Intermediate clinical endpoints for surrogacy in localised prostate cancer: an aggregate meta-analysis

Summary

Background:

Intermediate Clinical Endpoints (ICEs) in Cancer of the Prostate (ICECaP) established metastasis-free survival (MFS) as a surrogate for overall survival (OS) in localized prostate cancer (PCa) using 19, predominantly radiotherapy-based, trials. We sought to comprehensively assess aggregate trial-level performance of commonly reported ICEs across all randomized trials in localized PCa.

Methods:

For this meta-analysis, PubMed was searched for trials in localized PCa published between Jan 1970-Jan 2020. Eligible trials were required to be randomized, therapeutic, reported both OS and ≥1 ICE, sample size ≥70. Trials of metastatic disease were excluded. ICEs included biochemical-failure (BF), local-failure (LF), distant metastases (DM), BF-free survival (BFS), progression-free survival (PFS), and MFS. Candidacy for surrogacy was assessed using the second condition of the meta-analytic approach, correlation of the treatment effect of the ICE and OS, using R² weighted by the inverse variance of the log ICE hazard ratio.

Findings:

Seventy-five trials (53,631 patients) were included. Median follow-up was 9·1 years. BF (R² 0·38, 95%CI 0·11–0·64), BFS (R² 0·12, 95%CI 0·00–0·33), and LF (R² 0·09, 95%CI 0·00–0·37) had poor correlation with OS. PFS (R² 0·46, 95%CI 0·22–0·67) had moderate correlation. MFS (R² 0·78, 95%CI 0·59–0·89) had strong correlation. All biochemical-based endpoints had poor correlation regardless of treatment type, length of follow-up, or enrollment of high-risk patients. MFS had strong correlation across subgroups, especially in trials of radiotherapy with hormone therapy (R² 0·96, 95%CI 0·87–0·99) and radical prostatectomy (R² 0·95, 95%CI 0·76–0·99).

Interpretation:

Biochemical- and local-failure-based ICEs failed to meet the second condition of the meta-analytic approach and are not surrogate endpoints for OS in localized PCa. This study validates MFS as the only identified surrogate endpoint for OS to date.

Funding:

This work was funded in part by the Prostate Cancer Foundation (to DES), the P50 CA186786 (DES), R01 CA240991-05 (DES).

Introduction

The long natural history of localized prostate cancer poses challenges in clinical trials designed to improve overall survival (OS). For this reason, there has been great interest in the identification of intermediate clinical endpoints (ICE) that may serve as valid surrogate endpoints for OS. The international Intermediate Clinical Endpoints in Cancer of the Prostate (ICECaP) working group established metastasis-free survival (MFS) as a surrogate endpoint for OS for men with localized prostate cancer.(1)

ICECaP utilized the gold-standard two-stage meta-analytic approach to establish surrogacy of ICEs.(2) There are numerous strengths to this approach. It requires standardization of individual patient data to meet the first condition, correlation of the ICE and OS. This requirement for individual patient data, however, limited the ICECaP analyses to 22% (n=19) of the identified viable randomized controlled trials. Similarly, 15 trials were included in the recent biochemical-based event-free survival (EFS) analysis.(3)

Surrogate endpoints are context-specific and may not perform equally across treatment types. In the ICECaP MFS analysis, radiotherapy trials comprised 90% of the patients, precluding subset analyses solely in surgically treated patients. Similarly, there were no surgical trials in the more recent EFS analysis.(3) How ICEs perform in these various settings are unknown. We therefore leveraged the second condition of the two-stage meta-analytic approach, correlation of the treatment effect on the ICE and OS, to comprehensively assess ICEs across the most common treatments in localized prostate cancer. This approach allows for broad inclusion of published randomized trials in this disease space, and may be used as a screening tool to identify potential ICEs for subsequent individual patient-level validation.

