Efficacy of Abiraterone and Enzalutamide in Pre- and Postdocetaxel Castration-Resistant Prostate Cancer: A Trial-Level Meta-Analysis

We examined the comparative efficacies of first-line abiraterone and enzalutamide in pre- and postdocetaxel settings in castration-resistant prostate cancer (CRPC) through a trial level meta-analysis. A mixed method approach was applied to 19 unique studies containing 17 median overall survival (OS) estimates and 13 median radiographic progression-free survival (PFS) estimates. We employed a random-effects meta-analysis to compare efficacies of abiraterone and enzalutamide with respect to OS and PFS. In the predocetaxel setting, enzalutamide use was associated with an increase in median OS of 5.9 months (p < 0.001), hazard ratio (HR) = 0.81, and an increase in median PFS of 8.3 months (p < 0.001), HR = 0.47 compared to abiraterone. The advantage of enzalutamide improved after adjusting for baseline Gleason score to 19.5 months (p < 0.001) and 14.6 months (p < 0.001) in median OS and PFS, respectively. In the postdocetaxel setting, the advantage of enzalutamide use was nominally significant for median PFS (1.2 months p = 0.02 without adjustment and 2.2 months and p = 0.0007 after adjustment); there was no significant difference in median OS between the two agents. The results from this comprehensive meta-analysis suggest a survival advantage with the use of first-line enzalutamide over abiraterone in CRPC and highlight the need for prospective clinical trials.


Introduction
Prostate cancer is the most common cancer diagnosis in men and is projected to account for more than 160,000 new diagnoses in the United States in this year [1]. While localized prostate cancer has excellent prognosis, castrationresistant prostate cancer (CRPC) is uniformly lethal after a period of about 1-3 years [2]. Castration resistance represents the cumulative result of escape mechanisms deployed by the tumor to overcome androgen deprivation [3], clinically manifested by biochemical or radiographic progression despite castrate levels of serum testosterone [4]. In recent years, the FDA approvals of several new therapeutic agents for CRPC, such as the androgen receptor signaling axistargeting agents, abiraterone [5] and enzalutamide [6], have transformed the clinical management of advanced prostate cancer. Despite these advancements, the improvements in survival offered by these therapies are modest, on the order of several months, given the rapid and inevitable emergence of resistance. The issue of resistance is particularly relevant in CRPC with the rise of cross-resistance between abiraterone and enzalutamide [7][8][9][10], as well as between these agents and taxane therapies [11].
The availability of multiple CRPC therapies necessitates an understanding of optimizing the sequence in which these therapies are deployed, an area of major investigative effort [12,13]. Abiraterone and enzalutamide were initially approved for use in the postdocetaxel setting [14,15], with expanded indications for docetaxel-naïve CRPC [16][17][18]   and convenience of outpatient therapy compared to taxanes, abiraterone and enzalutamide are often deployed early in the CRPC disease course, with much of the decision-making as to which agent should be initiated, deferred to oncologist experience, toxicity profile, and patient preference. To date, there have been no direct prospective investigations comparing the efficacies of abiraterone and enzalutamide in CRPC patients, with only a few retrospective analyses [19] of single institution experiences.
Given that the emergence of cross-resistance and unique selective pressures exerted by abiraterone and enzalutamide may influence the efficacies of successive treatments, there is a need to determine whether these two therapies are truly interchangeable entities or whether there are differences in survival outcomes. We thus conducted a comprehensive triallevel meta-analysis to examine the comparative efficacies of abiraterone and enzalutamide in the first-line CRPC (predocetaxel) and second-line CRPC (postdocetaxel) settings by utilizing the published literature.

