According to the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) specification, a systematic search was conducted for RCT studies published prior to and including July 01, 2022. Databases searched included Medline, Embase, Cochrane Central, and ClinicalTrials.gov. The main search terms included prostate cancer, hormone sensitive, clinical trial, docetaxel, abiraterone, enzalutamide, apalutamide, darolutamide, radiotherapy, etc. Different combinations and strategies were applied during different database retrievals ( Appendix P 1-4 ). Only the results of RCTs published in English were included in this article, and a search was performed before the final statistical analysis to avoid omitting the latest published results.

The literature was screened according to the principles of Participants, Interventions, Comparisons and Outcomes (PICO). Inclusion criteria were as follows: study population: 1) Adult patients with mHSPC, aged ≥18 years; 2) Patients with initial onset or those who progressed after previous local therapy; 3) The duration of ADT treatment allowed in the stage of localized prostate cancer was no more than three years, and in the stage of metastatic prostate cancer was no more than six months; 4) Data from phase III RCTs. Interventions: 1) First-line treatments for mHSPC recommended in the latest guidelines of EAU and NCCN, including ADT combined with docetaxel, abiraterone acetate, enzalutamide, apalutamide, and RT+ADT (for LVD patients); 2) Recently reported triplet therapies, mainly NHAs+DOC+ADT. Control group: 1) According to the design of different studies, treatments in control groups included ADT alone, ADT plus placebo, or ADT plus traditional nonsteroidal antiandrogens (NSAA) (27, 28); 2) In studies with triplet therapies, the treatment in control groups was DOC+ADT. Outcomes: The main outcome was OS, and the secondary outcomes were progression-free survival (PFS) and safety indicators of high-grade AEs. Exclusion criteria: 1) Patients with metastatic or nonmetastatic castration- resistant prostate cancer; 2) Nonrandomized controlled studies or phase I/II RCTs; 3) Research that could not extract relevant data; 4) Duplicate studies.

The methodological quality of each study was independently assessed by two independent investigators using the revised Cochrane Collaboration Risk of Bias tool RoB1.0, and differences of opinion were resolved by consultation. The overall bias of the trial was assessed for each study from the following seven domains: random sequence generation (selection bias), allocation concealment (selection bias), blinding of participants and personnel (performance bias), blinding of outcome assessment (detection bias), incomplete outcome data (attrition bias), selective reporting (reporting bias), and other biases. The study as a whole was deemed as having a “low risk of bias” if all terms were at low risk of bias. The study as a whole was judged to be at high risk of bias if either item was judged to be at high risk of bias. Other conditions were judged to be at moderate risk of bias. We also assessed the certainty of evidence using the Confidence in Network Meta-Analysis framework (CINeMA).

Two independent researchers followed a preplanned search strategy that required each researcher to perform his research separately according to PRISMA guidelines (29). When multiple papers reported results from the same study at different stages, data was only extracted for the most recent results. Any discrepancies were resolved by consensus. Two investigators used predesigned forms for data extraction and recording. The characteristics of the included patients, including ECOG performance score, age, PSA value, Gleason score, visceral metastasis, tumor burden, proportion of patients with primary metastasis, previous ADT use, and previous/concurrent use of docetaxel, were recorded in detail. In terms of related efficacy results, HR values for OS and PFS were recorded, as well as HR values for different tumor burden subgroups. For safety outcomes, the number of patients with high-grade AEs were recorded, and OR values were calculated.

Three similar indicators related to PFS were used in different studies, including PFS (12, 13, 18, 30–32), radiographic progression-free survival (rPFS) (11, 14, 16, 17, 25, 33–36), and clinical progression-free survival (cPFS) (10, 11, 15). Since the three were similar in definition, and in most cases the imaging progression was earlier than the aggravation of clinical symptoms and death, we unified the three into one “generalized PFS” in the present meta-analysis (7). Traditional nonsteroidal antiandrogens (NSAA) have long been considered to result in no or very little improvement in OS in patients with metastatic prostate cancer (27, 28); therefore, we considered NSAA+ADT as the same treatment as ADT alone to pool more data in the present review. In the subgroup analysis, we focused on tumor burden (high- versus low-volume disease), as most RCTs suggested that tumor burden was a feature that influenced the efficacy of interventions. Following the CHAARTED study criteria (11, 36, 37), high volume disease (HVD) was defined as the presence of visceral metastases and/or four or more bone metastases, with at least one bone metastasis located outside the spine and pelvis. We also made subgroup analyses for de novo patients and patients with previous local treatment.

