Indirect treatment comparisons of darolutamide plus docetaxel and androgen deprivation therapy in patients with metastatic hormone-sensitive prostate cancer

Abstract

Androgen deprivation therapy (ADT) has long been the standard-of-care for metastatic, hormone-sensitive prostate cancer (mHSPC). The addition of docetaxel (DOC) and/or androgen receptor axis–targeted therapies (ARATs) such as darolutamide (DAR), enzalutamide (ENZ), apalutamide (APA), abirateone (ABI), and rezvilutamide (REZ) has been shown to significantly improve overall survival (OS) over standard-of-care in mHSPC, including standard of care plus DOC. We indirectly compared OS and progression-free survival (PFS) of DAR+DOC+ADT against approved comparators in China using Bayesian network meta-analysis. Comparator treatment data derived from published trials (identified via Medline, EMBASE, and Cochrane Library searches) and included ENZ, APA, DOC, ABI, and REZ, each with ADT. Sensitivity analysis assumed Standard Nonsteroidal Antiandrogen (SNA)+ADT efficacy was equivalent to ADT. Fixed and random effects analyses were performed in intention-to-treat (ITT) and high-volume populations. Results were summarized using hazard ratios (HRs) relative to DAR+DOC+ADT. HRs numerically favoured DAR+DOC+ADT on all comparisons. HRs on OS (fixed effects) strongly favoured DAR+DOC+ADT against DOC+ADT (0.68 [95% CrI: 0.57–0.80]), ADT (0.55 [0.44–0.67]), and SNA+ADT (0.44 [0.28–0.70]) in ITT population; similar results were observed in high-volume population in addition to APA+ADT (0.69 [0.50–0.96]). Excluding the comparison against ABI+ADT (ITT population; random effects), which was not statistically significant, HRs on PFS strongly favoured DAR+DOC+ADT on all comparisons. Outcome definition on PFS varied across trials and is a limitation in mHSPC comparisons. Results numerically favoured DAR+DOC+ADT on OS and PFS across all comparisons, with strong evidence against APA+ADT (in the high-volume population only), DOC+ADT, ADT, and SNA+ADT on OS and all comparators on PFS (excluding ABI+ADT in ITT). Findings support the continued use of DAR+DOC+ADT as frontline treatment in mHSPC, particularly in China where REZ is approved.

Supplementary Information

The online version contains supplementary material available at 10.1038/s41598-025-23491-0.

Introduction

Prostate cancer is the most common male malignancy and affects up to 1.4 million men worldwide each year^1,2. The 5-year survival rate is 98.2% in non-metastatic prostate cancer, and 30% in metastatic, hormone-sensitive prostate cancer (mHSPC)³. Androgen deprivation therapy (ADT) has long been the standard-of-care for mHSPC⁴. More recently, trials have shown that docetaxel (DOC) can be added to ADT to improve survival^5,6. The addition of the androgen receptor axis–targeted therapies (ARATs) abiraterone acetate (ABI), enzalutamide (ENZ), and apalutamide (APA) have similarly been shown to improve survival^7–12. Despite these advances in treatments for mHSPC, concerns have been raised about the safety of ARATs, such that the whole benefit-risk profile must be considered in treatment decisions^1,13. Research suggests ARAT-based doublet therapies have a more favourable benefit-harm balance than DOC-containing strategies, while triplet therapies (i.e., those in combination with DOC plus ADT) improve survival outcomes without a significant increase in severe adverse events¹⁴. Darolutamide (DAR) is a distinct androgen-receptor inhibitor with a design that limits potential for clinically relevant drug-drug interactions, or adverse events^15,16. The ARASENS phase 3 trial also found significant overall survival benefit of DAR over placebo when added to DOC plus ADT, with similar adverse events observed in both groups¹⁵.

With such favourable survival outcomes associated with the use of ARATs in combination with ADT, the treatment landscape for mHSPC is evolving and new treatment combinations are being developed and approved by international regulators. In 2022, rezvilutamide (REZ) received its first international approval in China for the treatment of patients with mHSPC with high-volume disease¹⁷. Rezviluatmide in combination with ADT showed promising OS and PFS outcomes compared with bicalutamide in the phase 3 CHART trial, which may support its use as frontline treatment for patients with high-volume mHSPC, both in China and other settings (once approved)¹⁸.

