Systematic review of early vs deferred hormonal treatment of locally advanced prostate cancer: a meta-analysis of randomized controlled trials


Steven J. Edwards, Outcomes Research, AstraZeneca UK Ltd, Horizon Place, 600 Capability Green, Luton, Bedfordshire LU1 3LU, UK.



To compare the effectiveness of hormonal treatment (luteinizing hormone-releasing hormone agonists and/or antiandrogens) as an early or as a deferred intervention for patients with locally advanced prostate cancer (LAPC), as radiotherapy is currently the standard treatment for LAPC, with hormonal treatment considered a reserve option.


We systematically reviewed randomized controlled trials (RCTs) in patients with LAPC treated with standard care (radical prostatectomy, radiotherapy, and/or watchful waiting) or standard care plus hormonal treatment. Outcomes assessed were mortality and objective disease progression. The meta-analysis used a fixed-effects model.


Of the 108 trials identified, seven met the inclusion criteria and were of sufficient quality to be included in the analysis. Early intervention with hormonal treatment significantly reduced all-cause mortality compared with deferred treatment (relative risk, RR, 0.86; 95% confidence interval, CI, 0.82–0.91; P < 0.001). Similarly, early vs deferred use of hormonal treatment significantly reduced: prostate cancer- specific mortality (RR 0.72; 95% CI 0.65–0.79); overall progression (RR 0.74; 0.69–0.78); local progression (RR 0.65; 0.57–0.73); and distant progression (RR 0.67; 0.61–0.74; all P < 0.001). Results were robust to changes in inclusion/exclusion criteria and use of a random-effects model for the meta-analyses. Heterogeneity and publication bias had no significant effect on the analyses.


Early intervention with hormonal treatment for patients with LAPC provides significantly lower mortality and objective disease progression than deferring their use until standard care has failed.


(locally) advanced prostate cancer


radical prostatectomy


randomized controlled trial


relative risk




nonsteroidal antiandrogen




The global incidence of prostate cancer has been estimated at >500 000 new cases each year [1]; this equates to ∼10% of all cancers in men. In 2000, it was estimated that there were 180 400 new cases of prostate cancer and 31 900 prostate cancer deaths in the USA [2]. In 2002, >30 000 men in the UK were diagnosed with prostate cancer, which has now overtaken lung cancer to become the most common cancer in men [3].

Although prostate cancer usually progresses slowly, a patient’s prognosis and subsequent treatment depends heavily on the grade of the tumour at diagnosis [4]. A 10-year prostate cancer-specific survival of >90% was reported in men with early, low-grade tumours, and of >75% among those with intermediate-grade tumours [5]. As would be expected, there is a significant decline in 10-year prostate cancer-specific survival among patients with more aggressive high-grade tumours.

Locally advanced prostate cancer (LAPC) is a serious condition in the UK, accounting for >27% of all new presentations of prostate cancer [6]. Of patients thought to have organ-confined (T1c or T2) prostate cancer, 25–40% fail after radical prostatectomy (RP) or radiation therapy [7]. Options for treating LAPC include watchful waiting, RP, and hormonal monotherapy or radiation therapy combined with androgen deprivation. Radiotherapy is the most commonly used treatment in conjunction with neoadjuvant, concomitant and/or adjuvant hormone treatment [8]. The aim of adding hormone treatment is first to reduce the risk of distant metastases, by sterilizing micrometastatic deposits at the time of diagnosis. Second, it reduces the risk of local recurrence, possibly by acting synergistically to enhance radiation-induced apoptosis [9,10].

The Veterans Administration Cooperative Urological Research Group [11–14] first investigated the concept of early vs deferred hormonal treatment in patients with APC and LAPC. The hormonal treatment used in these trials was diethylstilbestrol (DES) as adjuvant therapy to orchidectomy, vs orchidectomy, DES and placebo. In patients with stage III and IV disease there was no difference in prostate cancer-specific mortality. A post hoc analysis of the deaths in the trials showed a reduction in prostate cancer-specific mortality in patients taking adjuvant DES, but an increase in cardiovascular-specific mortality. Since this time, better tolerated hormonal treatments, (primarily LHRH agonists and nonsteroidal antiandrogens (NSAAs)), have become the standard for hormonal therapy.

We examined the impact on survival (overall and prostate cancer-specific) of androgen-deprivation therapy in men with LAPC, using a systematic review of the available RCTs comparing early hormonal treatment (LHRH agonist or NSAAs) with deferred treatment in patients with LAPC.


