Combined androgen blockade in prostate cancer: meta-analyses and associated issues

Authors


L. Klotz, MD, FRCS(C), Professor of Surgery, Sunnybrook & Women's College Health Sciences Centre, 2075 Bayview Avenue, #MG 408, Toronto, North York, Ontario M4N 3M5, Canada.
e-mail: laurence.klotz@swchsc.on.ca

Introduction

The concept of combined androgen blockade (CAB), also known as maximal androgen blockade, was widely accepted by urologists, after the confirmation of a clear survival benefit over monotherapy by a definitive study in 1989 [1]. However, further studies and meta-analyses have reported conflicting results and CAB has become the subject of widespread debate. The intention of this review is to comment on the latest update of the meta-analysis carried out by the Prostate Cancer Trialists Collaborative Group (PCTCG), recently published by Peto et al.[2], and to discuss this in the context of other meta-analyses and clinical trials.

The history of maximal androgen blockade

Castration by medical or surgical means suppresses androgen production by the testes, but has no effect on androgen production by the adrenal glands. Castration may be used in combination with antiandrogens. These competitively antagonize the androgen receptor and therefore block the action of androgens from any source. They may also block ligand-independent activation of the receptor. The first studies to report the use of CAB were published at the end of the 1970s, using bilateral orchidectomy combined with the steroidal antiandrogen cyproterone acetate (CPA) [3–5]; these found a slight benefit for CAB over castration alone. In the early 1980s, Labrie et al.[6] reported a small clinical series suggesting a dramatically improved efficacy with CAB. This finding provoked a sharp increase in research activity (strongly supported by the pharmaceutical industry), with over 20 randomized prospective trials comparing combination therapy with castration, published between 1982 and 1989 in Europe and North America [1,7–37]. There are few single questions in medicine which have been subjected to so many prospective randomized trials.

In 1989, the final analysis of the largest (613 patients) and most definitive study to that date [1] showed a 26% improvement in the survival of patients on the combination arm (LHRH agonist+flutamide). This trial changed practice, leading to the widespread use of CAB for patients with advanced prostate cancer. In addition, subgroup analysis revealed a greater benefit in the subset of patients with minimal metastatic disease and high performance status. This was widely interpreted as evidence for the benefit of the earlier introduction of androgen suppression; however, this question remains controversial. A recent meta-analysis comparing the early vs late introduction of CAB, also conducted by the PCTCG, reported that deaths from prostate cancer were reduced by about a third in the early treatment group, but that deaths from other causes increased [38].

Subsequent publications comparing CAB with monotherapy showed mixed results, but most of these studies had an inadequate follow-up and/or were too small to show definitively a difference between treatments. Larger studies have since been completed but the results remained conflicting, with some studies showing a benefit of CAB over monotherapy [39,40] and others the reverse [41].

Meta-analyses evaluating CAB

One approach to resolving the question of the value of CAB has been to use meta-analysis. By combining the data from many independent studies, a well-conducted meta-analysis has greater statistical power and thus is able to detect more modest, but real, differences between treatments than could have been detected in any one trial alone.

In 1989, an overview analysis (a meta-analysis which uses raw data from each of the investigators) of the multiple CAB trials was planned, with the co-operation of the trial investigators (the PCTCG), and was published in 1995 [42]. Although there was a trend towards a benefit from CAB, this was not statistically significant, leading the authors to conclude that the 22 trials showed no survival benefit for CAB. Four further meta-analyses of CAB vs monotherapy were published up to January 1999 [43–46]. All reported a survival benefit in favour of CAB (risk of death 6–22% lower than for castration alone), which was statistically significant in all but one study (Fig. 1).

Figure 1.

An overview of published meta-analyses of survival with CAB vs monotherapy; NSAA, nonsteroidal antiandrogen; nilut, nilutamide; flut, flutamide. *Derived from information in [2]; †relative risks using data from [2]; NSAA(PH), relative risks (proportional hazards) from discrete proportional hazards model; NSAA(LH), relative risks (log hazard ratio) from P values.