The network meta-analysis was performed using the “gemtc” and “rjags” packages of the R 4·0·5 software using the Monte Carlo Markov Chain (McMc) method in a Bayesian framework (38, 39). A network diagram was drawn for each intervention. The original hazard ratio (HR) values and 95% confidence intervals (CIs) of each RCT study (or their subgroup data) were extracted, and the consistency model was used (40). The number of model chains was 3, the initial value was 2·5, the sampling number was 10,000, and the number of iterations was 100 000 with a step size of 10. For closed-loop studies with both direct evidence and indirect evidence, the node-splitting model was used to test the consistency of closed-loop studies, and P<0·05 was considered statistically significant. The “mtc.anohe” command in the “gemtc” package was used to assess overall heterogeneity and was recorded with the variance parameter I². Heterogeneity was assessed visually using forest plots and I² statistics. If the heterogeneity was large (I²>50%, P<0·1), the source of heterogeneity was further analyzed.

The trace plotting method and density plotting method were used to evaluate the convergence of the model. The rank probability was calculated to determine the level of superiority and inferiority of different treatments according to P scores (41). We extracted the “number of patients with grade 3-5 AEs/total patients” in each group and calculated the odds ratio (OR) and 95% CI.

A total of 5138 articles and 169 clinical trials were retrieved from the literature search, and 68 of them were read in full text after browsing the titles and abstracts. Finally, ten multicenter phase III randomized controlled trials (21 references) (10–18, 25, 26, 30–37, 42, 43) were included in the analysis ( Figure 1 ). A total of 12,298 patients were included, with a median number of patients of 1,069 (range, 385-2061), median follow-up time of 44·6 months (range, 34-84 months), and publication years between 2013 and 2022. The ten trials involved a total of nine treatments, including ADT alone, DOC+ADT, abiraterone and prednisone(AAP)+ADT, ENZA+ADT, APA+ADT, RT+ADT, ENZA+DOC+ADT, AAP+DOC+ADT, and DARO+DOC+ADT. For OS, the overall risk of bias was low in five trials (CHAARTED, STAMPEDE-arm C, TITAN, HORRAD and ARASENS), while the remaining trials raised some concerns. All trials raised some concerns regarding generalized PFS except for the TITAN trial ( Appendix P 5-6 ).

The STAMPEDE study (12, 13, 18, 30, 31) allowed the recruitment of localized high-risk patients, and we only extracted data from mHSPC patients in this analysis. In the multiarm, multiphase STAMPEDE study, there was a period of overlap in the recruitment timing of arm C and arm G; thus data from this period could be used to directly compare AAP+ADT and DOC+ADT (32). The PEACE-1 study (25) used a 2 × 2 factorial design for newly diagnosed mHSPC patients. Only the data of the docetaxel population was extracted, and the triplet therapy “AAP+DOC+ADT” and the doublet therapy “DOC+ADT” were compared. The baseline characteristics of all included studies are listed in Table 1 and Appendix (P 7-8).

There was a certain proportion of patients with “early planned docetaxel used” in ENZAMET (15), ARCHES (16, 42), TITAN (17, 33) and STAMPEDE arm H (18) trials. There were obvious differences in HR values for OS between the docetaxel population and the population that did not use docetaxel ( Table 2 ), which suggested that the treatment sequence of docetaxel followed by NHAs may limit the effectiveness of NHAs (15, 16, 18, 33, 42). Thus, in our review, we strictly evaluated the impact of “planned docetaxel use” on the efficacies of NHAs or radiotherapy in relevant studies (15, 16, 18, 33, 42). We strictly excluded data from the docetaxel population in effectiveness assessment, unless data extraction was difficult in some subgroups. In the ENZAMET and ARCHES studies, 44·7% and 17·8% of patients received “early planned docetaxel chemotherapy”, respectively, and these patients actually received triplet therapy. We extracted data from these docetaxel populations and made an exploratory analysis on the comparative efficacy of triplet therapy of “ENZA+DOC+ADT” and doublet therapy of “DOC+ADT”. For similar data of the docetaxel population in the TITAN and STAMPEDE arm H studies, we did not conduct a similar analysis due to the small number of patients and incomplete data ( Appendix P 9 ).

Figure 2 shows a network diagram comparing different treatment options. Only data from 342 mHSPC patients from arms C and G of the STAMPEDE study, a multiarm, multistage RCT, were extracted for a head-to-head comparison (32). Consistency tests for this closed-loop comparison suggested that p values were all >0.1 ( Appendix P 11 ). The results of the heterogeneity test indicated that I^{2 =} 60.1% for the comparison of abiraterone and docetaxel, and the heterogeneity was derived from the direct comparison between the two agents ( Appendix P 12-18 ).

The STAMPEDE study had four comparison groups, arm C (12), arm G (13), and arm H (18), and a head-to-head subgroup (32). While in both ENZAMET (15) and ARCHES (16, 42) studies, independent analyses of the “docetaxel population” were performed. Therefore, a total of 15 comparison groups were obtained from the ten trials ( Appendix P 9 ), and the data of these 15 groups were all included in the network meta-analysis of OS. The forest plot demonstrated a comparison with ADT alone or DOC+ADT and is shown in Figures 3A, B .

