Impact of androgen suppression and zoledronic acid on bone mineral density and fractures in the Trans-Tasman Radiation Oncology Group (TROG) 03.04 Randomised Androgen Deprivation and Radiotherapy (RADAR) randomized controlled trial for locally advanced prostate cancer




  • To study the influence of adjuvant androgen suppression and bisphosphonates on incident vertebral and non-spinal fracture rates and bone mineral density (BMD) in men with locally advanced prostate cancer.

Patients and Methods

  • Between 2003 and 2007, 1071 men with locally advanced prostate cancer were randomly allocated, using a 2 × 2 trial design, to 6 months i.m. leuprorelin (androgen suppression [AS]) before radiotherapy alone ± 12 months additional leuprorelin ± 18 months zoledronic acid (ZdA), commencing at randomization.
  • The main endpoint was incident thoraco-lumbar vertebral fractures, which were assessed radiographically at randomization and at 3 years, then reassessed by centralized review.
  • Subsidiary endpoints included incident non-spinal fractures, which were documented throughout follow-up, and BMD, which was measured in 222 subjects at baseline, 2 years and 4 years.


  • Incident vertebral fractures at 3 years were observed in 132 subjects. Their occurrence was not increased by 18 months’ AS, nor reduced by ZdA.
  • Incident non-spinal fractures occurred in 72 subjects and were significantly related to AS duration but not to ZdA.
  • Osteopenia and osteoporosis prevalence rates at baseline were 23.4 and 1.4%, respectively, at the hip.
  • Treatment for 6 and 18 months with AS caused significant reductions in hip BMD at 2 and 4 years (P < 0.01) and ZdA prevented these losses at both time points.


  • In an AS-naïve population, 18 months of ZdA treatment prevented the sustained BMD losses caused by 18 months of AS treatment; however, the study power was insufficient to show that AS duration or ZdA influenced vertebral fracture rates.


Several trials have shown that androgen suppression (AS) therapy improves outcomes after radiotherapy for prostate cancer [1-6]. Two or more years of adjuvant AS have been shown to be more efficacious than treatment for ≤6 months [7, 8]. Longer durations of AS have also been shown to cause more long-term morbidity [9-13], including loss of bone mineral density (BMD) [14-19] sufficient to cause fragility fractures [12, 20, 21]. In designing the present trial to follow the Trans-Tasman Radiation Oncology Group (TROG) 96.01 trial [22], which had shown the advantage of adding 6 months of neoadjuvant AS therapy to radiotherapy, we chose to evaluate the addition of 12 months of adjuvant AS therapy after our 6-month neoadjuvant programme to determine whether it would significantly improve cancer control without significant long-term ill effects. In addition to preventing AS-induced loss of BMD, nitrogen bisphosphonates have shown efficacy in vitro in prostate cancer cell lines [23-25] and have reduced skeletal-related events in castrate-resistant metastatic prostate cancer [26]. We therefore incorporated 18 months of zoledronic acid (ZdA) into the present trial using a 2 × 2 factorial design. Fears concerning mandibular osteonecrosis using monthly dosing led us to use 4 mg i.v. every 3 months, as successfully used by Smith et al. [27]. The incorporation of fractures into the present trial's design as a secondary endpoint was also considered important because it is a known consequence of loss of BMD, and a source of considerable morbidity. Even more importantly, AS-induced fractures were reported in the BJUI recently to result in a 1.38-fold higher overall mortality risk [28].

The risk of AS-induced fractures in the 18-month AS trial arm was unclear when the trial was designed. Our expectation was that BMD would be reduced in subjects receiving 6 and 18 months of AS therapy without ZdA, and that an increased rate of fracture was likely in subjects receiving 18 months of AS. Studies conducted up until now [20, 27, 29], however, have indicated that there is less doubt that using the ZdA dosing scheme adopted would prevent losses in BMD and thereby AS-induced fractures too. In the present report, we describe the effects of the study drugs on BMD at the hip and incident thoraco-lumbar (TL) vertebral and non-spinal fractures.