Despite the relatively small number of active therapies in mHSPC, the relative effectiveness of these different treatment combinations has been the subject of much research in recent years, with studies relying on indirect treatment comparison (ITC) methodologies to compare treatments. Bayesian network meta-analysis (NMA) is the gold standard for ITC and has been endorsed internationally by the UK National Institute for Heath and Care Excellence (NICE), International Society for Pharmacoeconomics and Outcomes Research (ISPOR), and the World Health Organisation (WHO)^19–22. However, few studies have applied the methodology to establish relative efficacy in mHSPC, and instead applied a mix of Bucher comparisons and frequentist NMA. Within these, widely varying approaches were followed to connect evidence networks, compare treatment classes, address issues of heterogeneity, and account for correlations between treatment arms in multi-arm multi-stage (MAMS) adaptive design trials, such as the STAMPEDE trial^6,23. Studies have also varied in their focus on different populations, or subgroups; limited their comparisons to select therapies; or pooled treatments (e.g., ARATs) to compare relative effects within treatment classes. As a consequence, results from applied ITCs in mHSPC have largely varied. The optimal treatment choice is further dependent on patient characteristics. A meta-analysis suggests that patients who are younger, have a higher BMI, or have low-volume disease may benefit more from doublet therapies²⁴, while triplet therapies may be more suited for patients with high-volume disease¹⁴.

Nevertheless, there is a strong trend that DAR + DOC + ADT provides the greatest survival benefits for patients with mHSPC, and typically outperforms doublet therapies across a number of populations^25–27. Using Bayesian NMA, Zhou and colleagues found that DAR + DOC + ADT ranked first for improvement in OS among available systemic therapies for high-volume mHSPC, ahead of ABI + DOC + ADT and REZ + ADT. The treatment also ranked first on PFS, ahead of ABI + DOC + ADT and ENZ + ADT²⁷. In contrast, REZ + ADT ranked fifth on PFS in high-volume patients. Separately, Dou and colleagues compared four triplet therapies with DOC + ADT using Bayesian NMA (including ABI + DOC + ADT, APA + DOC + ADT, DAR + DOC + ADT, and ENZ + DOC + ADT), and found DAR + DOC + ADT emerged as superior on OS among patients with a high Gleason score ≥ 8, ECOG 0, as well as ECOG 1²⁶. In contrast, Lee and colleagues found that REZ + ADT ranked first on OS and PFS in patients with (high- and low volume combined) mHSPC, while DAR + DOC + ADT ranked fifth on both outcomes²⁸. However, unlike previous NMAs, where treatments largely connected to the evidence network via ADT monotherapy, Lee and colleagues anchored treatments on standard of care (SOC), which comprised evidence on ADT monotherapy and ADT + DOC from PEACE-I and ARASENS trials. The problem with this approach is that it likely biased against treatments that included DOC in the control and the intervention by cancelling out its effect. The authors also conducted separate subgroup analyses of high- and low-volume populations, whereby ABI + ADT and APA + ADT ranked first on OS in high- and low-volume populations, respectively; neither DAR + DOC + ADT nor REZ + ADT were included in the comparisons. Jian and colleagues similarly found REZ + ADT had the greatest potential to benefit patients with high-volume disease, after triplet therapies, but the NMA did not include DAR + DOC + ADT in the comparison as subgroup data were unavailable at the time of the analysis²⁹. As such, there remains incomplete and often contrasting evidence on the efficacy of available systemic therapies in mHSPC.

In this paper, we aimed to compare the relative efficacy of systemic therapies for mHSPC, comparing DAR + DOC + ADT with licensed and approved comparators in the Chinese setting where REZ + ADT has been approved, with implications for other settings where REZ may be approved in the future. Unlike a number of previous studies, we used Bayesian NMA to indirectly compare treatments and considered both intention-to-treat (ITT) and high-volume populations.

Materials and methods

Evidence network

We compared licensed and approved systemic therapies in the Chinese setting where REZ + ADT was recently approved. These were identified from a systematic literature review, details of which can be found in Online Resource 1, including key criteria for inclusion in in the clinical SLR (Table S1), Embase search (Table S2), Cochrane search (Table S3), Medline search (Table S4), and PRISMA flow chart (Figure S2) (publication forthcoming). The treatments (trials) included:

• ABI + ADT (LATITUDE, STAMPEDE G vs. A)^10,11.

• APA + ADT (TITAN)⁷.

• ENZ + ADT (ENZAMET, ARCHES)^9,30.

• DAR + DOC + ADT (ARASENS)¹⁵.

• REZ + ADT (CHART)¹⁸.

We excluded ABI + DOC + ADT on the basis that ABI + ADT is more widely used in its place in clinical practice in China³¹. A connected evidence network was available to indirectly compare all treatments, as shown in Fig. 1. Treatments largely connected to the network via ADT alone. REZ + ADT connected to the network via bicalutamide, which is a standard nonsteroidal antiandrogen (SNA) often prescribed for patients with mHSPC in China³¹. In sensitivity analysis, we assumed similar efficacy for SNA + ADT as ADT alone, similar to a previously published NMA by Menges and colleagues¹. This simplified the network for a comparison with REZ + ADT, as detailed in Online Resource 1 (Figure S1).