Methods

Search strategy and selection criteria

For this meta-analysis, randomized controlled trials in localized prostate cancer were identified using a systematic PubMed search in accordance with the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines,(4) and the data was analyzed according to the Reporting of Surrogate Endpoint Evaluation using Metanalyses (ReSEEM) guidelines (see Supplement for checklist).(5) Our study used a pre-specified analysis plan, although an online registry was not used. The search was supplemented with review of relevant abstracts from recent meetings, and was performed by LAG and DES, with conflicts resolved through discussion. Randomized trials published in English investigating a therapeutic option for localized prostate cancer from January 1970 to January 2020 were eligible that reported on OS and at least one ICE. All studies analyzed in ICECaP were included. Studies were excluded if they were screening, radiology or biopsy techniques, prevention, or non-medical intervention trials, enrolled metastatic or non-metastatic castrate resistant-prostate cancer patients, enrolled exclusively node positive disease, or if the sample size was less than 70 patients (this number chosen as ICECaP allowed only one trial with 62 patients).

Data Extraction

Seventy-five studies met eligibility criteria (Appendix page 14). Data was extracted by LAG, DW, GJ, SB, and NJ; variables extracted included treatment information, primary endpoint, single versus multicenter study, median follow-up, sample size, median age of patients enrolled, median baseline or entry prostate-specific antigen (PSA), risk groups eligible for inclusion (risk group defined by trial, most commonly National Comprehensive Cancer Network(6) or D’Amico(7)). For trials with multiple arms, separate randomizations were treated as separate comparisons. Hazard ratios (HR) for OS and the ICEs were collected. For trials that did not explicitly report HRs, Kaplan-Meier curves for relevant ICEs were extracted and digitized using WebPlotDigitizer(8) and HRs were estimated as previously described.(9)

Intermediate Clinical Endpoints

ICEs included endpoints of time to local failure (LF), time to biochemical failure (BF), time to distant metastases (DM), time to BF plus clinical failure (BF+CF; includes BF, LF, DM), BF-free survival (BFS; includes BF, LF, DM, or death), progression-free survival (PFS; LF, DM, or death), and MFS (DM or death). See Page 1 of Appendix for tabular list and definitions of all ICEs used. All endpoints were trial-defined, with the exception of trials included in the ICECaP analyses where endpoints were standardized and extracted.(1, 3)

Subgroup analyses

Preplanned subgroups across treatment type were assessed for the candidacy of each ICE. This included trials where men were treated with 1) radical radiotherapy, 2) radical prostatectomy, and 3) hormone therapy. These three groups are not mutually exclusive. Post-operative radiotherapy was not included in the radiotherapy subgroup, and instead remained in the surgical subgroup. Hormone therapy was defined as a first-generation luteinizing hormone-releasing hormone agonist, anti-androgen, or combination (of note: no trials were identified that included next-generation forms of hormone therapy). The hormone therapy subgroup included trials that assessed use or duration of hormone therapy (regardless of radical therapy), hormone therapy +/− radiotherapy, and orchiectomy trials.

Additional subgroups included trials that enrolled patients with high-risk disease and those with longer than the median follow-up of all eligible trials (>9 years). For any ICE analysis within a subgroup, correlation of treatment effect was only analyzed if there were at least five or more trials included in that particular subgroup in order to have confidence with the correlations identified (see Appendix Page 2 for the number of comparisons used for each subgroup and ICE, and which endpoints were not assessed).

Exploratory analyses

Within the radiotherapy subgroup, two smaller subgroups were assessed: 1) trials evaluating use or duration of hormone therapy and radiotherapy, and 2) trials evaluating radiotherapy dose-escalation, nodal irradiation, and hypofractionation.

Sensitivity analysis

As not all trials reported every ICEs, a sensitivity analysis was performed of trials that reported both BF and BFS to assess whether treatment effects on endpoints that include death as an event had stronger correlation of treatment effect estimates with OS, as would be expected. Leave-one out cross validation was performed as a sensitivity analysis. To minimize bias, only randomized trials with sample sizes ≥70 patients were included. Funnel plots were generated to show any potential source of reporting bias.