Patients and Methods

Subjects eligible for the present study had histologically confirmed adenocarcinoma of the prostate without lymph node or systemic metastases, T stage 2b–4 primary tumours or T stage 2a primary tumours of Gleason score ≥7 histology, and baseline PSA levels ≥10 ng/mL. The exclusion criteria included previous treatment with AS or bisphosphonates, previous pelvic RT, osteoporosis resulting in ≥30% vertebral compression fractures, or abnormal renal function (serum creatinine >2 × upper limit of normal). The trial was approved by the independent ethics committees of the participating centres and all the subjects provided written informed consent. Randomization was performed in Newcastle, Australia using the minimization technique, with stratification according to treatment centre, baseline PSA level (<10/10–20/ ≥20 ng/mL), Gleason score (≤6/ ≥7) and T stage (T2/T3,4), and use of a brachytherapy boost (yes/no), which are prognostic variables for the primary oncological endpoint of the trial, prostate cancer-specific mortality. Because BMD and fractures were not primary endpoints of the present TROG 03.04 Randomised Androgen Deprivation and Radiotherapy (RADAR) trial, known prognostic factors for these secondary endpoints were not included in the stratification scheme. Subjects were randomly assigned to four treatment arms in a 2 × 2 factorial design. All subjects received 6 months of leuprorelin (22.5 mg i.m. every 3 months), commencing at randomization, 5 months before radiotherapy to the prostate and seminal vesicles. After this they received no further treatment (i.e. short-term AS [STAS], the control arm) or an additional 12 months of leuprorelin, 22.5 mg i.m. every 3 months (i.e. intermediate term AS [ITAS]). In addition to AS treatment, subjects allocated to the two bisphosphonate treatment arms received ZdA 4 mg i.v. every 3 months for 18 months starting at randomization, (STAS+ZdA and ITAS+ZdA). The primary endpoint of the trial, prostate cancer-specific mortality, will be reported in 2014. The secondary endpoints reported in the present study are vertebral and non-spinal fractures, and BMD in a nested sub-study. These endpoints were reported after a minimum of 5 years’ follow-up as specified in the protocol.

Of these endpoints, the main outcome was incident vertebral fractures after 3 years’ follow-up. Subsidiary endpoints included change in BMD at the hip after 2 and 4 years of follow-up and incident non-spinal fractures over the entire study. In these substudies:

  1. All subjects had TL radiographs before treatment (0 years) and at 3 years. Estimation of loss of vertebral height attributable to wedge, bi-concave and compression fractures [30] for all thoracic vertebrae between T4 and T12, and lumbar vertebrae between L1 and L5 was carried out by each subject's institutional radiologist, blinded to the trial arm. A second centralized independent quantitative review of all TL films was conducted by a board-certified diagnostic radiologist in Wellington, New Zealand, also blinded to the trial arm. A prevalent fracture was defined as loss of vertebral height of ≥20% (i.e. Genant grade ≥1) [31] at baseline. An incident fracture was defined as one or more vertebrae with new loss of vertebral height at 3 years amounting to ≥1 Genant grade. Incident fractures were classified as ‘new’ if the vertebra was not fractured at baseline (i.e. Genant grade 0) or ‘worsening’ if a prevalent fracture existed at baseline. The trial's sample size was determined by the power to detect reductions in the primary trial oncological endpoint (prostate cancer-specific mortality) in the two 18-month AS trial arms. We therefore derived a posteriori estimates of power to detect changes in vertebral compression fractures attributable to the study drugs. When compared with the two 6-month AS arms (where the 3-year incident fracture rate was 17%), the present study had an 80% power to detect a 46% increase in fractures in the two 18-month AS arms to 25%. When compared with the two AS-only arms, the present study had an 80% power to detect a 39% reduction in fractures in the two AS+ZdA arms to 9.5%. If an interaction between the two trial factors existed and trial arms required comparison, the power to detect differences would decline. For example, the trial had a power of 80% to detect an increase in fractures of 68% in the ITAS arm when compared with the STAS arm.
  2. A nested sub-study of subjects underwent dual-energy radiograph absorptiometry (DEXA) scans of the total hip, femoral neck, greater trochanter and lumbar spine before treatment (0 years), at 2 years and 4 years. Because of funding constraints, only the first 240 subjects randomized were eligible for inclusion in the nested BMD sub-study. A total of 222 subjects with complete total hip data were eligible for analysis. This sub-study had an a priori power of 95% to detect an absolute reduction in BMD loss of 2.6% at 2 years (5% in the AS-only arms vs 2.4% in the ZdA arms) if 200 subjects were available for analysis.
  3. All subjects had non-metastatic, incident non-spinal fractures documented at each follow-up visit throughout the entire study period until the closeout date on 31 March 2012. No a priori sample size requirements were defined.


All patients were routinely followed up in clinic every 3 months for up to 30 months, then every 6 months up to 5 years, and annually thereafter for a further 5 years. At each visit new symptomatic incident fractures at any site were documented, clinician-assessed outcomes were collected and clinical examination was performed.

Quality Control Measures

Independent review of TL radiographs was performed as described above and a hip phantom was taken to all DEXA sites across Australia and New Zealand to verify uniform calibration. All DEXA outputs were reviewed by an independent expert blinded to trial arm.


Baseline characteristics were compared across treatment arms using the Kruskal–Wallis test for continuous variables and the chi-squared test for categorical variables.