Methods
The primary goal of this study is to compare median overall survival (OS) and median radiographic progression-free survival (PFS) of abiraterone and enzalutamide in both pre-and postdocetaxel settings. The treatment sequences examined were first-line abiraterone (denoted as AX) and first-line enzalutamide (EX) in the predocetaxel setting and docetaxelto-abiraterone (DA) and docetaxel-to-enzalutamide (DE) in the postdocetaxel setting. The median time-to-event estimates are defined as the time from the start of the treatment of interest, abiraterone, or enzalutamide, to the time at which 50% of the subjects in the study group have reached the outcome (death for OS, and disease progression for PFS). Disease progression is determined by radiographic evidence based on the RECIST [20] and PCWG2 [21] criteria.
Using the Entrez [22] package for Python, PubMed and Web of Science were queried initially for "abiraterone AND median AND survival" and "enzalutamide AND median AND survival" with a date range from January 1, 2008, to November 1, 2016 ( Figure 1). Querying through "median" estimates allowed us to maximize the number of suitable studies. Many studies had only a single cohort that met our inclusion criteria (described below), which may have been omitted under other queries such as "hazard ratio." A total of 265 peer-reviewed articles, in English, were curated. Studies were excluded if any of the following conditions were met: (1) greater than 20% of the study population utilized a prior therapy with another androgen deprivation agent (e.g., a DA study in which greater than 20% of patients had prior enzalutamide use); (2) no confidence intervals or bounds were reported for the median estimates; (3) disease progression in a PFS study was not evaluated through radiographic evidence; studies with composite radiographic and PSA progression estimates were also excluded. For studies with a series of publications, we use the most upto-date estimates in our analyses. To assess the potential of publication bias, we generated funnel plots of effect against precision for all treatment sequences.
In total, 6 clinical trials and 18 nonclinical trials (Table 1) qualified for inclusion using our criteria: 17 cohorts provided median OS estimates while 13 cohorts provided median PFS estimates. Of the 17 median OS estimates, 3 were for the AX sequence and 2 were for EX, and 10 were for DA and 2 were for DE. Of 13 cohorts with median PFS estimates, 3 were for the AX sequence and 4 were for EX and 3 were for DA and 3 were for DE ( Figure 1). Available confidence intervals or bounds and sample sizes were also recorded.
Most, but not all, of these studies also reported summary baseline characteristics such as race, mean or median age, proportion of patients with a Gleason score ≥ 8, and median baseline PSA score (ng/mL) ( Table 1). In addition to the The studies that provided two cohorts for our analyses are denoted ( * ) and correspond to those in Figures 2 and 3. Race is the predominant proportion within each cohort and age is either a mean or median measure. Blank cells indicate a lack of reporting in that category.
median time-to-event estimates, most studies we considered reported the proportion of patients with PSA decline ≥ 50% (Table 1). Unequal variance -tests were performed for these characteristics ( Table 2). Due to a lack of access to individual-level data, a trial-level meta-analysis was conducted using the rma.mv() function in the metafor package of R software [23]. Heterogeneity within treatment sequence groups ( 2 > 80% except for the DA sequence with respect to PFS) suggested a mixedeffects model when combining effect sizes. The median timeto-event estimate from each study was weighted by its inverse variance, where the variance was calculated from the reported 95% confidence interval (CI) as (UB − LB)/3.92, where UB and LB are the upper and lower bounds of the CI, respectively. Six studies only reported a lower bound presumably because early censoring prevented estimation of the upper bound; in these cases, the variance was calculated as (median-LB)/1.96. The rma.mv() function in the metafor package accounts for nonindependence in observed effects as we analyzed more than one median estimate from several studies For each of the two time-to-event outcomes (OS and PFS), a hazard ratio (HR) between two treatment sequences was calculated as the inverse ratio between their median survival estimates. It has been shown [24] that there is a high concordance between HR and inverse median survival ratio. In fact, under the exponential survival model (i.e., constant hazard function model), HR equals the inverse ratio of median survival time.

Results
As described in Methods and Figure 1, 265 publications were manually assessed and 24 cohorts met our inclusion criteria. Although we cannot rule out publication bias or lack thereof due to the limited number of studies per group, we do not discern any bias in the funnel plots (Supplemental S1 in Supplementary Material available online at https://doi.org/10.1155/2017/8560827).