Statistical analysis

Candidacy for surrogacy was evaluated at the trial level using the second condition of the two-stage meta-analytic approach assessing the magnitude of the correlation between treatment effect estimates on the ICE and OS. Correlation was quantified with Pearson’s R² calculated by weighting each trial by the inverse of the variance of the log of the ICE hazard ratio.(2) The ICE was deemed to be a surrogate for OS if the R² was ≥0·7. To obtain the surrogate threshold effect (STE), 95% prediction limits for the regression line of the effect of treatment on OS versus the effect of treatment on the surrogate were constructed, using the same trial level weights used in calculating R². The intersection of the upper 95% prediction limit with the horizontal line, representing a HR of 1 for OS, was defined as the STE, corresponding to no treatment effect on OS. For any subgroups where the upper 95% confidence interval did not cross 1, STE was not calculated. Egger’s tests were performed to assess reporting bias. All statistical analyses were performed using R version 3.6.2 (R Foundation for Statistical Computing, Vienna).

Role of the Funding Source

The funding sources had no role in study design, data collection, data analysis, data interpretation, or writing of the report. The authors had full access to all the raw data in the study and the corresponding author had final responsibility for the decision to submit for publication.

Results

A total of 4,221 studies were identified. After screening for localized or biochemically recurrent prostate cancer, 75 randomized trials published between March 1986 and January 2020 remained, composed of 53,631 patients (Figure 1). Most were multicenter trials (87%, 65 studies) and allowed high-risk patients to enroll (73%, 55 studies). Median trial-level follow-up was 9·1 years (interquartile range, 4·7), median age was 68·7 years, and median baseline PSA was 12·1 ng/mL (Table 1).

Figure 1:

PRISMA diagram. Other sources indicate records identified from review of trials references lists and meeting abstracts.

Table 1:

Trial information

	% of Patients (or median)	No. of Patients (or range)	No. of Trials	No. of Comparisonsa
Total	100%	53,631	75	87
Patient Characteristics
Age (years)	68·7	(60–76)	-
Tumor Characteristics
Risk Group
Low risk	5·4%	2,913	37	40
Intermediate risk	22·6%	12,116	35	38
High risk	22·4%	12,036	35	38
Not Reported b	47·2%	25,316	38	47
Gleason Score
≤ 6	19·2%	10,291	40	43
7	27·9%	14,961	44	49
≥ 8	11·7%	6,285	48	53
Not Reported b	31·1%	16,672	25	32
T-stage
T1–2	62·5%	33,522	60	71
T3–4	26·4%	14,152	60	71
Not Reported b	10·7%	5,752	16	16
Baseline PSAc (ng/mL)
Median	12·1	(4·6–51·1)	-
<10	23·6%	12,678	28	31
≥10 and < 20	11·5%	6,140	25	27
≥20	7·7%	4,137	30	33
Not Reported b	26·6%	14,242	16	23
Treatment Details d
Local Therapy
Radiotherapyc	63·8%	34,237	54	57
Surgery	22·8%	12,235	19	19
Systemic Therapy
Hormone Therapy	48·9%	26,238	29	35
Chemotherapye	3·0%	1,618	5	5
Trial Characteristics
Median follow-up (years)	9·1	(3·4–23·6)	-
Multi-center study	96·2%	51,611	65	75
High risk disease eligible for enrollment	76·8%	41,164	55	66

^a.Multiple trials contained multiple randomized arms, and thus the total number of treatment effects assessed were greater than the total number of trials.

^b.Not reported indicates number of trials/comparisons that did not report the information in the extracted format in that category; for example, not reported in Baseline PSA indicates trials that did not report median PSA or breakdown into the three categories identified.

^c.Median prostate specific antigen (PSA) and radiotherapy groups do not include post-operative radiotherapy trials (i.e. delivery of adjuvant or salvage radiotherapy)

^d.Radiotherapy and surgery groups are mutually exclusive. They are not mutually exclusive with the systemic therapy subgroups.

^e.Five trials with chemotherapy included patients that received chemotherapy with surgery or radiation therapy.