A further meta-analysis of randomized trials of CAB vs monotherapy was conducted for the USA Agency for Health Care Policy and Research and published on the Internet [47]. This analysis included 21 trials and a total of 6871 patients. There was no significant difference for overall 2-year survival, but overall 5-year survival was significantly better for CAB than for monotherapy, with a hazard ratio (HR) of 0.871 (95% CI 0.805–0.942).

The PCTCG recently updated its meta-analysis [2], which now includes 27 clinical trials and 8275 patients. This update reported a trend towards improved survival for CAB compared with monotherapy, with a HR of 0.958 (sd 0.026; Fig. 1), but this was not statistically significant. Subgroup analysis indicated that nonsteroidal antiandrogens may be associated with better survival than the steroidal antiandrogen CPA. The heterogeneity among the three treatments was statistically significant ( inline image = 9.4; P = 0.009), indicating that outcome is dependent on the antiandrogen used. The HR for studies using CPA was higher (and significant) than for nilutamide and flutamide (Fig. 1). If the CPA studies were excluded, CAB with nonsteroidal antiandrogens (flutamide plus nilutamide) was associated with a reduction in the risk of death of 8% (95% CI 3–13; 2P = 0.005), which translates to a small but significant improvement in 5-year survival of 2.9% over castration alone. Conversely, CAB with CPA is associated with an increased risk of death of 13% (95% CI 1–27; 2P = 0.04), which translates to a small but significant reduction in 5-year survival of 2.8% relative to castration alone.

Assessment of meta-analyses

The 1995 PCTCG overview [42] represented a major effort to collate the data from many trials. However, meta-analysis is by definition retrospective and subject to bias, as is any retrospective analysis. In recent years, methods for evaluating the quality of meta-analyses have been defined [48–50]. Table 1 lists the benefits and limitations of the five meta-analyses of CAB vs monotherapy, the results of which are shown in Fig. 1.

Table 1.  Benefits and limitations of the various meta-analyses of CAB compared with castration alone
Meta-analysisBenefitsLimitations
[2]Comprehensive, included
published and unpublished randomized trials
No report of initial protocol document
No study included potential confounding effect of
Access to individual patient data
27 trials, > 8200 patients, > 5900 events
most powerful meta-analysis
antiandrogen withdrawal
Appropriate statistical methodology used
[43]Quality of trials assessedIncluded only published data
20 randomized trials (19 were PCTCG trials) –
high statistical power
Not clear how data were analysed given there was no
access to individual patient data
Appropriate statistical methodology used
[44]A priori search criteria givenIncluded only published data – 13 trials
Appropriate statistical methodology usedOnly 7/13 trials available for re-analysis
Randomized trials only
Quality of trials assessed
No access to individual patient data – not clear how
HRs and P-values were calculated
[45]Randomized (nilutamide) trials onlyRelatively little power – seven trials
Valid statistical methodology usedOnly published trials included
 Apparent access to individual patient dataNo search/selection criteria given – one trial in patients
with painful metastases excluded as ‘study design was not
comparable’
Flaws in combination of various doses of nilutamide and
incorporation of ‘small’ trials into larger trials before
meta-analysis
[46]Randomized (flutamide) trials only includedOnly published trials included
 Valid statistical methodology usedOnly nine trials, most of which were small for assessing
survival
The largest trial (NCI 0105) may have had an undue
influence on the overall result
No access to individual data
Number of events not given, therefore maturity of data not
assessable

Study quality

The importance of formally assessing the quality and validity of the individual research studies included in the analysis is apparent and demonstrable [51,52]. Such assessments provide a measure of the extent to which bias has been minimized in any given trial, and methods to evaluate the quality of individual trials are well described [53]. These measures assess the extent to which:

  • • Subjects were randomized appropriately.
  • • The randomization was concealed from subjects and investigators.
  • • Treatment delivery was double-blind.
  • • The assessment of outcomes was blinded.
  • • All subjects enrolled in the trial were accounted for at its conclusion.