Fig. 1.

Evidence network for NMA on OS and PFS (relevant to the Chinese setting). ADT androgen deprivation therapy, NMA network meta-analysis, OS overall survival, PFS progression-free survival, SNA  standard nonsteroidal antiandrogen.

Data

Table 1 summarizes the input data (i.e., HRs) for each comparator used in the NMA on OS and PFS, as obtained from the SLR. All source data derived from pivotal randomised controlled trials (RCTs) assessing the efficacy of comparators (i.e., ENZ + ADT (ENZAMET, ARCHES)^9,30, APA + ADT (TITAN)^7,32, REZ + ADT (CHART)¹⁸, DOC + ADT (GETUG, CHAARTED, STAMPEDE arms C vs. A)^5,33–36, ABI + ADT (LATITUDE, STAMPEDE G vs. A)^{10,11,37–39}, and DAR + DOC + ADT (ARASENS)¹⁵; a detailed overview of included studies is provided in Table S5. We considered both ITT and high-volume populations; the latter derived from subgroup analyses, with the exception of the CHART trial which was entirely in high-volume patients (hence, ITT and high-volume HRs align for this study)¹⁸. High-volume mHSPC was defined as the presence of visceral metastases or ≥ 4 bone lesions with at least 1 lesion beyond the vertebral bodies and pelvis.

Table 1.

OS and PFS input data based on ITT and high-volume populations for NMA.

Trial name	Treatment arm 1	Treatment arm 2	N – arm 1	N – arm 2	OS (HR (95% CI)	PFS (HR (95% CI)
ARASENS15	DAR + DOC + ADT	DOC + ADT	ITT: 651 High volume: 497	ITT: 654 High volume: 508	ITT: 0.68 (0.57, 0.80) High volume: 0.67 (0.56, 0.80)*	ITT: 0.40 (0.35, 0.47) High volume: 0.43 (0.37, 0.50)*
CHART18	REZ + ADT	SNA + ADT†	ITT: 326 High volume: 326	ITT: 328 High volume: 328	ITT: 0.58 (0.44, 0.77) High volume: 0.58 (0.44, 0.77)	ITT: 0.44 (0.33, 0.58) High volume: 0.44 (0.33, 0.58)
ARCHES30	ENZ + ADT	ADT	ITT: 574 High volume: 354	ITT: 576 High volume: 373	ITT: 0.66 (0.53, 0.81) High volume: 0.66 (0.52, 0.83)	ITT: 0.63 (0.52, 0.76) High volume: 0.43 (0.33, 0.57)
CHAARTED5,36	DOC + ADT	ADT	ITT: 397 High volume: 263	ITT: 393 High volume: 250	ITT: 0.77 (0.65, 0.92) High volume: 0.63 (0.50, 0.79)	ITT: 0.62 (0.51, 0.75) High volume: 0.53 (0.42, 0.67)
GETUG-AFU 1534,35	DOC + ADT	ADT	ITT: 192 High volume: 92	ITT: 193 High volume: 91	ITT: 0.88 (0.68, 1.14) High volume: 0.78 (0.56, 1.09)	ITT: 0.69 (0.55, 0.87) High volume: 0.61 (0.44, 0.83)
LATITUDE10,37	ABI + ADT	ADT	ITT: 597 High volume: 487	ITT: 602 High volume: 468	ITT: 0.66 (0.56,0.78) High volume: 0.62 (0.52, 0.74)	ITT: 0.47 (0.39, 0.55) High volume: 0.46 (0.39, 0.54)
STAMPEDE arm G vs. A11,38,39	ABI + ADT	ADT	ITT: 960 High volume: 243	ITT: 957 High volume: 256	ITT: 0.60 (0.50, 0.71) High volume: 0.60 (0.46, 0.78)	ITT: 0.34 (0.29, 0.40) High volume: 0.46 (0.36, 0.58)
STAMPEDE arm C vs. A33	DOC + ADT	ADT	ITT: 362 High volume: 148	ITT: 724 High volume: 320	ITT: 0.81 (0.69, 0.95) High volume: 0.81 (0.64, 1.02)	ITT: 0.66 (0.57, 0.76) High volume: 0.68 (0.54, 0.85)
TITAN7,32	APA + ADT	ADT	ITT: 525 High volume: 325	ITT: 527 High volume: 335	ITT: 0.65 (0.53, 0.79) High volume: 0.70 (0.56, 0.88	ITT: 0.48 (0.39, 0.60) High volume: 0.53 (0.41, 0.67)
ENZAMET9	ENZ + ADT‡	SNA + ADT	ITT: 309 High volume: 114	ITT: 313 High volume: 118	ITT: 0.53 (0.37, 0.75) High volume: 0.65 (0.42, 0.99)	ITT: 0.34 (0.26, 0.44) High volume: 0.38 (0.27, 0.55)

*Based on an analysis of ARASENS IPD whereby patients were stratified for metastatic stage according to the TNM system (M1a, nonregional lymph node metastases only; M1b, bone metastases with or without lymph node metastases, or M1c, visceral metastases with or without lymph node or bone metastases) and alkaline phosphatase level.