All candidate ICEs were assessed in at least 12 trials and 11,396 patients (Table 2). Correlation of the treatment effect of each ICE with OS is seen in Figure 2. ICEs that included biochemical-failure had poor correlation with the treatment effect on OS: BF R² 0·38 (95%CI 0·11–0·64), BF+CF R² 0·28 (95%CI 0·004–0·65), BFS R² 0·12 (95%CI 0·003–0·33). Correlation with LF was also poor (R² 0·09, 95%CI 0·00–0·37). The treatment effect on DM alone, without death as an event, had weak correlation to the treatment effect on OS (R² of 0·24, 95% CI 0·01–0·56). Correlation of PFS was moderate (R² 0·46, 95%CI 0·22–0·67), and MFS had the strongest correlation (R² 0·78, 95%CI 0·59–0·89). A sensitivity analysis was performed using leave-one-out cross validation and the results remained consistent (Appendix page 5). Funnel plots showed no evidence of reporting bias (Appendix page 8).

Table 2:

Condition 2 surrogacy analysis

					Condition 2: correlation of treatment effects
True Endpoint	ICE	No. of Trials	No. of Comparisons	No. of Patients	R2 (95%CI)	STE	Regression Equation
OS	BF	25	27	17,086	0·38 (0·11–0·64)	0·31	Log(HR)OS = −0·0298 + 0·231*Log(HR)BF
OS	BF + CF	12	14	11,396	0·28 (0·005–0·65)	N/A	Log(HR)OS = −0·0203 + 0·181*Log(HR)BF+CF
OS	BFS	41	43	28,528	0·12 (0·003–0·33)	N/A	Log(HR)OS = −0·0478 + 0·166*Log(HR)BFS
OS	LF	21	24	13,753	0·09 (0·00–0·37)	N/A	Log(HR)OS = −0·0614 + 0·108*Log(HR)LF
OS	PFS	29	37	25,095	0·46 (0·22–0·67)	0·51	Log(HR)OS = 0·0061 + 0·428*Log(HR)PFS
OS	DM	18	21	13,623	0·24 (0·01–0·56)	N/A	Log(HR)OS = −0·0372 + 0·243*Log(HR)DM
OS	MFS	26	28	16,620	0·78 (0·59–0·89)	0·80	Log(HR)OS = −0·0261 + 0·661*Log(HR)DMFS

ICE: intermediate clinical endpoint; OS: overall survival; BF: biochemical failure: BF + CF: biochemical failure + clinical failure; BFS: biochemical-free survival; LF: local failure; PFS: progression-free survival; DM: distant metastases; MFS: metastases-free survival; STE: surrogate threshold effect

N/A indicates that the upper 95% confidence interval does not cross 1 and thus STE cannot be estimated

Figure 2:

Correlation between the treatment effect of the intermediate clinical endpoint and overall survival. Panels demonstrate (A) biochemical failure (BF), (B) biochemical failure and clinical failure (BF+CF), (C) biochemical failure free survival (BFS), (D) local failure (LF), (E) progression-free survival (PFS), (F) distant metastasis (DM), (G) metastasis free survival (MFS).

A total of 54 trials utilized primary radiotherapy in trial design, composed of 34,237 patients with a median follow-up of 8·5 years, median age of 69 years, and median PSA of 12·6 ng/mL. Additional characteristics for this group and all other subgroups are located in Appendix page 4. Findings for the radiotherapy trials mirrors the overall surrogacy, with poorest surrogacy for ICEs that included biochemical-failure (BF R² 0·21 (95%CI 0·004–0·53), BFS R² 0·21 (95%CI 0·02–0·46)) and LF (R² 0·05, 95%CI 0·00–0·33). PFS had good correlation (R² 0·72, 95%CI 0·47–0·86), and MFS had the strongest correlation (R² 0·78, 95%CI 0·56–0·90).