There is empirical evidence that estimates of effect of size are related to methodological rigour, such as the concealment of randomization [53]. For the question of CAB vs castration alone, where the results are so conflicting, it is especially important that the quality of the primary studies be assessed. A sensitivity analysis involves ranking the studies by their quality, and determining if there is a relationship between the quality of the included studies and the calculated benefit of therapy.

The inclusion of unpublished data, including abstracts, remains controversial. If unpublished work is included in a meta-analysis, as most meta-analysts and methodologists agree should occur [54], then there are standards which should be met to include the data. These include the following:

  • • The search for unpublished material must be explicit and systematic.
  • • Both published and unpublished studies must be subject to a uniform methodological evaluation of their quality.
  • • A sensitivity analysis that includes and excludes the unpublished data ought to be included to check for methodological quality.

In 1996, we conducted a formal critique and sensitivity analysis of the 1995 PCTCG meta-analysis [42], which identified several possible weaknesses in the study [55]. The PCTCG analysis discussed none of the issues of study quality described above. Whilst the quality of the studies may have been assessed, with an appropriate selection procedure in place, it has never been described and thus it remains uncertain. It may have been that some of the earlier trials were ‘marketing’ studies, with poor quality control of the data. The sensitivity analysis [55] showed the following:

  • • A meta-analysis based on the published trials, where survival data were extractable from the published data, showed a clear benefit in 2-year survival for combination therapy over castration alone.
  • • The unpublished trials, which tended to be considerably smaller, consistently failed to show a benefit.

While this last point may simply reflect publication bias (i.e. negative trials are less likely to be published), or that the smaller trials were under-powered, it is also possible that they had inferior methodological quality. However, the inclusion of unpublished trials with good methodology is vital for a valid meta-analysis.

The finding of a survival benefit for CAB over castration alone in the more rigorous trials is supported by a meta-analysis using published randomized controlled trials of nonsteroidal antiandrogens by the Metaworks Group [44]. This study also showed a substantial benefit of CAB, with clear confidence limits. The Metaworks' result may also reflect an improved benefit with nonsteroidal vs steroidal antiandrogens, as observed in the latest update of the PCTCG [2].

Trial design

The duration of follow-up may have an important influence on the results of a study. Patients with no metastatic disease have relatively long survival times and only a study with a very long follow-up can expect to record sufficient events to detect a difference. A trial is unlikely to detect a survival difference if most of the patients are expected to survive longer than its follow-up time. Trials with a short follow-up (sometimes described as ‘immature’) will report relatively few events compared with ‘mature’ trials with a long follow-up, in which most of the patients may have experienced the event being assessed. Some statistical methods are more sensitive to the effects of immature data than others. For example, the log-rank test is sensitive to late differences, and may produce a suboptimal outcome if applied to immature data. Many of the trials reviewed in the 1995 PCTCG analysis [42] were relatively immature and based their results on the log-rank statistic. In the 2000 PCTCG analysis [2], updated follow-up information was obtained for 13 of the original 22 trials and the follow-up was typically ≈ 5 years. The log-rank analysis, using exact dates of randomization and death, was calculated separately for each trial, thereby allowing for differences in follow-up duration. The HR (relative risk) is the chance of an event (e.g. death) occurring in one treatment group compared with another, over the whole follow-up. A HR of < 1 indicates a lower chance of the event occurring in the first treatment group than in the second. As the HR takes into account events occurring throughout the follow-up, it provides a more valid measure of treatment differences.

The effect of new knowledge

Finally, the treatments evaluated in randomized trials may sometimes be overtaken by changes in medical understanding and clinical practice. Most of the patients in the trials published to date have been maintained on antiandrogen therapy until death. In 1993, Kelly and Scher [56] first reported a secondary response to the discontinuation of flutamide, which has since been reported with the other antiandrogens [57–60]. It is possible that the benefit of CAB occurs during the hormone-dependent phase of the disease, and that a survival benefit is dependent in part on the antiandrogen being discontinued during the hormone-refractory phase. The studies included in the 1995 and 2000 PCTCG overviews [2,42] were all initiated before this phenomenon was reported.