The CENTRAL, BIOSIS, DARE, EMBASE and MEDLINE databases were searched for relevant abstracts and papers; all searching was completed in July 2006. The following search terms were included in the search strategy: surgery (including ‘orchiectomy’ or ‘orchidectomy’ or ‘castration’); radiotherapy (or brachytherapy); chemotherapy; watchful waiting (or active surveillance or active watching); and hormonal therapy, including antiandrogens (androcur or bicalutamide or climen or cyproterone or dutasteride or finasteride or flutamide or ketoconazole or nilutamide or spironolactone) or the LHRH agonists (buserelin or goserelin or leuprolide or triptorelin); and clinical trial (or clinical trial, phase III or clinical trial, phase IV or controlled clinical trial or randomized controlled trial); and ‘prostate cancer’ (or ‘locally advanced prostate cancer’). No restriction was placed on the language used in the publications.

The criteria for selecting trials for inclusion in the review were: a study design that was randomized and controlled; direct comparison of hormonal therapy (i.e. LHRH agonist or NSAA) with deferred treatment (i.e. hormonal therapy after watchful waiting, radiotherapy or RP) in patients with LAPC.

The main criterion for determining the quality of trials was the method of randomization and concealment of allocation of the treatments used, as this is the aspect of RCT design likely to introduce the most bias [15–17]. This was classified as: adequate; possibly adequate; or inadequate.

The primary outcome of interest was overall mortality. Secondary outcomes assessed were: prostate cancer-specific mortality, overall progression, local progression, and distant progression. It was planned that overall adverse events and withdrawal rates would also be assessed. Data were taken from qualifying trials, and recalculated if not presented in an intention-to-treat (ITT) format. For the purposes of this review, we defined ITT as, ‘patients being analysed in the treatment arm that they entered at randomization, regardless of whether they discontinued, received the incorrect treatment or withdrew before completion of the trial’. If results were presented in a different format, they were recalculated where possible. Patient numbers reinstated into the analysis were assumed to have the worst possible outcome for each of the analyses, e.g. in the primary analysis they were assumed to have died.

Where a RCT included patients with different stages of prostate cancer, e.g. LAPC and APC, only the LAPC data were extracted. If there were multiple publications of the same study, the latest publication was used. If the latest publication did not provide data for all outcomes, then previous publications were used if they contained the missing data. Individual trials could only contribute to the analyses for which they reported the appropriate outcomes. Data were initially extracted by one reviewer, and subsequently checked independently by two other reviewers.

The meta-analyses used the Mantel-Haenszel method (a fixed-effects model); this model was used rather than a random-effects model in the primary analysis, as the weighting favours larger trials in a fixed-effects model, rather than the more equal weighting across all trials used in a random-effects model [18]. The summary effect estimate was calculated as a relative risk (RR) with 95% CI, as a RR is more easily interpreted than other summary effect estimates such as an odds ratio [19]. Publication bias was assessed by using a regression of normalized effect vs precision, using the method proposed by Egger et al.[20]. For all analyses we used StatsDirect version 2.5.6 [21].

For the primary analysis, prospective subgroup analyses for clinical heterogeneity were planned based on: median duration of follow-up (e.g. 5, 10 years, etc.); hormonal treatment regimen (e.g. neoadjuvant, adjuvant, etc.); non-hormonal treatment (e.g. radiotherapy, watchful waiting, etc.). If significant statistical heterogeneity was detected by a chi-square test, subgroup analyses were planned to identify which trial(s) were responsible for the heterogeneity.

The following alternative analyses were conducted: (i) as quality scoring of RCTs involves a degree of subjectivity it was planned to assess what effect including all trials regardless of quality assessment would have on the analysis; (ii) using the DerSimonian and Laird method [22] (a random-effects model) for the analysis.


Of the 108 papers identified in the search of bibliographic databases, eight were appropriate for quality assessment (Fig. 1). The characteristics of the eight trials identified are given in Table 1[23–34] and the data extracted and recalculated by ITT (where required) are presented in Table 2. The quality assessment of trials was ‘possibly adequate’ concealment of allocation to treatment. This was typically due to the brevity of description in the methods sections of each trial, i.e. there was insufficient detail to have complete confidence in the methods used. The one trial that failed to proceed beyond this stage had an imbalance of risk factors in the two treatment groups (e.g. Gleason score) [26]. However, the effect on the analyses of including this trial was tested in an additional sensitivity analysis.

Figure 1.

Results of a search of BIOSIS, CENTRAL, DARE, EMBASE and MEDLINE, for RCTs comparing early with deferred treatment in patients with LAPC.