^†SNA = Bicalutamide.

^‡Results from the non-docetaxel patient subgroup, thus not using the reported ENZ + DOC + ADT vs. SNA + ADT hazard ratios. ABI = abiraterone acetate; ADT = androgen deprivation therapy APA = apalutamide; DAR = darolutamide; DOC = docetaxel; ENZ = enzalutamide; HR = hazard ratio; ITT = intention-to-treat; OS = overall survival; PFS = progression-free survival; REZ = rezvilutamide; SNA = standard nonsteroidal antiandrogen.

While OS was defined similarly across trials, the definition of PFS varied somewhat (Table S6). Most studies relied on first documented radiographic progression (or death) to define PFS^7,10,18,30; three studies used biochemical progression (including biochemical failure)^11,33,35; two studies defined PFS using clinical assessment^5,9. In contrast, ARASENS used time to castration resistant prostate cancer or death (CROD) to measure disease progression¹⁵. Although differing approaches were used to measure PFS, studies largely relied on RECIST to define disease progression⁴⁰, including ARASENS¹⁵.

Statistical analysis

Proportional hazards NMA

We used Bayesian NMA to compare OS and PFS based on published HRs in ITT and high-volume populations. We followed the Bayesian generalised linear model framework described in the NICE Decision Support Unit (DSU) Technical Support Documents (TSD) on evidence synthesis and commonly employed in the NMA literature^19,41. A detailed description of the methods used to assess relative effectiveness is presented in Online Resource 1 (Table S7), along with a description of the adjustment used to account for correlations between arms in the STAMPEDE trial, which used MAMS design, that was informed by Vale and colleagues (Table S8, Table S9, and Table S10)⁴².

Assessing model fit

The preferred model was chosen based on goodness of fit. This was measured using the posterior mean of the residual deviance, which is a measure of the magnitude of the difference between the observed data and the model predictions for those data⁴³. Smaller values are preferred, and in a well-fitting model the posterior mean residual deviance should be close to the number of data points. The deviance formula for a Binomial likelihood is provided in the NICE DSU TSD 2⁴¹.

In comparing models, difference of ≥ 5 points for posterior mean residual deviance and deviance information criteria (DIC) were considered meaningful⁴⁴, with lower values favoured. In the event that the random effects models did not converge due to lack of data, we employed predictive distributions for the heterogeneity variance as informative priors. For continuous outcomes we used those developed by Rhodes and colleagues based on 6,492 meta-analyses⁴⁵.

Statistical analysis

All methods were implemented in the OpenBUGS Bayesian software, with code modified from the NICE DSU TSD 2^19,46. This uses Markov Chain Monte Carlo (MCMC) simulation to sample from the posterior distribution of NMA models. We used 30,000 burn-in simulations and 30,000 sampling simulations, with 3 chains. Convergence was assessed by visual inspection of the trace plots and the Brookes-Gelman-Rubin (BGR) Rhat statistic, which have been reported for model parameters⁴⁶.

The posterior distributions of relative treatment effects (i.e., HR of OS/PFS) between interventions were summarized by their median and 95% credible intervals (CrIs), which were constructed from the 2.5th and 97.5th percentiles of the MCMC samples. Surface under the cumulative ranking curve (SUCRA), with values close to 1.0 being favourable, was also calculated. A summary of the complete range of analyses conducted, and associated assumptions, are summarised in Table S11.

Results

OS

Model assessment statistics are reported in Table S12. The DIC favoured a fixed effects model due to having a lower value but there was limited difference with the random effects model. The total residual deviances indicated that both models fit the data well and convergence was achieved. Coupled with limited heterogeneity in outcome definition, for example, a fixed effects model was chosen in both populations.

Fixed effects comparing DAR + DOC + ADT with other treatments in the Chinese setting on OS are presented in Table 2; Fig. 2 for both ITT and high-volume populations. There was strong evidence of a difference of DAR + DOC + ADT against DOC + ADT (HR: 0.68, 95% CrI: 0.57, 0.80), ADT (HR: 0.55, 95% CrI: 0.44, 0.67), and SNA + ADT (HR: 0.44, 95% CrI: 0.28, 0.70) in the ITT population. Similar findings were observed in the high-volume population, as well as against APA + ADT (HR: 0.69, 95% CrI: 0.50, 0.96) in high-volume population. As per the SUCRA values, which indicate the probability of treatment being the best (with values close to 1.0 being favourable), DAR + DOC + ADT ranked highest in both populations, while REZ + ADT ranked in fifth and second place in ITT and high-volume populations, respectively. Complete cross tables comparing all interventions on OS are presented in Table S13, Table S14, Table S15, and Table S16; random effects models are presented in Table S17, which showed similar results to the fixed effects model, with the exception that there was no evidence of a difference in OS against APA + ADT in the high-volume population.