Given the lower overall correlation of MFS in radiotherapy trials (R² 0·78) compared to the ICECaP report (ICECaP R² 0·92), the cohort was separated into trials testing radiotherapy-specific questions (i.e. dose-escalation, nodal irradiation, and hypofractionation; 19 trials with 13,605 patients, median follow-up 6·8 years) and those that tested the use or duration of hormone therapy (21 trials, 14,374 patients, median follow-up 9·1 years). The latter group of trials primarily comprised the original ICECaP cohort with 13 of these trials represented in ICECaP. Notably, the former subgroup (radiotherapy-specific questions) showed poor correlation of BF, BFS, and MFS, and had an inverse correlation between these ICEs and OS (BF R² 0·04 (95%CI 0·00–0·53), BFS R² 0·15 (95%CI 0·00–0·53), MFS R² 0·31 (95%CI 0·00–0·79)). A further sensitivity analysis was performed on the BFS endpoint in the radiotherapy-specific trials with longer follow-up (>9 years). The estimated R² was stronger at 0·72 (95%CI 0·12–0·94), but remained inversely correlated to the treatment effect on OS (Appendix page 9). In contrast, MFS in radiotherapy-based trials testing the use or duration of hormone therapy had a strong correlation to OS (R² 0·96, 95%CI 0·87–0·99, Table 3).

Table 3:

Surrogacy analysis of subgroups

ICE	R2 (95%CI) for each subgroup
	All radiotherapy trials	Radiotherapy with hormone therapy	Radiotherapy-specific trials*	Hormone therapy trials	Surgical trials	Trials with median f/u > 9 yrs	Trials that allowed high risk enrollment
BF	0·21 (0·004, 0·53)	0·26 (0·00, 0·70)	0·04 (0·00–0·53)	0·51 (0·07, 0·82)	0·76 (0·19, 0·95)	0·46 (0·12, 0·74)	0·27 (0·02, 0·60)
STE	N/A	N/A	N/A	0·30	0·36	0·42	0·23
BFS	0·21 (0·02, 0·46)	0·53 (0·13, 0·81)	0·15 (0·00–0·53)	0·47 (0·13, 0·74)	0·04 (0·00, 0·54)	0·07 (0·00, 0·38)	0·16 (0·008, 0·41)
STE	0·22	0·47	N/A	0·44	N/A	N/A	N/A
LF	0·05 (0·00, 0·33)	0·23 (0·00, 0·68)	N/A	0·33 (0·003, 0·71)	N/A	0·14 (0·00, 0·49)	0·13 (0·00, 0·50)
STE	N/A	N/A	N/A	N/A	N/A	N/A	N/A
PFS	0·72 (0·47, 0·86)	0·83 (0·58, 0·94)	N/A	0·47 (0·16, 0·72)	0·63 (0·13, 0·89)	0·35 (0·07, 0·64)	0·65 (0·40, 0·81)
STE	0·68	0·74	N/A	0·50	0·34	0·41	0·62
DM	0·04 (0·00, 0·38)	0·03 (0·00, 0·55)	N/A	0·09 (0·00, 0·53)	0·96 (0·80, 0·99)	0·26 (0·001, 0·63)	0·23 (0·00, 0·62)
STE	N/A	N/A	N/A	N/A	0·68	0·26	N/A
MFS	0·78 (0·56, 0·90)	0·96 (0·87, 0·99)	0·31 (0·00–0·79)	0·96 (0·87, 0·99)	0·95 (0·76, 0·99)	0·73 (0·39, 0·90)	0·82 (0·64, 0·91)
STE	0·79	0·90	N/A	0·90	0·77	0·79	0·83

ICE: intermediate clinical endpoint; OS: overall survival; BF: biochemical failure: BF + CF: biochemical failure + clinical failure; BFS: biochemical-free survival; LF: local failure; PFS: progression-free survival; DM: distant metastases; MFS: metastases-free survival; STE: surrogate threshold effect

^*Trials testing radiotherapy dose, fractionation, and field size.

N/A indicates that the upper 95% confidence interval does not cross 1 and thus STE cannot be estimated

A total of 29 trials used hormone therapy as a component of treatment, composed of 26,238 patients with a median follow up of 9·7 years, median age of 69 years, and median PSA of 15·5 ng/mL. ICEs with biochemical-failure had moderate correlation, although improved from the overall cohort: BF R² 0·51 (95%CI 0·07–0·82), BFS R² 0·47 (95%CI 0·13–0·74). LF remained poor, with R² of 0·33 (95%CI 0·003–0·71). PFS had moderate correlation (R² 0·47, 95%CI 0·16–0·72) and MFS had the strongest correlation (R² of 0·96, 95%CI 0·87–0·99, Table 3).