Interaction between the trial factors (duration of AS and use of ZdA) for the study's main endpoint (incident vertebral fractures at 3 years) was assessed using multiple logistic regression. Subsequent trial arm comparisons were undertaken using multiple logistic regression adjusted for baseline fracture risk using a composite variable based on the method used by Shao et al. [28]. High fracture risk was defined as the presence of one or more of the following factors at baseline: age ≥80 years, history of non-spinal or symptomatic vertebral fractures, diabetes, alcohol consumption of 21 or more standard drinks per week, current smoker, and asymptomatic TL fractures. All other subjects were classified as low risk. Other known risk factors including hepatic dysfunction, corticosteroid use and rheumatoid arthritis were not included in the risk composite owing to very small numbers.

The subsidiary (explanatory) endpoint, BMD at the hip, was assessed by comparing changes in hip BMD from baseline to 2 years and from 2 to 4 years in the four trial arms using paired T-tests. Multiple logistic regression models were derived to assess the influence of treatment arm on BMD, adjusting for baseline fracture risk (low vs high) using the composite variable described above, and baseline BMD T score at the hip (normopenic vs osteopenic/porotic).

Assessment of trial factor interactions and trial arm comparisons of time to incident non-spinal fractures were undertaken using Cox multivariable analyses. These models were adjusted for baseline fracture risk as described above.

A P value <0.05 was considered to indicate statistical significance because all analyses were explanatory. Analyses were by intention-to-treat and were performed in Stata Version 11.2 [32].


Between October 2003 and August 2007, 1071 eligible subjects with locally advanced prostate cancer from 23 Australian and New Zealand cancer treatment centres were randomized. Baseline vertebral fractures and BMD data are shown alongside relevant characteristics of the subjects and their treatment allocation in Table 1. There were no statistical differences between these characteristics in the four treatment arms. Median follow-up time from randomization was 5.6 years.

Table 1. Pretreatment characteristics and incident fractures in the study period according to treatment arm (N = 1071)
  1. *Comparisons exclude missing values. High fracture risk at baseline defined by presence of one or more of the following conditions: age ≥80 years, history of non-spinal or symptomatic vertebral fracture, diabetes, alcohol consumption of ≥21 standard drinks per week, current smoker, asymptomatic TL fractures. All other subjects were defined as low risk. IQR, interquartile range; BMI, body mass index.

Median (IQR) age, years

Age group, n (%)

69.3 (63.9–73.3)69.1 (63.8–73.3)68.6 (63.3–73.2)68.2 (62.6–72.3)68.7 (63.4–73.0)0.39
≤70 years161 (60.1)156 (58.2)165 (61.6)181 (67.8)663 (61.9)0.12
>70 years107 (39.9)112 (41.8)103 (38.4)86 (32.2)408 (38.1) 
T stage, n (%)      
T2170 (63.4)171 (63.8)170 (63.4)169 (63.3)680 (63.5)1.00
T3,498 (36.6)97 (36.2)98 (36.6)98 (36.7)391 (36.5) 
Gleason score, n (%)      
<726 (9.7)25 (9.3)25 (9.3)25 (9.4)101 (9.4)0.73
7155 (57.8)155 (57.8)138 (51.5)151 (56.6)599 (55.9) 
>787 (32.5)88 (32.8)105 (39.2)91 (34.1)371 (34.6) 
PSA group, n (%)      
<10 ng/mL74 (27.6)74 (27.6)72 (26.9)73 (27.3)293 (27.4)1.00
10–20 ng/mL107 (39.9)107 (39.9)104 (38.8)107 (40.1)425 (39.7) 
≥20 ng/mL87 (32.5)87 (32.5)92 (34.3)87 (32.6)353 (33.0) 

Median (IQR) BMI

BMI group, n (%)