Table 1.  RCTs of early vs deferred hormonal treatment of LAPC identified by the search
Trial codeLatest publicationTrial designPatient age, yearsComparisonMedian FU, years (all patients)Sponsor
  1. NCI, National Cancer Institute of the USA; RT, radiotherapy; GOS, goserelin 3.6 mg s.c. every 28 days; FLU, flutamide 250 mg oral three times daily; BIC, bicalutamide 150 mg once daily; BUS, buserelin 6.3 mg; CPA, cyproterone acetate; WW, watchful waiting; AA, antiandrogen; BOx, bilateral orchidectomy; *The two LHRH groups were combined in the analysis; †SC, standard care (RP or RT or WW).

MRC[23]Open, randomized (minimization [31])AdultsWW vs GOS + AA (or BOx) until progression (M0 patients only)Not stated (≤11)MRC
RTOG 86–10[24]Open, randomized (scheme [32])71 (49–88)RT vs RT + GOS + FLU (started 2 weeks before RT and continuing during RT for 112 days in total)6.7NCI
EORTC 22863[25]Open, randomized (minimization [33])71 (51–80)RT vs RT + GOS (starting on day 1 of RT) for 3 years + CPA oral for 28 days5.5AstraZeneca
EST-3886[26]Open, randomized66 (45–78)WW vs GOS (or BOx) until progression7.1NCI
TTROG 96.01[27]Open, randomized (minimization)68 (41–87)RT vs RT + GOS + FLU for 3 months (starting 2 months before RT) vs RT + GOS + FLU (starting 5 months before RT)*5.9AstraZeneca and Schering-Plough funded data analysis
EPC[28]Double-blind, double-dummy, randomized67 (38–93)SC vs SC + BIC for ≥2 years (trials 23, 25) or ≥5 years (trial 24)7.4AstraZeneca
RTOG 85–31[29]Open, randomized (scheme [32])AdultsRT vs RT + GOS (starting in last week of RT) until progression7.6NCI
EORTC 30891[30]Open, randomized73 (52–81)WW vs BUS + CPA oral for 14 days (or subcapsular Ox) until progression7.8NCI
Table 2.  Data extracted and recalculated as ITT (if necessary) from the RCTs of early vs deferred hormonal treatment of LAPC identified by the search
MRC [23]RTOG 86–10 [24]EORTC 22863 [25]EST-3886 [26]*TTROG 96.01 [27]EPC [28]RTOG 85–31 [29]EORTC 30891 [30]
  • *

    Only included in the sensitivity analysis;

  • data extracted from 1999 publication Messing et al.[34].

Early, n
Patients randomized256234207495421367488492
 Overall150120 5415 115 351280269
 Cancer 81 60 16 8 59 14393106
 Overall19010015 53 439382
 Local 80  7 39  92129
 Distant 96 90 26 45139128
Deferred, n
Patients randomized24723720851276 1315489493
 Overall177143 9126 71 366327297
 Cancer125 81 5222 42 183134 112
 Overall168 3142 74 521257
 Local105 25 34 139199


In the primary analysis, the meta-analysis of all-cause mortality of early vs deferred hormonal treatment of patients with LAPC (Fig. 2a) shows a lower RR in favour of early hormonal treatment (0.86, 95% CI 0.82–0.91; P < 0.001). In the secondary analyses, when assessing prostate cancer-specific mortality (Fig. 2b) rather than all-cause mortality, there was a greater decrease in RR with early than with deferred hormonal treatment in patients with LAPC (0.72; 0.65–0.79; P < 0.001). When comparing early vs deferred hormonal treatments of LAPC in terms of objective disease progression, overall progression (Fig. 2c) was associated with a significant reduction in risk in favour of early treatment (0.74; 0.69–0.78; P < 0.001). Similarly, when assessing local progression (RR 0.65; 0.57–0.73; P < 0.001) or distant progression (0.67; 0.61–0.74; P < 0.001), the results show a significant reduction in risk with early vs deferred hormonal treatment of patients with LAPC (Fig. 2d,e, respectively).

Figure 2.

Meta-analysis of: a , all-cause mortality; b , prostate cancer-specific mortality; c , overall progression; d , local progression; and e , distant progression, of early vs deferred hormonal treat in patients with LAPC.

Adverse events were reported inconsistently in some trials [23–25,28,30] and not reported in others [27,29]. The most common adverse events reported are summarized in Table 3. Similarly, withdrawal rates were only given as an explicit outcome in EPC 2005 [35] (53.3% vs 49.7%, early vs deferred treatment, respectively). There appeared to be an underlying assumption in the RCTs assessed that few if any patients withdrew once treatment had commenced. This prevented the calculation of an overall summary effect estimate of withdrawals.