Nineteen trials utilized radical prostatectomy as a component of treatment, composed of 12,235 patients with a median follow-up of 9·4 years, median age of 64·6 years, and median PSA of 7·8 ng/mL. Six of the 19 trials involved post-operative radiotherapy (3,490 patients). BFS poorly correlated with OS (R² 0·04, 95%CI 0·00–0·54) and PFS had moderate correlation (R² 0·63, 95%CI 0·13–0·89). MFS had strong correlation with OS (MFS R² 0·95 (95%CI 0·76–0·99), Table 3).

For trials with a median follow-up time >9 years (35 trials, 29,367 patients, median follow-up 10·5 years), the correlation results remained consistent with BFS performing poorly (BFS R² 0·07, 95%CI 0·00–0·38), and MFS having the strongest correlation (R² 0·73, 95%CI 0·39–0·90).

When restricting only to the subset of trials that allowed enrollment of high-risk patients (55 trials, 41,101 patients, median follow-up 9·1 years), biochemical-failure endpoints (BF R² 0·27 (95%CI 0·02–0·60), BFS R² 0·16 (95%CI 0·008–0·41)) and LF (R² 0·13, 95%CI 0·00–0·50) had poor correlation. MFS had the strongest correlation (R² 0·82, 95%CI 0·64–0·91, Table 3).

As BFS is a composite of BF and death (as well as LF and DM), it should have a stronger correlation than BF alone to OS. Although BF and BFS typically had similar R² across groups, BF appeared to be more strongly correlated than BFS to OS. We hypothesized this was due to different study compositions used for each endpoint, as multiple trials did not report both endpoints. Thus, we analyzed studies that reported both BF and BFS within each trial. Thirteen trials were identified, with a total of 8,305 patients. In this subgroup, the relationship between BF and OS and BFS and OS was as expected, with BFS having a higher correlation with OS than BF (BF R² 0·28 (95%CI 0·004–0·65), BFS R² 0·40 (95%CI 0·04–0·73)).

Surrogate Threshold Effects (STE) were calculated for all subgroups where the upper 95%CI crossed with the horizontal line of OS of 1. STE for BF overall was 0·31, and for all subgroups examined (radiotherapy, hormone therapy, median follow-up greater than 9 years, high-risk only) BF STE ranged from 0·23–0·42, indicating that a treatment effect of 58% or more would be needed to have a high confidence that the BF treatment effect would translate into an improvement in OS. The STE for MFS overall was 0·78, and for all subgroups examined (radiotherapy, hormone therapy, surgery, median follow-up, and high-risk trials) ranged from 0·77–0·90, indicating that a treatment effect of at least 23% on MFS would demonstrate improvement in OS across all subgroups. The MFS STEs for the hormone therapy trials and the radiotherapy plus hormone therapy trials were both 0·90, slightly higher than ICECaP’s STE of 0·88 (Table 3).(1)

Discussion

This analysis confirms the findings of the ICECaP group that MFS remains a surrogate for OS(1) and adds important additional information. First, we validate MFS as a potential surrogate endpoint across all eligible published randomized trials in localized prostate cancer, and specifically in trials that utilized surgery as well as hormone therapy. Second, to our knowledge, we demonstrate for the first time that the treatment effect on local failure had no correlation to the treatment effect on OS and should not be used as a surrogate endpoint for OS. Third, this study raises concerns for the use of all tested biochemical-based endpoints as surrogate endpoints for OS in future randomized trials. This held true even when restricting trials to those with long-term follow-up (>9 years) or those that included high-risk disease. Finally, we demonstrate that MFS may not be a surrogate endpoint for radiotherapy-specific trials testing dose-escalation, nodal irradiation, or hypofractionation. Interestingly, in radiotherapy plus hormone therapy trials the treatment effect correlation between MFS and OS was very strong.