The importance of antiandrogen

As described earlier, the recent update of the PCTCG meta-analysis [2] indicated that the choice of antiandrogen had a significant effect on the survival benefit obtained from CAB compared with castration. The nonsteroidal antiandrogens flutamide and nilutamide were associated with an 8% decrease in the risk of death (statistically significant for flutamide but not for nilutamide), whereas the steroidal antiandrogen CPA was associated with a statistically significant 13% increase in the risk of death. These results support the position that the choice of antiandrogen has an effect on the outcome.

These findings are supported by other meta-analyses which have reported statistically significant survival advantages for CAB with nonsteroidal antiandrogens. The improvement in survival compared with castration was 10% for flutamide [46] and 16% for nilutamide (Fig. 1) [45].

The two most recent trials published, both fairly large and meticulously conducted, investigated the value of CAB using orchidectomy in conjunction with nonsteroidal antiandrogens. The intent was to exclude the possibility that blocking ‘flare’ with LHRH agonist treatment was the sole reason for perceived survival differences with CAB. The long-term efficacy of CAB with nilutamide has been compared with castration in stage D2 prostate cancer [40,61]. This study consists of a 9-year follow-up of 457 patients with advanced prostate cancer randomized between orchidectomy and nilutamide or placebo. There was a significant benefit in progression-free, cancer-specific and overall survival in the CAB group, with relative risk ratios of 0.72, 0.77 and 0.80, respectively. This translated into a median overall survival of ≈ 28 vs 23 months. The survival difference increased over time, i.e. at 8 years there was ≈ 30% greater survival in the CAB group. The morbidity of nilutamide in this study was small. In addition, a retrospective stratification analysis showed a survival benefit of almost 3 years for CAB in the favourable disease group, reminiscent of the initial leuprolide/flutamide study [1].

In contrast, the recently reported Southwest Oncology Group study of orchidectomy with or without flutamide in 1387 patients with metastatic prostate cancer showed no additional benefit with CAB [41], although the reduction in risk of death (9%) was in line with the most recent meta-analysis [2]. This study showed no statistically significant difference in time to progression or overall survival in the two groups, although there was an improvement in PSA progression rate in the flutamide arm which did not translate into a difference in disease progression or survival. The size of this trial, the largest of the 26 testing the concept of CAB, commands respect. Some differences in quality of life favouring the castration arm, particularly in the domains of gastrointestinal toxicity and emotional functioning, were reported in a companion publication [62].

How are these varying results reconciled? There are four possibilities. First, the leuprolide/flutamide study used daily LHRH injection rather than surgical castration [1]. It is possible that compliance was imperfect in patients taking daily leuprolide. This may have resulted in repeated mini-flares as patients resumed and discontinued treatment for short periods. In those patients it is conceivable that the benefit of an antiandrogen in blocking these mini-flares would be substantial. There are also recent data suggesting that some patients experience a rise in testosterone level with each injection of depot LHRH analogue, even when this is on schedule [63].

Second, it is possible that antiandrogen treatment has a different role when combined with LHRH agonist than when combined with orchidectomy. Blocking flare is clearly warranted. The use of antiandrogen combined with LHRH agonists would be important in patients on an intermittent protocol, where flare occurs with each reintroduction of therapy.

Third, it is possible that there are important differences in the study populations between the nilutamide study [61] and the flutamide study [41], particularly with respect to the extent of disease.

A fourth, and compelling, possibility is that there are subtle differences between different nonsteroidal antiandrogens that may account for the apparent difference in detected efficacy.