Table 2.

OS hazard ratios and SUCRA for ITT and high volume populations (fixed effects).

Treatment	ITT population (fixed effects)				High volume population (fixed effects)
	HR (95%CrI)	Rank number	SUCRA	Mean rank (95%Crl)	HR (95%CrI)	Rank number	SUCRA	Mean rank (95%Crl)
DAR + DOC + ADT	Reference	1	0.93	1.508 (0.999, 4.000)	Reference	1	0.95	1.363 (1.000, 2.675)
ABI + ADT	0.87 (0.69, 1.10)	2	0.73	2.901 (1.000, 5.000)	0.79 (0.60, 1.04)	3	0.70	3.072 (1.150, 5.000)
APA + ADT	0.84 (0.63, 1.12)	3	0.67	3.337 (1.000, 5.000)	0.69 (0.50, 0.96)	5	0.48	4.656 (2.000, 7.000)
ENZ + ADT	0.83 (0.62, 1.12)	4	0.64	3.496 (1.000, 5.001)	0.73 (0.53, 1.02)	4	0.57	3.977 (2.000, 6.000)
REZ + ADT	0.76 (0.44, 1.30)	5	0.53	4.274 (1.000, 7.000)	0.82 (0.45, 1.51)	2	0.71	3.061 (1.000, 7.000)
DOC + ADT	0.68 (0.57, 0.80)	6	0.34	5.612 (4.000, 6.552)	0.67 (0.56, 0.80)	6	0.41	5.147 (3.000, 7.000)
ADT	0.55 (0.44, 0.67)	7	0.14	7.051 (6.000, 8.000)	0.48 (0.38, 0.61)	8	0.08	7.443 (6.000, 8.003)
SNA + ADT	0.44 (0.28, 0.70)	8	0.03	7.821 (6.984, 8.004)	0.48 (0.28, 0.82)	7	0.10	7.281 (4.218, 8.002)

Treatments are listed in the order of their SUCRA ranking in the ITT population. SUCRA values close to 1.0 are favourable. HR values below 1.00 favour reference (i.e., DAR + DOC + ADT). Values are highlighted in bold if the 95% CrI exclude 1.00 .

ABI = abiraterone acetate; ADT = androgen deprivation therapy APA = apalutamide; DAR = darolutamide; DOC = docetaxel; ENZ = enzalutamide; HR = hazard ratio; ITT = intention-to-treat; OS = overall survival; REZ = rezvilutamide; SNA = standard nonsteroidal antiandrogen; SUCRA = surface under the cumulative ranking curve.

Fig. 2.

Forest plot depicting (A) overall survival and (B) progression-free survival in both ITT and high-volume populations. ABI = abiraterone acetate; ADT = androgen deprivation therapy APA = apalutamide; DAR = darolutamide; DOC = docetaxel; ENZ = enzalutamide; HR = hazard ratio; ITT = intention-to-treat; PFS = progression-free survival; REZ = rezvilutamide; SNA = standard nonsteroidal antiandrogen; SUCRA = surface under the cumulative ranking curve.

PFS

Model assessment statistics are reported in Table S18. The DIC favoured a random effects model in ITT population due to > 5 point different in total residual deviance. We therefore selected random effects in the ITT population, and this is supported by heterogeneous reporting of the PFS outcome and differences in baseline characteristics. In the high-volume population, there was limited difference between the fixed and random effects model; hence, fixed effects model was chosen due to the reduced heterogeneity in patient population, as well as due to having a lower DIC.

Random and fixed effects on PFS comparing DAR + DOC + ADT with other treatments in ITT and high volume populations, respectively, are presented in Table 3; Fig. 2. Excluding ABI + ADT, there was strong evidence of a difference of DAR + DOC + ADT against all other treatments including APA + ADT (HR: 0.56, 95% CrI: 0.32, 0.95), ENZ + ADT (HR: 0.42, 95% CrI: 0.24, 0.73), DOC + ADT (HR: 0.40, 95% CrI: 0.29, 0.57), REZ + ADT (HR: 0.33, 95% CrI: 0.15, 0.71), ADT (HR: 0.27, 95% CrI: 0.18, 0.40), and SNA + ADT (HR: 0.14, 95% CrI: 0.07, 0.28) in the ITT population. In the high-volume population, there was evidence of a difference of DAR + DOC + ADT against all comparator treatments. In both populations, DAR + DOC + ADT ranked highest, as per SUCRA value, while REZ + ADT ranked sixth and fourth in ITT and high-volume populations, respectively. Complete cross tables are presented in Table S19, Table S20, Table S21, and Table S22 with fixed and random effects models in ITT and high-volume populations presented in Table S23; these results were broadly comparable with the exception that there was no evidence of a difference in PFS against REZ + ADT in the high-volume population, while there was evidence of a difference against ABI + ADT in the ITT population.