27.7 (24.8–30.0)28.1 (25.8–30.1)27.8 (25.3–30.5)27.4 (24.7–29.9)27.8 (25.2–30.1)0.35
≤25 kg/m269 (25.8)45 (16.8)60 (22.4)70 (26.2)244 (22.8)0.20
>25 and <30 kg/m2129 (48.1)141 (52.6)128 (47.8)128 (47.9)526 (49.1) 
≥30 kg/m266 (24.6)69 (25.8)74 (27.6)62 (23.2)271 (25.3) 
Missing4 (1.5)13 (4.9)6 (2.2)7 (2.6)30 (2.8) 
Current smoker, n (%)      
No240 (90.0)231 (86.2)236 (88.1)229 (85.8)936 (87.4)0.44
Yes26 (9.7)35 (13.1)32 (11.9)38 (14.2)131 (12.2) 
Missing2 (0.8)2 (0.8)0 (0.0)0 (0.0)4 (0.4) 
Alcohol consumption, n (%)      
<21 standard drinks/week210 (78.4)207 (77.2)209 (78.0)211 (79.0)837 (78.2)0.99
≥21 standard drinks/week57 (21.3)57 (21.3)58 (21.6)55 (20.6)227 (21.2) 
Missing1 (0.4)4 (1.5)1 (0.4)1 (0.4)7 (0.7) 
TL fractures at baseline, n (%)      
<20%209 (78.0)209 (78.0)220 (82.1)217 (81.3)855 (79.8)0.53
≥20%58 (21.6)58 (21.6)47 (17.5)50 (18.7)213 (19.9) 
Missing1 (0.4)1 (0.4)1 (0.4)0 (0.0)3 (0.3) 
Number of TL fractures ≥20% at baseline, n (%)      
None209 (78.0)209 (78.0)220 (82.1)217 (81.3)855 (79.8)0.79
One only37 (13.8)36 (13.4)33 (12.3)34 (12.7)140 (13.1) 
Multiple21 (7.8)22 (8.2)14 (5.2)16 (6.0)73 (6.8) 
Missing1 (0.4)1 (0.4)1 (0.4)0 (0.0)3 (0.3) 
Fracture risk, n (%)      
Low110 (42.2)113 (42.2)115 (42.9)118 (44.2)456 (42.6)0.89
High153 (55.6)149 (55.6)150 (56.0)143 (53.6)595 (55.6) 
Missing5 (2.2)6 (2.2)3 (1.1)6 (2.3)20 (1.9) 
Incident vertebral fractures at 3 years, n (%)      
No179 (81)186 (84.9)205 (89.9)203 (85.7)773 (85.4)0.07
Yes42 (19.0)33 (15.1)23 (10.1)34 (14.3)132 (14.6) 
History of non-spinal fracture (self-reported), n (%)      
No234 (87.3)238 (88.8)227 (84.7)225 (84.3)924 (86.3)0.64
Yes28 (10.5)26 (9.7)34 (12.7)32 (12.0)120 (11.2) 
Missing6 (2.2)4 (1.5)7 (2.6)10 (3.8)27 (2.5) 
Number of incident non-spinal fractures (full study period), n (%)10 (3.7)19 (7.1)24 (9.0)19 (7.1)72 (6.7) 
4-year cumulative probability of non-spinal fractures (95% CI), %3.1 (1.6–6.1)5.6 (3.4–9.3)7.0 (4.5–10.9)5.5 (3.3–9.1) 0.26

Protocol treatment compliance for AS was 99% for subjects allocated to STAS therapy and 85% for those receiving ITAS therapy. Treatment compliance for ZdA was 77%. The main reason for missed doses of ZdA was to avoid an increased risk of mandibular osteonecrosis resulting from unplanned dental extractions.

Main Study Endpoint: Incident TL Vertebral Fractures at 3 Years

At baseline, 1068 subjects had evaluable TL radiographs. Prevalent fractures were evenly balanced across the treatment arms with a nonsignificant preponderance of fractures in the two STAS arms. Distribution according to vertebral body affected was typical for subjects aged >60 years [31, 33, 34], i.e. bi-modal at T8 and T12.

At 3 years, 905 subjects had TL radiographs available for assessment of incident vertebral fractures. Of the 166 inevaluable subjects, 58 had commenced secondary therapeutic interventions, 36 had died, 28 had withdrawn from the study, and 44 were not evaluable for other miscellaneous causes.

Of the 905 evaluable subjects, 132 (14.6%) experienced incident fractures (i.e. new or worsening prevalent fractures) during the 3-year study period (Table 1).

The interaction term between the two factors tested in the trial (AS duration and the use of ZdA) was near significant at P = 0.077; therefore, this endpoint was assessed in all four treatment arms. In Fig. 1 the proportion of subjects who developed incident vertebral fractures is shown stratified by treatment arm and baseline fracture risk. The odds of vertebral fracture for subjects with high baseline fracture risk were 1.92 (P = 0.001) compared with subjects with low risk. The importance of baseline fracture risk is clearly shown in all four treatment arms, but does not explain the relatively small proportion of subjects experiencing fracture after treatment with ITAS (without ZdA). This finding was confirmed in multivariable analysis where the odds of 3-year vertebral fracture were 0.48 (P = 0.009) in the ITAS arm compared with the control arm (STAS) when adjusted for baseline fracture risk. The use of ZdA resulted in a nonsignificant decrease in 3-year fractures in the STAS+ZdA arm (odds ratio [OR] 0.75, P = 0.26) and the ITAS+ZdA arm (OR 0.69, P = 0.15) when compared with STAS.

Figure 1.

Proportion of incident vertebral fractures at 3 years according to treatment arm and baseline fracture risk (mean ± se). High fracture risk was defined by the presence of one or more of the following conditions at baseline: age ≥80 years, history of non-spinal or symptomatic vertebral fracture, diabetes, alcohol consumption of ≥21 standard drinks per week, current smoker, asymptomatic TL fractures. All other subjects were defined as low risk.