The ICECaP group pre-specified an R² of greater than 0·7 as the definition of a surrogate endpoint.(10) In the current study, LF, BF, and BFS all had substantially lower R² than 0·7 (all <0·4). This confirms the recent radiotherapy-based ICECaP analysis that demonstrated that event-free survival (a composite of BF, LF, DM, or death—same as BFS in the present study) had a weak correlation with OS (in that study, R² 0·35, 95%CI 0·01–0·60).(3)

ICEs in trials that primarily tested modalities of radiotherapy (i.e. dose-escalation, nodal irradiation, and hypofractionation) demonstrated a slight inverse correlation of each ICE with OS. The correlation became stronger when analyzing trials with >9 years of follow-up (R² 0·72, but inverse correlation). The goal of dose-escalation is primarily to reduce LF, and we demonstrate for the first time that LF had no correlation with OS in any of the subgroups examined (R² 0·01–0·23). This may explain why every dose-escalation trial to date, even those powered for OS, have not improved OS.(11–13) Based on mounting evidence of the inability of biochemical-based and now LF endpoints to function as surrogate endpoints, an expectation of a positive impact on OS from therapies adopted as standard of care based on benefits to BF and LF requires revision.

Surrogate endpoints are context-specific. An important finding of our study was to ensure that MFS remains a valid surrogate endpoint post-surgery, as this was not able to be shown in ICECaP. This was because only 8% of patients in the original ICECaP MFS study, and 0% in the follow-up EFS study, were from surgical trials. The reason surrogate endpoints may be treatment-specific is due the differences in salvage treatment options. For example, prostatectomy failures are typically salvaged with radiotherapy, whereas failures after radiotherapy historically are less commonly salvaged with local therapy. Fortunately, our analysis confirms that MFS is a strong candidate for surrogacy irrespective of primary radical treatment. Other ICEs in the surgical subgroup often had wide confidence intervals given the overall paucity of the conduct of surgical randomized trials.

Valid surrogate endpoints are important for optimal clinical trial design, as it offers a means to measure an endpoint that may occur within a shorter timeframe than the ultimate endpoint of interest. Long-term follow-up of patients is resource-intensive and challenging. Additionally, use of surrogate endpoints could allow new therapeutic options to reach patients sooner. Estimating time savings with the use of a validated surrogate endpoint is complex and no simple generalization can be made. The potential time saved depends on the estimated treatment effect on the ICE and OS, accrual rate, timing of events in the natural history of the disease, type I and II error desired, among numerous other variables.

Our study specifically focused on surrogate endpoints of OS. There are other important endpoints that are worthy of further investigation, including quality of life, treatment burden, and financial toxicity, that may help patients make optimal treatment decisions. Thus, this data is best applied to future superiority trials attempting to provide clinically meaningful improvement in oncologic outcomes. It should also be noted that another reason that MFS remains a surrogate endpoint, and BFS is not, is that it is estimated that >70% of MFS events are from death as the first event. In contrast, >75% of events that comprise the BFS endpoint are estimated to come from recurrence (most commonly BF).(3) Although improvements in BF may be important, we must not forget that non-prostate cancer deaths remain the dominant mode of death, and our therapies can worsen other-cause mortality. For example, the addition of hormone therapy to salvage radiation in patients with a PSA of <1.5 ng/mL increased other-cause mortality and cardiac events.(14) Thus, the poor correlation of BFS to OS may also be due to treatment intensification resulting in worsened other-cause mortality without greater improvements in prostate cancer-specific mortality.(15)

Limitations of this study are worthy of discussion. The goal of this study was to comprehensively screen all commonly reported ICEs and assess their potential as surrogate endpoints in clinically relevant subgroups using aggregate data. This was an intentional tradeoff to overcome the inability to acquire individual patient data to this scale, which is a global challenge in biomedical research. To minimize risk of bias in individual studies, all eligible trials were included in the analysis; however, risk of bias across studies remains due to selective reporting and changes in measurement of endpoints. Additionally, using this approach does not allow harmonization of exact definition of endpoints, such as PSA cut points for BF. Furthermore, there are real-world limitations to the determination of LF, as it is based on frequency and type of assessment (e.g. digital rectal exam vs imaging) with variable pathologic confirmation. Similarly, follow-up imaging is based on physician discretion and may have changed across time. Novel molecular imaging is unlikely to have been used in almost any trial included, and further study is needed to understand its role and impact on surrogate endpoints. For example, it is probable that MFS based on PET/CT imaging may have weaker correlation given it more closely approximates BFS.