Table 3.

PFS hazard ratios and SUCRA for ITT (random effects) and high volume (fixed effects) populations.

Treatment	ITT population (random effects)				High volume population (fixed effects)
	HR (95%CrI)	Rank number	SUCRA	Mean rank (95%Crl)	HR (95%CrI)	Rank number	SUCRA	Mean rank (95%Crl)
DAR + DOC + ADT	Reference	1	0.99	1.067 (0.997, 2.000)	Reference	1	1.00	1.014 (0.9967, 1.004)
ABI + ADT	0.68 (0.42, 1.07)	2	0.83	2.174 (1.000, 3.000)	0.56 (0.44, 0.72)	3	0.68	3.224 (2.000, 5.000)
APA + ADT	0.56 (0.32, 0.95)	3	0.70	3.077 (2.000, 5.000)	0.49 (0.35, 0.68)	5	0.50	4.475 (2.000, 6.000)
ENZ + ADT	0.42 (0.24, 0.73)	4	0.51	4.435 (3.000, 6.000)	0.60 (0.43, 0.85)	2	0.75	2.743 (1.998, 5.000)
DOC + ADT	0.40 (0.29, 0.57)	5	0.46	4.799 (3.000, 6.000)	0.43 (0.37, 0.50)	6	0.35	5.560 (4.000, 6.002)
REZ + ADT	0.33 (0.15, 0.71)	6	0.33	5.718 (3.000, 7.000)	0.52 (0.29, 0.92)	4	0.57	3.995 (2.000, 6.000)
ADT	0.27 (0.18, 0.40)	7	0.18	6.748 (6.000, 7.004)	0.26 (0.21, 0.32)	7	0.10	7.291 (6.997, 8.003)
SNA + ADT	0.14 (0.07, 0.28)	8	0.002	7.983 (7.997, 8.003)	0.23 (0.14, 0.38)	8	0.04	7.697 (7.000, 8.004)

Treatments are listed in the order of their SUCRA ranking in the ITT population. SUCRA values close to 1.0 are favourable. HR values below 1.00 favour reference (i.e., DAR + DOC + ADT). Values are highlighted in bold if the 95% CrI exclude 1.00.

ABI = abiraterone acetate; ADT = androgen deprivation therapy APA = apalutamide; DAR = darolutamide; DOC = docetaxel; ENZ = enzalutamide; HR = hazard ratio; ITT = intention-to-treat; PFS = progression-free survival; REZ = rezvilutamide; SNA = standard nonsteroidal antiandrogen; SUCRA = surface under the cumulative ranking curve.

Sensitivity analysis

In sensitivity analysis we simplified the evidence network by assuming SNA + ADT was equivalent to ADT to allow a more direct comparison with REZ + ADT, similar to a previous NMA¹. Broadly similar results were observed on OS and PFS comparing fixed and random effects models. Darolutamide plus DOC + ADT consistently ranked highest on all outcomes and across all populations. Model assessment statistics, along with complete cross tables, HRs, and SUCRA ranking are presented in Table S24 to Table S37.

Discussion

We compared the relative effectiveness on OS and PFS of DAR + DOC + ADT against licensed and approved therapies in the Chinese setting, including recently approved REZ + ADT, and considered both ITT and high-volume populations. Across both populations and outcomes, results numerically favoured DAR + DOC + ADT on all comparisons. There was strong evidence of a difference in OS against DOC + ADT, ADT, and SNA + ADT in ITT and high-volume populations, as well as APA + ADT in the high-volume population. On PFS, there was strong evidence of a difference of DAR + DOC + ADT against all other treatments including APA + ADT, ENZ + ADT, DOC + ADT, REZ + ADT, ADT and SNA + ADT in the ITT population, except ABI + ADT. Despite the statistical uncertainty on a number of comparisons on OS, as demonstrated by the overlapping 95% credible intervals, the SUCRA results provide a clear hierarchy of the best treatment choice: DAR + DOC + ADT consistently ranked as the highest-performing treatment across all outcomes and populations. With SUCRA values as high as 0.99 (ITT population) and 1.0 (high-volume population) on OS, for example, the DAR triplet therapy was strongly supported as the optimal treatment choice. Rezvilutamide plus ADT performed well next to other comparators in high-volume population, ranked second (with a SUCRA value of 0.71) and fourth (with a SUCRA value of 0.57) in OS and PFS, respectively. However, there remained a strong difference in favour of DAR + DOC + ADT on PFS, and a numerical advantage on OS. These findings were further supported by sensitivity analysis which assumed SNA + ADT was equivalent to ADT. These findings have a substantial clinical impact, reinforcing the known benefits of ARATs in delaying castration-resistant prostate cancer and prolonging life. Specifically, this research demonstrates that the triplet therapy of DAR + DOC + ADT consistently ranks highest in OS and PFS among key ARATs, providing clinicians and patients with a clear, first-line treatment option in mHSPC.