Subsidiary Endpoints

Figure 2 shows the percent change in total hip BMD at 2 and 4 years from baseline by treatment arm for the 222 subjects in the BMD sub-study. It can be seen that AS resulted in BMD loss at the hip, and increasing AS duration produced greater and more sustained losses. Subjects receiving 6 months AS (STAS) experienced a mean decrease of 2.6% in hip BMD at 2 years (P < 0.001), but some recovery was evident at 4 years, although an overall loss of 1.7% was maintained (P = 0.001). BMD loss at 2 years was almost doubled in subjects receiving an additional 12 months AS (ITAS) (4.1%, P < 0.001) with little recovery occurring in the next 2 years (3.7%, P < 0.001).

Figure 2.

Percent change in total hip bone mineral density at 2 and 4 years from baseline by treatment arm.

Figure 2 also shows that 18 months of ZdA was sufficient to prevent ADT-induced BMD losses in both AS groups. Compared with baseline, the mean total hip BMD increases at 2 and 4 years were 0.6% (P = 0.18) and 1.8% (P = 0.003) in the STAS+ZdA arm and 0.5% (P = 0.41) and 1.2% (P = 0.09) in the ITAS+ZdA arm, respectively.

Baseline TL fractures were more common in the nested BMD sub-study than in the entire study population (23.4% and 19.9% respectively), but were similarly distributed across the treatment arms (Tables 1, 2); this difference was nonsignificant. In multivariable analysis, the duration of AS and ZdA use were not significantly associated with incident vertebral fractures at 3 years. Compared with the control arm (STAS) the odds of 3-year fracture were: 0.95 (P = 0.93) in subjects receiving STAS+ZdA, 0.57 (P = 0.30) in subjects receiving ITAS and 0.52 (P = 0.25) in those receiving ITAS+ZdA. BMD T scores at the hip in the osteopenic/porotic range emerged as a more potent predictor of 3-year vertebral fractures (OR 2.66, P = 0.016) than baseline vertebral fractures (OR 1.70, P = 0.19).

Table 2. Pretreatment characteristics of BMD nested sub-study and incident fractures in the study period according to treatment arm (N = 222)
  1. *Comparisons exclude missing values. High fracture risk at baseline defined by presence of one or more of the following conditions: age ≥80 years, history of non-spinal or symptomatic vertebral fracture, diabetes, alcohol consumption of ≥21 standard drinks per week, current smoker, asymptomatic TL fractures. All other subjects were defined as low risk. IQR, interquartile range; BMI, body mass index.

Median (IQR) age

Age group, n (%)

69.5 (65.1–72.8)67.1 (62.7–72.6)66.2 (59.8–72.2)69.1 (66.1–72.2)68.3 (62.7–72.3)0.21
≤70 years39 (61.9)36 (62.1)35 (67.3)32 (65.3)142 (64.0)0.92
>70 years24 (38.1)22 (37.9)17 (32.7)17 (34.7)80 (36.0) 
T stage, n (%)      
T235 (55.6)41 (70.7)39 (75.0)32 (65.3)147 (66.2)0.14
T3,428 (44.4)17 (29.3)13 (25.0)17 (34.7)75 (33.8) 
Gleason score, n (%)      
<76 (9.5)6 (10.3)8 (15.4)3 (6.1)23 (10.4)0.69
738 (60.3)35 (60.3)29 (55.8)35 (71.4)137 (61.7) 
>719 (30.2)17 (29.3)15 (28.9)11 (22.5)62 (27.9) 
PSA group, n (%)      
<10 μg/L18 (28.6)20 (34.5)12 (23.1)16 (32.7)66 (29.7)0.79
10–20 μg/L22 (34.9)23 (39.7)22 (42.3)19 (38.8)86 (38.7) 
≥20 μg/L23 (36.5)15 (25.9)18 (34.6)14 (28.6)70 (31.5) 

Median (IQR) BMI, kg/m2

BMI group, n (%)