Except for radiotherapy (HR 0·92, 95% CI 0·76-1·11), the other seven treatment options were all significantly better than ADT in terms of OS benefits. Compared with docetaxel chemotherapy, only the triplet therapies of DARO+DOC+ADT (HR 0·68, 95% CI 0·53-0·88) had significant OS benefits. Additionally, the triplet therapy of ENZA+DOC+ADT was not superior to ENZA+ADT (HR 1·07, 95% CI 0·70-1·64). The detailed HR data of the comparison between each two treatment groups can be found in Appendix (P 21-24). The ranking and ranking probability of competing interventions are shown in Appendix (P 25-32). The triplet treatment DARO+DOC+ADT had the highest probability of ranking the best. The convergence of the models was well according to trace and density plots ( Appendix P 33-34 ).

In addition to the HORRAD (43) and ARASENS (26) trials, 13 of the 15 comparison groups reported results of PFS or similar indicators ( Appendix P 9 ). Specifically, six comparison groups (GETUG-AFU15 (10, 36), LATITUDE (14, 35), ARCHES (16, 42) and its docetaxel subgroup, TITAN (17, 33), and PEACE-1 (25) reported results of rPFS, four comparison groups (STAMPEDE arm C (12), arm G (13), arm H (18), and AAP-DOC head-to-head comparison subgroup (32) reported results of PFS, and cPFS was reported in four comparison groups (GETUG-AFU15 (10), CHAARTED (11), ENZAMET (15), and ENZAMET’s docetaxel subgroup). We unified these indicators as a “generalized PFS” (We chose rPFS, rather than cPFS, in GETUG-AFU15 for subsequent comparison).

The results showed that except for radiotherapy (HR 0·96, 95% CI 0·80-1.16), all other measures were significantly better than ADT ( Figures 3C, D ). In addition, except for ADT (HR 1·50, 95% CI 1·32-1·72) and radiotherapy (HR 1·44, 95% CI 1·14-1·83), which were inferior to docetaxel chemotherapy, other treatments were all superior to docetaxel chemotherapy. The ranking probability and HR values for the pairwise comparisons of each treatment measure can be found in Appendix (P 21-32). The triplet therapy ENZA+DOC+ADT had the highest probability of ranking the best. Besides, we also made respective analyses for rPFS, PFS and cPFS (The forest plots can be seen in Appendix P 20 ).

OS data of high- or low- volume subgroups were extracted from 11 of the 15 comparison groups and 8 treatments were included ( Appendix P 9 ). The analysis results of mHSPC with high volume disease (HVD) showed that radiotherapy (HR 1·07, 95% CI 0·76-1·51) had no benefit compared with ADT, the benefit of ENZA+DOC+ADT (HR 0·71, 95% CI 0·41-1·22) had no significant difference compared with ADT, while the other five treatments all had significant benefits compared to ADT ( Figure 3E ). In terms of possible ranking, the best was AAP+DOC+ADT (HR 0·52, 95% CI 0·33-0·83), followed by AAP+ADT (HR 0·61, 95% CI 0·47-0·79). The results of low tumor disease (LVD) showed that APA+ADT (HR 0·52, 95% CI 0·26-1·05) and ENZA+ADT (HR 0·53, 95% CI 0·31-0·88) possibly ranked the highest ( Figure 3F ). However, it has to be emphasized that ARASENS trial did not stratify the patients according to disease volume.

Only six studies reported HR data in patients with visceral metastases, involving five interventions ( Appendix P 9). The results suggested that DARO+DOC+ADT (HR 0·42, 95% CI 0·13-1·34) had the highest probability of ranking the best, while the new antiandrogens, especially enzalutamide (HR 1·11, 95% CI 0·58-2·09), performed less well. ( Figures 3G and Appendix P 25-32 )

In terms of safety indicators, we selected grade 3-5 AEs, as there were more studies reporting grade 3-5 AEs than serious adverse events (SAEs). In particular, two important trials related to triplet therapies, PEACE-1 (25) and ARASENS (26), both reported grade 3-5 AEs. The CHAARTED study (11) reported grade 3-5 AEs in the docetaxel group, but corresponding data in the control group were lacking; thus, these data were not used. No data from GETUG -AFU15 (10, 36) were used, as they reported a different safety indicator of “AE events/person month”. Finally, we included a meta-analysis of grade 3-5 AEs from nine studies ( Appendix P 45 ) involving eight different interventions.

The results showed that the risk of grade 3-5 AEs in triplet therapies was relatively high, especially that of AAP+DOC+ADT ( Figure 3H ). The order of security was: ADT alone (comparator) > RT+ADT (1·03, 0·57-1·87) > APA+ADT (1·39, 0·76-2·55) > ENZA+ADT (1·81, 1·18-2·77) > AAP+ADT (1·92, 1·27-2·95) > DOC+ADT (2·28, 1·26-4·16) > DARO+DOC+ADT (2·59, 1·11-6·09) > AAP+DOC+ADT (3·56, 1·51-8·43).

The certainty of the evidence for outcomes of interest as measured with CINeMA varied from high to very low. Full information on CINeMA is described in the appendix ( Appendix P 46–64 ).