Table 3.  The most common adverse events reported in the trials included in the meta-analysis of early vs deferred hormonal treatment of patients with LAPC
TrialAdverse events (early vs deferred)
MRC [23]Pathological fractures (1% vs 2%), ureteric obstruction (9% vs 11%), extraskeletal metastases (7% vs 11%)
RTOG 86–10 [24]Treatment terminated in 10% in the early group for flutamide toxicity (5% diarrhoea, 1% hot flushes, 1% liver abnormalities, and 3% other reasons, e.g. rash, nausea, syncope, etc.)
EORTC 22863 [25]Hot flushes 2%; depression 1% and mastodynia/galactorrhoea 1% in the early group
EPC [35]Breast pain (74.8% vs 9.5%), gynaecomastia (66.6% vs 10.8%)
EORTC 30891 [30]Pain (33.5% vs 42.2%), ureteric obstruction (4.7% vs 11.4%)

It was not possible to proceed with the planned sensitivity analysis of including trials that failed the quality screen, as all trials were considered to have the same level of methodological quality. However, a sensitivity analysis was conducted to determine the effect of using all eight trials identified by the search. This resulted in small numerical differences to the overall estimates but did not change the direction or the significance of any differences (Table 4). Similarly, the use of a random-effects model had little impact on the overall results (Table 4).

Table 4.  A summary of meta-analyses including the trial [26] (excluded due to an imbalance of risk factors in the two treatment groups with LAPC) or using a random-effects model
OutcomeIncluding [26]Random-effects model
RR (95% CI)PRR (95% CI)P
  • *

    The outcome was unchanged with [26] as the excluded trial did not report these data and so could not be included in the analysis.

 All-cause mortality0.86 (0.81–0.90)<0.0010.85 (0.80–0.91)<0.001
 Prostate cancer-specific mortality0.71 (0.64–0.78)<0.0010.70 (0.59–0.83)<0.001
 Overall progression0.73 (0.68–0.77)<0.0010.68 (0.55–0.84)<0.001
 Local progression*0.65 (0.57–0.73)<0.0010.65 (0.55–0.77)<0.001
 Distant progression*0.67 (0.61–0.74)<0.0010.67 (0.58–0.78)<0.001

A chi-square test was used to investigate possible statistical heterogeneity in the primary analysis. The meta-analysis of all-cause mortality did not fail the test for homogeneity (Q 9.28; d.f. = 5, P = 0.16).

The trials identified had a range of median follow-up of 5.5–7.8 years (excluding the MRC trial that did not report these data). As such, the planned subgroup analysis based on median duration of follow-up was not conducted, as the follow-up in the individual trials was deemed too similar for any meaningful comparison.

However, the trials did fall into two discrete groups when considering neoadjuvant [24,27] and adjuvant [23,25,28–30] hormonal treatment. The reduction in all-cause mortality associated with neoadjuvant or adjuvant treatment was very similar to that in the combined primary analysis (RR 0.84; 95% CI 0.73–0.97; P = 0.016; and 0.87; 0.82–0.92, P < 0.001, respectively). Even including the excluded adjuvant trial [26] made very little difference to the overall significant reduction in all-cause mortality (RR 0.86; 95% CI 0.81–0.91, P < 0.001).

When considering the type of standard care given in the ‘deferred’ hormonal treatment group, the only treatments that were exclusively used in more than one trial were radiotherapy [24,25,27,29] and watchful waiting [23,30]. In the radiation subgroup, the reduction in all-cause mortality associated with early vs deferred hormonal treatment in patients with LAPC was greater than in the combined primary analysis (RR 0.69; 95% CI 0.61–0.78, P < 0.001), while in the watchful-waiting subgroup the reduction in all-cause mortality associated with early vs deferred hormonal treatment in patients with LAPC was less than in the combined primary analysis (RR 0.88; 95% CI 0.81–0.96, P = 0.003).

The few trials identified in the search made any assessment of publication bias problematic. However, in the primary analysis, there was no apparent asymmetry in a funnel plot of the data, and an assessment of regression of normalized effect vs precision detected no significant publication bias (bias −2.52; 95% CI −5.50 to 0.45, P = 0.08).