Despite the relatively small number of active therapies in mHSPC, the relative effectiveness of these different treatment combinations has been the subject of numerous ITCs in recent years. Rather than contributing to any potential overlap and duplication, the findings from this analysis are unique in that we included evidence on REZ + ADT, which is a relatively new therapy that has only been approved for use in China to date¹⁷. This is the first NMA to consider the efficacy of systemic therapies in both ITT and high-volume populations, to the best of our knowledge. Unlike a number of previous NMAs in mHSPC, which relied on Bucher comparisons and frequentist methods, we applied Bayesian NMA to establish relative effectiveness, which is internationally recommended methodology for ITC, endorsed by NICE in the UK, ISPOR, and the WHO^19–22. Unlike frequentist methods that rely on inferential statistics, the Bayesian NMA approach offers a unified and flexible framework for analysing complex evidence networks, combining both direct and indirect evidence in the same network. Through probabilistic rankings, results are clinically intuitive and often more useful for decision-making than traditional p-values and confidence intervals.

Our findings may therefore be considered more robust than those from a frequentist comparison. Our findings also align with other Bayesian NMAs in mHSPC^26,27. Similar to Zhou and colleagues, we found DAR + DOC + ADT ranked first for improvement in OS among available systemic therapies for high-volume mHSPC, followed by REZ + ADT (ranked third by Zhou and colleagues, after ABI + DOC + ADT; the triplet therapy was not considered in this analysis as it is not widely used in China). We similarly found DAR + DOC + ADT ranked first on PFS, ahead of ENZ + ADT and ABI + ADT, which were similarly ranked behind other triplet therapies by Zhou and colleagues, along with REZ + ADT²⁷. In the comparison of triplet therapies by Dou and colleagues, DAR + DOC + ADT emerged as superior on OS, similar to this analysis. Unlike this analysis, which focussed on ITT and high-volume populations, Dou and colleagues compared OS in patients with a high Gleason score ≥ 8, ECOG 0, and ECOG 1. Combined, the findings support the continued use of DAR + DOC + ADT in mHSPC, particularly among specific subgroups, such as those with high-volume disease and a high Gleason score ≥ 8²⁶.

Although there has been some duplication of NMA in mHSPC, studies in this area have largely varied due to differing terms of review aims, eligibility criteria and included data, statistical methodology, reporting and inference, as highlighted by Fisher²³. These variations in study design and research aims have made it difficult to meaningfully establish best performing treatments. Added to this, a number of studies have pooled the effect of ARAT treatments in ITCs, rather than establishing the effectiveness of individual treatment combinations. Ciccarse and colleagues merged ABI and DAR as androgen receptor signalling pathway inhibitors (ARSi) and compared them in combination with DOC + ADT to DOC + ADT⁴⁷. The authors found a clear survival advantage for the addition of ARSi. Roy and colleagues separately compared triplet of ADT, ARAT, and DOC with ADT + ARAT⁴⁸. The authors found non-significant OS benefit with triplet therapy, while ADT + DOC and ADT alone were associated with an increased risk of death. While the evidence is sometimes inconsistent, there is strong evidence that triplet therapies provide more favourable survival benefits than doublet therapies in mHSPC, while DAR + DOC + ADT performs better than other ARATs in mHSPC, as supported by this analysis.

We acknowledge there are certain limitations associated with this work. Though the proposed method of Vale and colleagues to account for correlations between the arms of the STAMPEDE MAMS trials is possible for the ITT analyses, it was not possible for the high-volume subgroup analyses because correlation between STAMPEDE arms is unknown in subgroups⁴². We therefore assumed proportional hazards. For some comparisons, we had to assume that the number of common events on the common treatment arms was equal to the number of events in the earlier trial. This was because the STAMPEDE authors had not reported these overlap event counts and Vale and colleagues was published after the later STAMPEDE publications. As we could not adjust for correlations in the high-volume population, the precision of the comparisons involving the STAMPEDE arms (i.e., DOC + ADT and ABI + ADT) may be overestimated, so these results should be interpreted with some caution.