The baseline characteristics-age, proportion with Gleason score ≥ 8, and baseline median PSA score-were comparable between AX and AE and between DA and DE for   both OS and PFS analyses ( Table 2). As expected, there was a negative correlation between the proportion of patients with a Gleason score ≥ 8 and median baseline PSA score for both OS ( = −0.22, = 0.45) and PFS ( = −0.44, = 0.28) (Supplemental Figures S2 and S3). The DE cohorts had a significantly higher proportion of patients with PSA decline ≥ 50% than the DA cohorts for both OS and PFS ( Table 2).
The EX cohorts also had a higher proportion of patients with PSA decline ≥ 50% than the AX cohorts, and the difference was significant for PFS ( Table 2). The PSA advantages of enzalutamide over abiraterone are consistent with our main results with respect to median survival. The combined estimates for median OS for all four treatment sequences are shown in Figure 2. The median OS  Figure 3. The patterns are similar to those for OS. The median PFS for EX was 15.8 months (95% CI 14.3-17.2), while that for AX was 7.4 months (95% CI 6.2-8.7), showing a significance difference of 8.3 months between the sequences ( < 0.0001). The corresponding HR was 0.47. The advantage of EX is also increased after adjusting for the baseline Gleason score, representing a 14.6-month improvement compared to the AX group ( < 0.001) ( Table 3). There was a nominally statistical difference in median PFS between DA (5.9 months, 95% CI 5.2-6.5) and DE (7.1 months, 95% CI 6.2-8.0) (1.2 months; = 0.02); the HR was 0.82. The adjustment for the baseline Gleason score also made the difference between DE and DA

Discussion
In this meta-analysis, we compared the efficacies of abiraterone and enzalutamide by pooling results from 19 published studies, which yielded 24 cohorts with median OS and/or median PFS estimates. We found that treatment with first-line enzalutamide was associated with improved outcomes both in terms of OS (HR = 0.81) and PFS (HR = 0.47) compared to first-line abiraterone in the predocetaxel CRPC setting. First-line enzalutamide treatment was associated with a median OS advantage of 5.9 months and a median PFS advantage of 8.3 months; these advantages were   further improved to 19.5 and 14.6 months, respectively, after baseline Gleason score was taken into account. We note that the greatest median survival estimates for AX belonged to the clinical trial COU-AA-302 for both OS and PFS and that, in the case of OS, COU-AA-302 showed much longer median survival than all other studies which, we suspect, is due to the highly selective nature of clinical trial investigations. The single-center studies that decrease the combined estimate for predocetaxel abiraterone may be more reflective of realworld experiences. In the postdocetaxel setting, enzalutamide showed a small but statistically significant (especially after adjusting for baseline Gleason score) advantage over abiraterone with respect to PFS.
A recent pooled analysis of only major phase III clinical trials PREVAIL, AFFIRM, COU-AA-301, and COU-AA-302 conducted by Chopra et al. [25] yielded similar but less significant findings. Enzalutamide was suggested to be superior to abiraterone with respect to radiographic PFS in both pre-and postdocetaxel settings; their results for the OS were not statistically significant although the direction of effect was in agreement with our findings. In that study, adjustments for baseline measures were not considered even though Gleason score has been shown to be strongly predictive of survival outcomes in CRPC [26]. Our finding of the association between enzalutamide use and longer survival for both OS and radiographic PFS, especially after adjusting for baseline Gleason score, underscores the need for prospective studies comparing the two drugs and suggests that abiraterone and enzalutamide should perhaps not be considered as interchangeable AR-targeting agents.