26.5 (24.9–30.0)27.6 (25.2–30.1)28.2 (25.6–31.7)28.0 (25.6–30.5)27.7 (25.2–30.5)0.42
≤25 kg/m217 (27.0)11 (19.0)10 (19.2)11 (22.5)49 (22.1)0.86
>25 and <30 kg/m230 (47.6)27 (46.6)24 (46.2)22 (44.9)103 (46.4) 
≥30 kg/m215 (23.8)13 (22.4)18 (34.6)15 (30.6)61 (27.5) 
Missing1 (1.6)7 (12.1)0 (0.0)1 (2.0)9 (4.1) 
Current smoker, n (%)      
No53 (84.1)49 (84.5)49 (94.2)42 (85.7)193 (86.9)0.52
Yes8 (12.7)7 (12.1)3 (5.8)7 (14.3)25 (11.3) 
Missing3 (3.2)2 (3.5)0 (0.0)0 (0.0)4 (1.8) 
Alcohol (standard drinks/week)      
<21 standard drinks/week50 (79.4)45 (77.6)45 (86.5)34 (69.4)174 (78.4)0.27
≥21 standard drinks/week12 (19.0)11 (19.0)7 (13.5)14 (28.6)44 (19.8) 
Missing1 (1.6)2 (3.5)0 (0.0)1 (2.0)4 (1.8) 
Hip BMD      
Median (IQR) hip BMD, gm/cm21.07 (0.93–1.18)1.01 (0.91–1.14)1.03 (0.95–1.13)1.04 (0.95–1.11)1.03 (0.93–1.14)0.48
Hip T score, n (%)      
Normopenic49 (77.8)42 (72.4)38 (73.1)38 (77.6)167 (75.2)0.90
Osteopenic13 (20.6)16 (27.6)13 (25.0)10 (20.4)52 (23.4) 
Osteoporotic1 (1·6)0 (0·0)1 (1.9)1 (2.0)3 (1.4) 
TL fractures at baseline, n (%)      
<20%49 (77.8)39 (67.2)43 (82.7)38 (77.6)169 (76.1)0.35
≥20%14 (22.2)18 (31.0)9 (17.3)11 (22.5)52 (23.4) 
Missing0 (0.0)1 (1.7)0 (0.0)0 (0.0)1 (0.5) 
Number of TL fractures ≥20% at baseline, n (%)      
None49 (77.8)39 (67.2)43 (82·7)38 (77·5)169 (76·1)0.38
One only11 (17.5)9 (15.5)5 (9·6)7 (14·3)32 (14·4) 
Multiple3 (4.7)9 (15·5)4 (7·7)4 (8·2)20 (9·0) 
Missing0 (0.0)1 (1.7)0 (0.0)0 (0.0)1 (0.5) 
Fracture risk, n (%)      
Low24 (38.1)19 (32.8)23 (44.2)16 (32.7)82 (36.9) 
High28 (44.4)29 (50.0)25 (48.1)28 (57.1)110 (49.6) 
Missing11 (17.5)10 (17.2)4 (7.7)5 (10.2)30 (13.5) 
Incident vertebral fractures at 3 years, n (%)      
No41 (78.8)37 (77.1)41 (85.4)37 (84.1)156 (81.2)0.70
Yes11 (21.2)11 (22.9)7 (14.6)7 (15.9)36 (18.8) 

At baseline, 120 subjects had a self-reported history of non-spinal fractures. Incident fractures occurred in 72 subjects during the entire study period (median [interquartile range] follow-up 5.6 [4.7–6.5] years). The interaction term for the two factors tested in the trial was near significant at P = 0.06. Further analysis was therefore undertaken by trial arm. The cumulative probability of non-spinal fractures by treatment arm is shown in Fig. 3. At 4 years, the cumulative probabilities were STAS 3.1% (95% CI 1.6–6.1), STAS+ZdA 5.6% (95% CI 3.4–9.3), ITAS 7.0% (95% CI 4.5–10.9), and ITAS+ZdA 5.5% (95% 3.3–9.1).

Figure 3.

Cumulative probability of non-spinal fractures by treatment arm.

In multivariable analysis adjusted for baseline fracture risk, subjects receiving ITAS were at the highest risk of experiencing non-spinal fracture during the entire study period (HR 2.55, P = 0.013) when compared with the control arm (STAS). Subjects in both ZdA arms had nonsignificantly higher risks than those receiving STAS. Hazard ratios (HRs) were 1.94 (P = 0.09) in subjects receiving STAS+ZdA and 1.96 (P = 0.08) receiving ITAS+ZdA. Baseline fracture risk was an independent predictor of non-spinal fractures (HR 1.75, P = 0.027).

Study Drug Toxicities

Osteonecrosis of the mandible was observed in two subjects treated with 18 months of ZdA, one in each of the AS arms (both of whom fully recovered). ZdA was not associated with a decline in renal function irrespective of AS duration. The frequency of grade one-corrected serum calcium levels <2.15 mmol/L in subjects ranged from 2.7 to 8.8% between randomization and 24 months; however, no adverse events occurred as a result.


The present study has produced apparently conflicting findings. Our nested BMD study provided a result that we were expecting, namely that ITAS causes greater reductions in BMD at the hip than STAS, and that 18 months of ZdA prevents these losses; however, compared with STAS, ITAS did not increase incident vertebral fractures in the entire study cohort over the 3-year study period. In fact incident vertebral fractures were significantly less frequent in the ITAS trial arm than in the STAS (control) arm in multivariable analysis adjusted for baseline fracture risk. Moreover, 18 months of ZdA did not significantly reduce fractures in subjects receiving 6 or 18 months’ AS.