This systematic review showed that early intervention with hormonal treatments significantly reduces the risk of overall mortality (a relative reduction of 14%, P < 0.001). This result was stable to changing the inclusion/exclusion criteria (i.e. when adding in the excluded study [26] the relative reduction remained at 14%, P < 0.001), and using a random-effects model rather than a fixed-effects model (the risk reduction increased to 15%, P < 0.001). Heterogeneity and publication bias also had no significant role in the primary analysis.

A visual inspection of the forest plots identified the EORTC 22863 trial as ‘different’ from the other trials included in the analysis, in that early treatment over deferred is noticeably more beneficial than in the other comparisons. The primary analysis did not fail the test for homogeneity, showing that the results of the EORTC 22863 trial are within the boundaries of probability, based on the other trials combined in the analysis. The ‘visual’ difference is perhaps due to the relatively few events for a trial of this size.

A subgroup analysis based on neoadjuvant and adjuvant hormonal treatment had little or no impact on the reduction in mortality (16%, P = 0.016; and 13%, P < 0.001, respectively). One interpretation of these results could be that neoadjuvant and adjuvant treatment offer similar reductions in mortality. A RCT was conducted that directly compared neoadjuvant-concomitant with neoadjuvant-concomitant plus adjuvant treatment in patients with LAPC. The trial design had all patients receive goserelin and flutamide until radiotherapy was completed, with follow-on therapy either as no further hormonal treatment or continued hormonal treatment for 2 years [36]. There was no significant difference in the two groups in terms of overall survival at 5 years (although there were significant advantages for continued hormonal treatment in disease-free survival, prostate cancer-specific survival, biochemical failure, distant metastases, and local progression). A retrospective subgroup analysis showed a significant advantage for continued hormonal treatment in overall survival at 5 years in patients with a Gleason score of ≥8 (81.0% vs 70.7%, P = 0.044) [36].

In the present study the radiation subgroup meta-analysis made a substantial difference to the primary analysis; the reduction in risk of all-cause mortality increased to 31% (P < 0.001), possibly indicating a synergistic effect with radiotherapy and early hormonal treatment of LAPC. The watchful-waiting subgroup meta-analysis had a slightly lower reduction in risk of all-cause mortality than the primary analysis (a difference of 2% in favour of the primary analysis) perhaps due to patients selected for watchful-waiting trials having less nodal involvement in their disease [23,30] compared with trials comparing active treatments with hormonal therapy [24–29].

The secondary outcome measures, prostate cancer-specific mortality and progression, showed similarly impressive reductions with early vs deferred hormonal treatment (relative reductions of 28% for prostate cancer-specific mortality, 26% overall progression, 35% local progression, 33% distant progression; all P < 0.001).

The present study has several limitations; principally, the outcomes assessed were not all present in all RCTs, and the outcomes that were reported did not necessarily feature in the latest publication of that trial. The lack of consistent reporting of adverse events made it impossible to assess this outcome across all trials. Perhaps the most reasonable interpretation of the results is that hormonal treatments typically have additional adverse events compared to standard care for LAPC, with the profiles depending on the hormonal treatment given (e.g. bicalutamide, goserelin, flutamide, etc.).

Similarly, the lack of data on withdrawals available in the RCTs assessed prevented any calculation of the RR of withdrawals between early and deferred hormonal treatment. However, the use of an ITT analysis, using a conservative approach towards missing data, hopefully militates against this weakness in available data.

Since the literature review and subsequent analysis was conducted, an update to the EST-3886 trial was published [37], the results of which show a significant benefit for early vs deferred hormonal treatment in overall survival, prostate cancer-specific survival, and progression-free survival [37]. As this trial was only included in a sensitivity analysis, and as the results complement the results presented here, the impact of this update makes no difference to the conclusions drawn from the present study.

Research on prostate cancer has provided clinicians with additional information on the use of hormonal treatments at the various stages of severity of the disease. In 2001, Wilt et al.[38] published a systematic review of early vs deferred hormonal treatment in APC, showing significant benefits for early treatment in patients followed for a median of ≥5 years. The present study shows that hormonal treatment has a similar role in LAPC. Further research should be conducted to determine if hormonal treatment offers these benefits if used at earlier stages of prostate cancer.

In conclusion, early intervention with hormonal treatment for patients with LAPC provides important reductions in all-cause mortality, prostate cancer-specific mortality, overall progression, local progression, and distant progression compared with deferring their use until standard care has failed to halt the disease. Early introduction of hormone therapy should be offered to men at highest risk of disease progression and cancer-related death.


This research was supported by AstraZeneca UK Ltd; S.J.E. is a full-time employee of AstraZeneca UK Ltd.