Antiandrogens differ in their affinity for the androgen receptor [64]. When androgen receptors are mutated, a common phenomenon in advanced prostate cancer, antiandrogens may function as androgen receptor agonists [65]. In addition, androgen-independent growth may occur because of androgen-independent activation of the androgen receptor by growth factors and other molecules (e.g. protein kinase A) [66]. Antiandrogens inhibit this androgen-independent activation of the androgen receptor. Recent data indicate that the protein kinase A-induced androgen-independent activation of the androgen receptor is differentially blocked by different nonsteroidal antiandrogens [67,68]. Suppression of such androgen-independent receptor activation may explain the different effects of antiandrogens. This hypothesis is also supported by the marked second-line responses frequently induced by a change in the antiandrogen, e.g. from flutamide to bicalutamide [69–71]. Thus the apparent paradox of the positive orchidectomy-nilutamide trial [61] and the negative orchidectomy-flutamide [41] trial may be ascribable to the different effect of the two drugs. An exploratory analysis comparing different nonsteroidal antiandrogens in a 2 × 2 factorial design (leuprolide vs goserelin and flutamide vs bicalutamide) recently suggested a worse survival in the leuprolide-flutamide arm than in the other three [72]. If confirmed by further randomized controlled trials, differences in efficacy between nonsteroidal antiandrogens could become an important influence in the physician's choice of drug.

Patient selection may also play a role. Two studies have shown a greater benefit of CAB in minimal metastatic disease [1,40]; this remains unconfirmed. A recent study of patients progressing on androgen ablation monotherapy showed that androgen-receptor gene amplification detected in tumours was associated with a favourable treatment response to second-line CAB, compared to those with no androgen-receptor amplification [73]; 21% of responders had amplification compared with 3% of those not responding (P = 0.016). This suggests that it may be possible to use molecular markers to identify subsets of patients who would benefit to a greater degree from CAB.

Implications for the physician

The balance of the available evidence suggests that CAB using nonsteroidal antiandrogens is associated with a modest survival benefit over castration alone. There is a possibility that efficacy may vary among different nonsteroidal antiandrogens. The most recent PCTCG meta-analysis [2] reported a small but significant survival advantage (8%) for CAB using nonsteroidal antiandrogens, compared with a moderate and significant survival disadvantage (13%) for CAB using CPA. Indeed, Labrie and Candas [74], in a commentary on the updated PCTCG analysis, state that the conclusion of the meta-analysis should be that CAB using nonsteroidal antiandrogens prolongs life and reduces the death rate from all causes by 8–10% relative to castration alone. The choice of antiandrogen may also be affected by tolerability issues. For example, diarrhoea appears to be more common with flutamide [41,72], while delayed light/dark adaptation and alcohol intolerance are greatest with nilutamide [75]; neither are seen as major issues with bicalutamide [75]. Clinically relevant outcome measures should be used for clinical studies. Survival is often expressed as a reduction in absolute mortality at a given time (e.g. 5 years), or an increase in the median time to death. Such measures are limited because they reflect the situation only at one point (5 years, or the point at which the Kaplan–Meier curve crosses 50% on the vertical axis). Cause-specific mortality is the percentage of patients who die from a specific cause such as disease progression. However, as the remaining lifetime of patients with advanced prostate cancer is often short, all-cause mortality is usually accepted as the most definitive measure of survival. Other outcome measures include time to progression, objective response, pain, quality of life, and biochemical response. Such measures have been reported in some trials and meta-analyses of CAB vs monotherapy, and have generally indicated a benefit in favour of CAB [1,11,39,40,44].

Statisticians deal with populations, while physicians deal with individuals. A marginal statistical benefit in a patient population may represent a significant benefit for an individual patient. It is reasonable to offer CAB, using a nonsteroidal antiandrogen, to selected patients. Further research will be required to identify those patients who benefit most from CAB. Physicians will need to consider individual benefits and preferences, including the compromise between quality of life, cost and survival when selecting the most appropriate therapy for an individual patient.

Conclusion

Individual studies may provide conflicting results and often lack statistical power, and thus meta-analyses are useful. The most recent meta-analysis reported an underlying variability among trial outcomes. There was a significant reduction in the risk of death with a nonsteroidal antiandrogen in CAB, but a significant increase in the risk of death with a steroidal antiandrogen in CAB.

Interesting differences in efficacy are emerging among the various nonsteroidal antiandrogens, which is linked to their differential effects on the wild-type and mutant androgen receptors. Tolerability also varies among agents and this may influence the choice of therapy in an individual patient. The choice of therapy remains very much a matter for the individual patient and his physician.

Ancillary