The first question raised is whether the study sample size was an important contributor to this outcome. It was estimated that the study was underpowered to detect an increase smaller than 46% in incident vertebral fractures in the two ITAS arms. The observed difference of a −4.79% decrease in fractures in the two ITAS arms (99% CIs −10.85%,1.27%) suggests that any true increase in fractures in the ITAS arms is likely to be smaller than 46% (data not shown). This degree of limitation is important as will be discussed later on. Since an increase in trial sample size was not an option, it will be asked whether an increase in study power could have been achieved by increasing the 3-year study duration. Unfortunately, relevant data did not become available until 2005 when the RADAR trial was well under way. These data, from Shahinian et al. [12] and Smith et al. [35], indicated that doubling the study duration from 3 to 6 years would increase the absolute difference in fractures between the STAS and ITAS (without ZdA) trial arms 2.2–2.5-fold (our estimates); however, this opportunity was not taken. We believed that vertebral study dropout rates would be very high from the 4th follow-up year onward because of the use of salvage AS and through other causes such as death or withdrawal because of intercurrent medical disorders. Indeed at 3 years of follow-up, 15% of subjects had dropped out of the study.

The observed difference is also compatible with an imbalance in explanatory factors across the treatment arms. This is quite plausible because the RADAR trial was not deliberately designed to assess differences in fractures and, therefore, was not stratified for any of the known baseline factors that can contribute to fractures. This may be quite important because in the nested BMD sub-study, we found that osteoporotic or osteopenic T scores at baseline were an even stronger predictor of incident vertebral fractures than baseline fracture risk. This raises the possibility that, had baseline measures of BMD at the hip been available in the entire study cohort, the unexpected study result could have been explained. As it stands, our exploratory analyses were not able to explain the study result and this leads to the question of what other important baseline risk factors were missing from our dataset. The only relevant epidemiological study in Australian men is the Dubbo Osteoporosis Epidemiology Study [36]. This study identified predictive factors for osteoporotic fractures in elderly men that were not collected in our study, including quadriceps strength and sway factor (a measure of postural stability). Another known risk factor for fractures missing from our dataset was family history of fractures [37, 38].

Given that the expected influence of the study drugs on vertebral fractures in the present study was so limited, the second question is raised as to whether the expectation was reasonable that incident vertebral fractures would be increased in the ITAS arm when compared with the STAS (control) arm over a 3-year time period. The literature regarding vertebral fractures after 18 months AS is weak. Shahinian et al. [12] found that fracture rates were not higher after 4 months AS, but were after longer durations, particularly after 9 months or more. The relative risk of fractures in men receiving >9 months of AS compared with those receiving 4–9 months of AS was in the range 1.33–1.4 (our estimation). Smith et al. [35]made a similar finding for AS durations beyond 1 year. Those receiving >1 year of AS had a relative risk of fractures of ∼1.15 (our estimation) compared with those receiving <1 year. Shao et al. [28] found that the risk of subsequent fracture in men receiving adjuvant AS suppression with curative RT (as in the present study) or prostatectomy treatment was lower than in men receiving AS alone in the palliation of cancers considered untreatable by RT or prostatectomy. The latter group of men would most likely have had a higher risk of fractures and have received longer durations of AS. We estimated that the relative risk of fractures would have been almost double in subjects receiving long-term AS or orchidectomy for the palliation of advanced disease when compared with those receiving shorter courses as an adjuvant measure. The relative risk of fractures in subjects with apparently localized cancers who receive 18 months of AS in the adjuvant setting when compared with 6 months is therefore unknown. Since subjects receiving adjuvant AS are likely to be at lower risk of fracture than subjects with advanced disease we believe that these three studies suggest that 18 months adjuvant AS will increase the risk of fractures when compared with 6 months adjuvant AS by <1.5-fold. This is a difference that is smaller than the present study had the power to detect, as stated earlier. We must therefore emphasize that the negative fracture findings in the present study should not be taken to mean that 18 months’ adjuvant AS does not cause more fractures than 6 months’ AS. Our BMD findings indicate that losses in BMD are prolonged after both 18 and 6 months’ AS and warn that fracture risk will be increased even by AS durations as short as 4 months, as indicated by Shahinian et al. [12].

