A total of 394 articles were retrieved originally after the electronic literature search. After screening the titles and abstracts, 303 articles were excluded resulting in 91 articles that were potentially relevant for inclusion into the study. An additional 68 articles were excluded from 91 articles for the following reasons: duplicate date, preliminary meeting reports, and not reported/could not obtain usable data that can help infer the results in the current article. As a result, 23 RCTs with 29908 patients that fulfilled the inclusion criteria were identified. (Fig A.1) Among these, 19 trials were phase III, and 4 trials were phase II RCTs. The baseline characteristics are listed in Table A.1. In addition, 7 trials investigated the role of new agents in combination with docetaxel in chemotherapy-naive CRPC patients. The study (MP Wirth (2004)) included 3 trials, and hence, we calculated it as 3 independent trials in our analysis. All trials were reported in full-text and most of trials had a low risk of bias (Fig A.2).5, 7–13, 18–32
In the OS analysis, 22 trials were included. Of these, 7 investigated the new agents combined with docetaxel, and 15 trials investigated the new agents alone. Figure 1A shows the relative comparisons analyzed in the network. HR and the 95% CI were explicitly reported in all the published articles. Compared to the placebo, both random-effects Bayesian and frequentist network meta-analyses found that only enzalutamide (HR = 0.71, 95% CI: 0.52–0.95, HR = 0.71, 95% CI: 0.52–0.95, respectively) and abiraterone (HR = 0.73, 95% CI: 0.58–0.89, HR = 0.73, 95% CI: 0.62–0.85, respectively) had a significant impact on OS (Fig. 2A). In addition, the results of two frame networks were similar. Combined with the convergent and lever graph, the network showed consistent results (Fig A.3). Although none of the trials compared the efficacy between enzalutamide and abiraterone, we used our network to explore the comparison. As shown in Fig. 2A, the indirect comparison of HR was 0.99 (95% CI: 0.72–1.40) with no significance. Similarly, no significant outcome was detected in Bayesian and frequentist frame network meta-analysis comparing enzalutamide and abiraterone (HR = 0.97, 95% CI: 0.68–1.40, HR = 0.98, 95% CI: 0.74–1.30, respectively); the P-value in both treatments were similar.
In the subgroup analysis, only 3 articles analyzed the subgroup of age (< 75, ≥ 75), ECOG score (0,1), and Gleason score (≤ 7, ≥ 8), respectively. Therefore, a network was created in these subgroups, which provided interesting results. As shown in Fig. 3, both enzalutamide and abiraterone had significant improvements in the OS in patients with age < 75 years. However, only enzalutamide had the same trend in patients with age ≥ 75 years. Similar results were observed in the ECOG score subgroup, and the P-scores of enzalutamide were higher than abiraterone in all comparisons. In the Gleason score subgroup, only enzalutamide, ipilimumab, and orteronel were investigated for efficacy without abiraterone. The results indicated that enzalutamide could improve the OS significantly in both Gleason ≤ 7 and Gleason ≥ 8 subgroups.
All new agents in combination with docetaxel did not show any significant improvements in the OS in network meta-analysis when compared to the placebo and docetaxel (Fig A.4A). The results also indicated that chemotherapy-naïve CRPC patients had to postpone the initiation of cytotoxic chemotherapy. For the median time to the initiation of cytotoxic chemotherapy and the median time to decline of the FACT-P global score, there were 5 and 4 novel agents, respectively, involved in the analysis. Enzalutamide, abiraterone, ipilimumab, and tasquinimod could improve the median time to chemotherapy, while enzalutamide and bicalutamide could improve the median time to decline of the FACT-P global score (Fig A.5).
In the PFS analysis, 16 trials were included. Of these, 5 investigated the new agents combined with docetaxel, and 11 trials investigated the new agents alone. Figure 1B showed the network of comparisons. A total of 14 trials provided explicit HR and 95% CI. The remaining 2 trials only have the value for the median time to PSA progression, which was used as a substitute. Compared to the placebo, both Bayesian and frequentist random-effects network meta-analyses suggested that enzalutamide (HR = 0.21, 95% CI: 0.11–0.38, HR = 0.20, 95% CI: 0.15–0.29, respectively) and abiraterone (HR = 0.30, 95% CI: 0.13–0.71, HR = 0.30, 95% CI: 0.18–0.49, respectively) had significant improvements in PFS (Fig. 2B). The results of our analysis were consistent (Fig A.6). The results of the two frame networks were similar. Also, bicalutamide had significant benefits in PFS in frequentist network, and the other new agents did not present any clinical value. As shown in Fig. 2B, the indirect comparison of HR between enzalutamide and abiraterone was 0.57 (95% CI: 0.46–0.70). However, the differences were insignificant in both Bayesian and frequentist network meta-analysis (HR = 0.69, 95% CI: 0.24–2.00, HR = 0.68, 95% CI: 0.37–1.20, respectively). Furthermore, the P-score of enzalutamide was 0.98, and that of abiraterone was 0.84.
All new agents in combination with docetaxel did not show any significant improvements in PFS in Bayesian network meta-analysis. Only bevacizumab combined with docetaxel showed a significant benefit in PFS in frequentist network meta-analysis (HR = 0.80, 95% CI: 0.71–0.91) (Fig A.4B).
In rPFS analysis, 8 trials were included. However, none investigated the new agents combined with docetaxel. Figure 1C showed the network of comparisons. Eight trials provided explicit HR and 95% CI. Compared to the placebo, both network meta-analyses suggested that enzalutamide (HR = 0.21, 95% CI: 0.11–0.38, HR = 0.20, 95% CI: 0.15–0.29, respectively), abiraterone (HR = 0.30, 95% CI: 0.13–0.71, HR = 0.30, 95% CI: 0.18–0.49, respectively), and bicalutamide (HR = 0.30, 95% CI: 0.13–0.71, HR = 0.30, 95% CI: 0.18–0.49, respectively) effectuated significant improvements in rPFS (Data were shown in Fig. 4A). Moreover, tasquinimod and orteronel had a positive outcome in rPFS only in frequentist network. Different from the above results, enzalutamide was more advantageous than abiraterone in rPFS in indirect comparison and Bayesian and frequentist network meta-analysis (HR = 0.69, 95% CI: 0.24–2.00, HR = 0.68, 95% CI: 0.37–1.20, HR = 0.68, 95% CI: 0.37–1.20, respectively). The P-score of enzalutamide was 1.00, and the score of abiraterone was 0.67.
As most trials reported that grade ≥ 3 AEs, a network meta-analysis was performed. As shown in Fig. 1D and 4B, compared to the placebo, ipilimumab and tasquinimod led to more AEs in both models. Additionally, abiraterone and orteronel led to more AEs in frequentist network. However, no significant differences were detected between enzalutamide and abiraterone (Fig. 4B). We summarized the rankings of eight novel treatments strategies in terms of OS and AEs based on P-values (Fig. 5). Enzalutamide and abiraterone had similar benefits/harm and ranked as the best and second-best treatment, respectively, while. Ipilimumab and orteronel were ranked as the worst and second worst in terms of AEs.