Heterogeneity in outcome definition on PFS existed across trials. In the ITT population, we accounted for such heterogeneity by selecting random effects model; however, the results still numerically favoured DAR + DOC + ADT, with strong evidence of a difference observed on all comparisons, except a comparison with ABI + ADT in the ITT population. We acknowledge that combining disparate endpoints (on PFS) may be problematic as clinical events, such as biochemical progression/radiographic progression/composite failure-free survival, may occur at different stages of the disease and may not be interchangeable. The combination of these endpoints may violate the transitivity assumption, so relative effects may not be reliably estimated. However, a range of different endpoints have been used across trials so limiting comparisons of PFS to strict outcome definitions would drastically reduce the scope of the evidence networks for a comparison of PFS. For instance, only three trials (ARCHES, LATITUDE, and TITAN) relied on rPFS to measure disease progression, so comparisons would be limited to ENZ + ADT, ABI + ADT, and APA + ADT (and REZ + ADT [CHART] if SNA + ADT was assumed equivalent to ADT). Two trials relied on cPFS (ENZAMET and CHAARTED) while one trial each used FFS with biochemical progression (STAMPEDE) and castration-relapse or death (ARASENS). As limiting the evidence networks to trials and treatments with strictly comparable outcome definitions on PFS was not feasible, we performed both fixed and random effects models, with random effects models chosen in the base case analysis where applicable due to the heterogeneity in outcome definition. We found results were broadly comparable across both models, suggesting some reliability in estimated relative effects.

A further limitation is that comparisons were limited to treatments currently used in China. As such, we excluded ABI + DOC + ADT on the basis that ABI + ADT is more widely used in its place in clinical practice in China³¹. The overall ranking of treatments may have been different if ABI + DOC + ADT were included in the NMA. However, a previous NMA comparing triplet therapies found DAR + DOC + ADT ranked higher than ABI + DOC + ADT on OS⁴⁹. A comparison on PFS was not included in the NMA so it is unclear how its inclusion might have affected the rankings presented here. Further research is warranted to assess the relative effects of the DAR triplet therapy against the ABI triplet therapy on PFS in other settings where ABI + DOC + ADT may be considered a relevant comparator.

While our analysis focused on the Chinese setting, where REZ + ADT is approved for high-volume disease, the results may still be generalizable to other countries. This is supported by the NMA’s use of published Phase III trial data from a well-defined patient population. However, a key limitation remains the lack of widespread clinical data on REZ in diverse, non-Chinese populations. Although the CHART trial was multinational, the vast majority of participants were Chinese, meaning its efficacy and safety outside of this primary trial population may require further evaluation. Lastly, while we considered both overall (ITT) and high-volume patient populations in this paper, the impact of patient-specific characteristics on treatment outcomes can affect the optimal treatment choice (Fisher et al. 2025), suggesting a need for further research in other populations to better personalize treatment and optimize outcomes for patients with mHSPC.

Conclusion

Results numerically favoured DAR + DOC + ADT on OS and PFS across all comparisons, with statistical significance demonstrated against APA + ADT (in the high-volume population only), DOC + ADT, ADT, and SNA + ADT on OS and all comparators on PFS (excluding ABI + ADT in ITT population only, which was not statistically significant). Reinforcing the known benefits of ARATs in delaying castration-resistant prostate cancer and prolonging life, these findings support the continued use of DAR + DOC + ADT as frontline treatment for patients with mHSPC, particularly in China where REZ was recently approved. The findings have implications for other settings where REZ may be approved in the future.

Supplementary Information

Below is the link to the electronic supplementary material.

Author contributions

All authors contributed to the study conception and design. Material preparation, data collection and analysis were performed by Philip Orishaba, Howard Thom, and Noman Paracha. The first draft of the manuscript was written by Christopher G Fawsitt and all authors commented on previous versions of the manuscript. Haiyin Wang was responsible for main conceptual ideas, review and quality assurance. All authors read and approved the final manuscript.

Funding

This work was supported by Bayer Pharmaceuticals, Basel, Switzerland.

Data availability

The datasets generated during and/or analysed during the current study are available from the corresponding author on reasonable request.

Declarations

Competing interests

Financial interests: HT owns shares in the consulting company Clifton Insight, and PO and CF are employees of Clifton Insight, which has received fees from Amicus, Baxter, Bayer, Daiichi-Sankyo, Eisai, Lundbeck, Merck, Novo Nordisk, Pfizer, Roche, and UCB. RX and YZ are employees of the Bayer Healthcare Company Ltd. Non-financial interests: HW and NX declare that they have no conflict of interests. All the remaining authors declare no conflict of interest.