The effectiveness of ZdA in the doses used in the present study in preventing fractures in men with apparently localized cancers given 18 months of AS is less open to doubt. The handful of randomized trials that have tested the value of ZdA and other antiresorptive agents indicate that 18 months’ ZdA will prevent the majority of fractures that have an osteopenic contribution in men receiving 18 months’ AS [19, 27, 39]. The two recent large-scale trials conducted by Smith et al. [40, 41] showed that denosumab and toremifene were effective in preventing BMD loss, and approximately halved incident vertebral fractures, in men at far greater risk of vertebral fractures than our subjects. These men, who had already received sometimes lengthy durations of AS for advanced prostate cancer, had much higher baseline hip T score rates of osteopenia (56.3–63.1%) and osteoporosis (14.7–23.6%) than in those in the present nested BMD sub-study where the corresponding rates were 23.4 and 1.4%. In these trials, therefore, changes in BMD attributable to the study drugs corresponded well with vertebral fractures. Although low BMD at the hip is a risk factor for vertebral fractures, the inverse relationship between BMD and fractures is not straightforward. Some authors believe that there is little evidence to support the proposition that BMD change is a surrogate for fracture risk and that vertebral fracture risk, and its reduction by intervention, corresponds poorly to changes in BMD [42]; however most authors accept that low BMD contributes to <50% of all incident fractures [36, 43, 44]. In our nested sub-study, changes in BMD caused by the study drugs did not correspond with fractures. This leads us to speculate that a relationship between decreasing BMD and increasing vertebral fracture exists above a certain threshold in vertebral fragility. Below this limit, modest BMD decreases (∼5%) will not increase vertebral fractures. Unfortunately, too few fractures occurred in the present sub-study to corroborate this suggestion. Irrespective of this speculation, we must emphasize that the fracture results from the present trial should not be taken to mean that ZdA does not have a prophylactic role in the prevention of AS induced fractures. The present study does confirm, however, that baseline BMD at the hip is a strong predictor of subsequent fracture, and that ZdA can prevent the prolonged loss of BMD causes. In fact our study indicates that consideration of its prophylactic use in men with low T scores at the hip and evidence of loss of vertebral height of ≥20% in one or more TL vertebrae would be very wise.

Although the hazard of non-spinal fractures in subjects receiving ITAS was greater than in those receiving STAS (HR 2.55, P = 0.013), as might be expected, the hazard was almost as great in subjects receiving STAS + 18 months ZdA as in those receiving ITAS. This unexpected finding may also be a power issue resulting from the relatively small event rate (i.e. 72 subjects with non-spinal fractures in the entire cohort over a median follow-up duration of 5.6 years, as well as unknown baseline risk factors for fracture.

A final limitation in the study design needs to be acknowledged. The same funding constraints that restricted DEXA studies to a nested sub-study of subjects powered to detect potentially clinically relevant study drug effects also prevented serial measures of bone turnover markers and oestradiol levels. All of the limitations in our study design discussed above lead us to recommend that future randomized studies of adjuvant AS-induced fractures in men with apparently localized prostate cancer build serial DEXA and bone turnover markers into the study design in all subjects. Moreover, future studies should be stratified for baseline fracture risk and BMD at the hip, at the very least. Our study also warns that sample size requirements may be high and study duration should be longer for reliable fracture rate effect sizes to be estimated.

It will be asked what the practising urologist or oncologist will learn from this study, in the absence of data from very much larger prospective randomized trials. As discussed above, our BMD study has been more helpful in this regard than our fracture study. There is no doubt that adjuvant AS will remain in common use in the many men who are still diagnosed with locally advanced cancers; therefore, it is important to know what to do in this setting. We feel that it is necessary, amongst other investigations, to assess renal function and vitamin D levels, order TL X-rays for evidence of loss of vertebral height of ≥20%, and a DEXA scan, that includes the hip, to determine if there is evidence of osteopenia or osteoporosis (particularly T scores <2 [45]). As Shahinian [46] has suggested, prophylactic vitamin D and calcium administration may be wise in all cases. In our view, prophylactic ZdA would also be sensible in all cases with evidence of low BMD at the hip and one or more TL vertebral compressions ≥20%; however, as suggested by Michaelson et al. [39], annual dosing may be all that is required.


This study was funded by the National Health and Medical Research Council (Canberra, ACT, Australia), Novartis Pharmaceuticals, Abbott Pharmaceuticals (both Sydney, NSW, Australia), New Zealand Health Research Council, New Zealand Cancer Society, University of Newcastle (Australia), Hunter Medical Research Institute (New Lambton Heights, NSW, Australia) Calvary Mater Radiation Oncology Fund (Waratah, NSW, Australia), and Maitland Cancel Appeal (Rutherford, NSW, Australia). We thank Rosemary Bradford (Prostate Cancer Trials Group, University of Newcastle) for her skilful preparation of the report.

Conflict of Interest

James Denham: Novartis Pharmaceuticals – grant. Abbott Pharmaceuticals – supply of leuprolide in subjects with T2 cancers during conduct of the study. Nigel Spry: Abbott Pharmaceuticals – personal fees. AstraZeneca – travel assistance. Outside submitted work. Kevin Lynch: Novartis – full time employee until 02/2008 during conduct of the study. Jean Ball: Novartis Pharmaceuticals – Educational grant during conduct of study.


bone mineral density


androgen suppression


zoledronic acid




dual-energy radiograph absorptiometry


short-term AS


intermediate-term AS


odds ratio