It is important to note that our study seeks to identify differences in outcomes of abiraterone and enzalutamide utilized in the CRPC trajectory but does not directly address the issue of optimal sequencing of AR therapies in relation to one another or with docetaxel. We were also unable to directly compare efficacies of one AR agent after the other (i.e., abiraterone after enzalutamide or vice versa) given the lack of studies on these sequences fitting our inclusion criteria. To date, there have been some retrospective studies reporting single-center experiences with the sequencing of abiraterone and enzalutamide. Maughan et al. [19] suggested enhanced PFS using the abiraterone-to-enzalutamide sequence over the enzalutamide-to-abiraterone sequence, suggesting that the former may maximize the therapeutic benefit of both therapies while minimizing cross-resistance. A second retrospective study also revealed similar findings [27]. Our finding of potential enzalutamide superiority in the firstline CRPC setting is not necessarily at odds with those results, as the AX and EX cohorts received a heterogeneous set of therapies after abiraterone or enzalutamide failure, and does not necessarily reflect outcomes for when one AR agent is followed directly by the other. In addition, due to the lack of access to individual patient-level data from the studies, we were unable to identify the subgroups of patients that benefited most from enzalutamide in the first-line setting. It is entirely possible that certain unknown patient characteristics are accounting for the superior survival in predocetaxel enzalutamide-treated patients. Mechanistically, these patients may have derived more benefit from firstline enzalutamide given that, perhaps, CYP17-driven adrenal androgen production (target of abiraterone therapy) was not the major driver of their disease.
Our findings, combined with the fact that several studies have suggested an attenuated response to the second AR agent compared to treatment naïve cases [7][8][9][10], are reasons to pursue prospective trials aiming to optimize treatment sequence in CRPC. To this end, the optimal sequencing of AR-targeting agents in CRPC is being assessed by ongoing prospective studies such as NCT02125357, a phase II randomized study of abiraterone-to-enzalutamide versus enzalutamideto-abiraterone in chemo-naïve CRPC patients. Preliminary results suggest that enzalutamide use is associated with superior PSA response compared to abiraterone use first line [28]. This ongoing study also includes efforts on biomarker identification using circulating tumor DNA (ctDNA) to assess genomic alterations in genes such as AR, p53, and BRCA. Such predictive biomarkers are an important asset to clinical decision-making and treatment selection in the era of noninvasive tumor profiling. To that end, AR splice variant-7 (AR-V7) is a ligand-independent variant of the androgen receptor that has emerged as both an underlying mechanism of resistance and a promising predictive biomarker in CRPC. While there are over 20 known AR splice variants, AR-V7 has established clinical relevance with its detection in clinical specimens associated with inferior responses to abiraterone and enzalutamide [29][30][31]. Taken together, the comparative efficacy of abiraterone and enzalutamide must be assessed in relation to known and emerging biomarkers of resistance in CRPC.
Clinical decisions on the sequencing of therapies in CRPC remain largely consensus-based rather than evidence-based, given the lack of prospective head-to-head trials assessing efficacies of agents in relation to the sequence in which they are deployed. Here, we present a trial-level meta-analysis using data from prospective trials and retrospective studies, suggesting that enzalutamide use is associated with longer median OS and PFS compared to abiraterone in the first-line (predocetaxel) setting and that this survival improvement is further accentuated when baseline Gleason score is taken into account. These findings highlight the limitations in using a consensus-based approach to treatment selection in treatment naïve CRPC patients and the need to pursue prospective trial validation. However, until further work is done to confirm optimal treatment selection and treatment sequencing, biomarkers in the management of metastatic CRPC, clinical factors such as comorbid conditions, cost considerations, patient preference, and side effect profiles should continue to guide the clinician's decision on treatment sequencing of systemic therapies for men with metastatic CRPC.

Conflicts of Interest
Emmanuel S. Antonarakis is a paid consultant/advisor to Janssen, Astellas, Sanofi, Dendreon, Medivation, and ESSA; he has received research funding to his institution from Janssen, Johnson & Johnson, Sanofi, Dendreon, Aragon, Exelixis, Genentech, Novartis, and Tokai; and he is the coinventor of a biomarker technology that has been licensed to Tokai and Qiagen. Other contributing authors have no conflicts of interest, including specific financial interests or relationships and affiliations relevant to the subject matter or materials discussed in the manuscript.