**p < 0.01

Discussion

ADT has been increasingly used as a salvage therapy for men experiencing biochemical recurrence following primary therapy of prostatectomy or radiotherapy for localized prostate cancer [9]. To our knowledge, this is the first report on the patterns of sADT utilization among a community-based cohort in the USA. Our study has several important findings. First, we found that a short PSA doubling time was strongly associated with sADT use, which is clinically sensible. Rising PSA levels and corresponding shorter PSA doubling times are indicators of biochemical or clinical recurrence, which are linked to a higher risk of death from prostate cancer [23]. Salvage radiotherapy (for the primary prostatectomy group) and adjuvant ADT were also independently associated with sADT use. Although these therapies are given to men having some signs of disease recurrence after surgery or to men having radiotherapy but with higher baseline risk of recurrence [24, 25], we found both were also associated with greater receipt of sADT after adjustment for factors associated with greater pre-treatment risk of recurrence.

As expected, we found that higher AUA risk group was predictive of sADT regardless of the primary therapy from the adjusted results. In the sensitivity analysis with only baseline characteristics, where we replaced the risk group by its composite measures (baseline PSA, Gleason score, and T stage), we found that baseline PSA was independently associated with sADT when its value is larger than 10 ng/mL, likely because of the high-risk nature of their tumors. Our findings for greater sADT use among men with other higher-risk prognostic characteristics such as higher Gleason score and T stage are consistent with our findings for baseline serum PSA levels, suggesting that men with a higher risk of recurrence/progression at initial diagnosis are more likely to receive sADT.

We found that older age and more comorbidity are associated with sADT use regardless of primary therapy, after adjusting for all other characteristics. It is possible that increasing age is related to more reluctance to administer local salvage therapy. The higher use of sADT among men with more comorbidity could reflect the fact that these men interact with their healthcare system more frequently and therefore are biased toward receiving interventions for multiple diseases. Such a finding could also indicate a potential overuse for older men with more comorbid conditions, who might suffer the adverse effects of ADT including cardiovascular disease and diabetes mellitus [12, 13]. However, this requires additional study.

We identified racial/ethnic differences in receipt of sADT with African-Americans more likely to receive sADT than whites regardless of primary therapy. This finding differs from those of Carson et al. [26] who found that black men were less likely to receive ADT than white men after adjustment for baseline risk. This discrepancy may be explained by the following three reasons. First, our study included men diagnosed with localized prostate cancer, while Carson et al. included men only diagnosed with stage IV prostate cancer. Second, our sample included more than 60 % of men with age <65, whereas the SEER-Medicare sample of Carson et al. has men only with age 65 and above. Third, our cohort is more contemporary with men diagnosed after 2000, while Carson's cohort was diagnosed from 1991 to 1999. However, findings of both studies reflect racial differences in the treatment of prostate cancer that need to be addressed.

The present study has several strengths. Using a community-based sample from a well-defined population, we included a large cohort of approximately 36,000 men with localized prostate cancer treated with primary prostatectomy or radiotherapy and had follow-up data for up to 15 years. We have extensive data on baseline prognostic factors and longitudinal laboratory test results including PSA values.

However, our findings are subject to limitations. First, we defined sADT based on the primary therapy, stage of cancer, and timing of ADT following primary therapy. Our electronic data did not allow us to specifically determine disease recurrence/progression, to differentiate adjuvant ADT from sADT with certainty, and to fully document the reasons for sADT initiation. For example, 4 % of our sample received sADT but did not have PSA data available prior to their receipt of sADT. They may have received PSAs that were not available in the electronic health records, or their sADT could have been triggered by other reasons, such as symptomatic local recurrence or metastatic disease, which we were unable to ascertain. Therefore, our study is only able to describe the pattern of sADT use and its associated characteristics. Second, our sample consisted of men enrolled in managed care plans. This may potentially limit the generalizability of our findings with respect to the use of sADT to fee-for-service settings. However, the populations covered by the three health plans are sociodemographically diverse, and they provide services to Medicare and Medicaid patients. Third, the analysis may be subject to confounding from unobserved factors, including provider-level clustering, physicians' or patients' preferences, lack of indication of sADT, and other unmeasured patient factors such as BMI, frailty, and health behaviors, which might have influenced sADT use.

Conclusions

Our study showed that approximately 13 % of men with localized prostate cancer who completed primary therapy with prostatectomy or radiotherapy received sADT within an average of 6 years following their primary therapy. In men with possible recurrent or progressive prostate cancer, sADT use was higher in those at greater risk of death from prostate cancer as reflected by initial diagnosis with higher-risk disease, shorter longitudinal PSA doubling time, and the need for other salvage therapies post-primary treatment. These patterns appear to be largely consistent with current recommendations [9]. The use of sADT after primary prostatectomy or radiotherapy is the standard care for men with detectable metastases, but the balance of benefits to risks remains uncertain for men with biochemical recurrence only, particularly among men who are older and have more comorbid conditions. More evidence regarding the benefits and risks of sADT for biochemical recurrence versus symptomatic recurrence is needed.

Appendix.

PSA following primary therapy

We classified men's highest post-primary therapy PSA values (ng/mL) into three categories: (1) at least one PSA ≥10, which may trigger the use of ADT [27]; (2) no PSA ≥10, but ≥2.0 above nadir (the lowest PSA level observed) for those with primary radiotherapy or ≥0.2 for those with primary prostatectomy [28]; and (3) all PSA <2.0 above nadir after radiotherapy or <0.2 after prostatectomy. These categories were mutually exclusive based on the highest PSA observed.

We estimated the PSA trajectory by calculating the PSA doubling time [21, 29] from the time of nadir after radiotherapy or undetectable PSA after prostatectomy to (1) the first sADT use or (2) the end of follow-up for patients without sADT use. The PSA doubling time was the natural log of 2 (=0.693) divided by the slope of a linear regression of the log(PSA) over time. The PSA slope was estimated with the use of linear least squares when three or more PSA values were available, or by calculation using the formula $\frac{log{2}_{\ast }(T2-T1)}{log\text{PSA}2-log\text{PSA}1}$ when only two PSA values were available, where PSA1 and PSA2 were obtained at times T1 and T2, respectively [29]. The PSA doubling time was finally categorized into four levels (in months): <9.0, 9.0–<15.0, 15.0–<36.0, and ≥36.0 (indicating no PSA rise).

Handling of missing data

A substantial proportion of cases (20 %) had at least one or more of the key clinical prognostic variables (clinical stage, Gleason score, or baseline PSA) missing. We performed multiple imputations using all other covariates to predict values for these variables. We constructed five imputed datasets, each having estimates for the missing values for PSA, Gleason score, and T stage. We then pooled the estimates and corresponding SEs across the five imputations using Rubin's method [30]. All model results used these imputed datasets; multivariable models using only the complete cases did not show any significant deviations from the results shown.