Information on type and time to salvage local and/or systemic therapies were not reported by most trials and were likely heterogeneous across patients; these may also impact the surrogacy of ICEs on OS. Recording of salvage treatments, timing of therapies, and testosterone data, would greatly help the development and understanding of potential future surrogate endpoints. Finally, prostate-cancer specific mortality and other-cause mortality were not reported in many trials and not able to be assessed as an endpoint in this analysis. However, despite these limitations, our results were highly consistent across treatment subgroup, length of follow-up, and inclusion of high-risk disease.

In conclusion, this study confirms that MFS remains the only validated surrogate endpoint of OS to date for localized and recurrent prostate cancer. We show that this remains true across multiple subgroups of patients and treatments. Currently, the optimal endpoints for future clinical trials intended to demonstrate improvement in OS are direct measurement of OS or use of MFS as a surrogate for OS; improvements in solely LF or BF are unlikely to translate into improvements in OS. Ongoing investigation and development of novel endpoints and incorporation of biomarkers are needed to identify additional surrogate endpoints.

Research in context.

Evidence before this study

Surrogate endpoints allow for efficient clinical trial design. The use of surrogate endpoints in localized prostate cancer is particularly relevant given the long natural history of the disease and subsequent challenges in required follow-up time in therapeutic trials powered for improvements in overall survival. The international Intermediate Clinical Endpoints in Cancer of the Prostate (ICECaP) working group established metastasis-free survival as a surrogate endpoint for overall survival for men with localized prostate cancer, and more recently demonstrated that event-free survival (a biochemical failure-based endpoint) is not a surrogate for overall survival. The ICECaP studies serve as the gold-standard to date, but the requirement of individual patient level data limited the number of trials that could be analyzed. We sought to expand upon these analyses by including all eligible clinical trials, focusing on the second condition of the two-stage meta-analytic approach for surrogacy, the correlation of the treatment effect of the intermediate clinical endpoint and overall survival. We searched PubMed for therapeutic randomized trials in localized prostate cancer published between 1970–2020. We used the MeSH Terms “random allocation” and “prostatic neoplasms”, filtered for Clinical Trials. Eligible trials were required to be randomized, testing a therapeutic intervention, had ≥70 patients, were a primary report of a trial, and reported at least one intermediate clinical endpoint in addition to overall survival.

Added value of this study

First, we were able to validate metastasis-free survival as a surrogate endpoint across all eligible published randomized trials in localized prostate cancer, but specifically now in trials that exclusively utilized surgery as well as hormone therapy. Second, we demonstrate for the first time, to our knowledge, that the treatment effect on local failure has almost no correlation to the treatment effect on overall survival, and thus should not be used as a surrogate endpoint. Third, this study raises concerns for the use of all tested biochemical-based endpoints as surrogate endpoints for overall survival in future randomized trials. Finally, we demonstrate that the treatment effect of biochemical failure had an inverse correlation to that of overall survival for radiotherapy-specific trials testing dose-escalation, nodal irradiation, or hypofractionation, and requires further investigation.

Implications of all the available evidence

Our findings have several potential implications for interpretation of prior clinical trials results, current clinical trial design, and future research. We confirm that metastasis-free survival remains the only validated surrogate endpoint to date for localized prostate cancer, and this should be the endpoint on which new therapies aimed to improve overall survival are approved and adopted. The findings demonstrated here raise the question of the impact of treatment paradigm shifts that were made based on unvalidated surrogate endpoints. Ongoing investigation and development of novel endpoints and incorporation of biomarkers are needed to identify potential earlier surrogate endpoints.