The association between radiographic response and overall survival in men with metastatic castration-resistant prostate cancer receiving chemotherapy




In men with metastatic castration-resistant prostate cancer (CRPC), the association of measurable tumor responses with overall survival (OS) is unknown. The authors retrospectively evaluated the TAX327 phase 3 trial to study this relation.


Eligible patients for this analysis included those with World Health Organization (WHO)-defined measurable metastatic disease randomized to receive either docetaxel or mitoxantrone. OS was estimated by using the Kaplan-Meier method, and the prognostic relation of WHO-defined radiologic response with OS was performed by using Cox proportional hazards regression. Landmark analyses evaluated survival from baseline and at 2, 3, 4, and 6 months after baseline.


Four hundred twelve patients enrolled on the TAX327 trial had measurable tumors. Thirty-seven patients exhibited a complete or partial objective response (CR/PR, 9.0%), 116 had stable disease (SD, 28.2%), 99 had progressive disease (PD, 24%) and 160 (38.8%) did not have a after-baseline objective assessment. Partial responders demonstrated longer median OS (29.0 months) than patients with SD (22.1 months) or those with PD (10.8 months) or those who were not assessed (12.7 months). These results remained after landmark analysis. The authors found a significant association between ≥30% prostate-specific antigen (PSA) declines and radiologic response, with ≥30% PSA declines occurring in all patients with CR/PR, 79.8% of patients with SD, and 34.4% with PD. Radiologic response remained a significant but modest post-treatment prognostic factor for OS after adjusting for treatment, pain response, and ≥30% PSA decline (P = .009).


In men with metastatic CRPC and measurable disease receiving chemotherapy, objective tumor response was prognostic for OS and appeared to complement PSA assessment. Cancer 2011. © 2011 American Cancer Society.

The determination of objective response to therapy in men with metastatic castration-resistant prostate cancer (CRPC) is plagued by the presence of nonmeasurable bony metastatic disease in the majority of patients who are beginning chemotherapy.1, 2 Retrospective studies of the Southwest Oncology Group (SWOG)-9916 and the TAX327 trials, which compared docetaxel-based with mitoxantrone-based chemotherapy, demonstrated that a prostate-specific antigen (PSA) decline ≥30% or ≥50% within 3 months had a moderate degree of surrogacy for extended overall survival (OS) in the setting of cytotoxic chemotherapy.3, 4 Retrospective analyses of these trials have also demonstrated that an increase in PSA at 3 months correlates with poor overall survival (OS).5, 6 A recent retrospective study demonstrated that progression-free survival (PFS), defined by a composite endpoint (progression by bone scan, PSA criteria by PSA working group, and measurable tumor progression), also correlates with poorer OS.7 However, assessment of bone-scan progression may be hampered by the bone healing or flare response, and PSA progression may not apply in the setting of biologic agents.8 Tumor progression may not be optimally reflected by clinical progression (pain progression or skeletal events).9 Although circulating tumor cells (CTCs) and functional imaging are emerging as promising surrogates for clinical outcomes, further validation is required, and their role in clinical decisions is unclear.10-12

An inexpensive, readily available and objectively measured clinical parameter that enhances the ability to predict enhanced OS is desirable. A small, but substantial proportion, of men have measurable soft tissue, lymph node, or visceral metastasis. Although objective evaluation of measurable soft tissue or visceral disease is feasible, the correlation of objective responses with long-term outcomes is unclear.13-15 Measurable tumor response may complement and enhance the power of PSA decline and/or pain improvement as an intermediate surrogate for long-term outcomes.4, 16 We, therefore, conducted a retrospective study of the TAX327 trial to identify a potential association between measurable tumor responses (with or without PSA decline) and OS.


Patient Population in the TAX327 Trial and Eligibility for Retrospective Analysis

The TAX327 trial was a randomized 3-arm trial of 1006 men that compared first-line docetaxel-prednisone every 3 weeks (D3W), weekly docetaxel-prednisone (DW), or mitoxantrone-prednisone (MP) every 3 weeks in patients with progressive metastatic CRPC. Patients were recruited from March 2000 through June 2002 after institutional internal review board approval at participating institutions.1, 17 Eligible men had metastatic prostatic adenocarcinoma and evidence of progression despite androgen-deprivation therapy defined clinically or by radiographic or PSA criteria. Treatment was continued for up to 30 weeks or until toxicity, progression, or death. Imaging studies were performed at intervals of 6 to 9 weeks and repeated after 4 weeks to identify those with a response. For patients with at least 1 bidimensionally measurable lesion, tumor response was evaluated by World Health Organization (WHO) criteria, for which the product of the largest diameter and its perpendicular are summed for predefined lesions.18 Four WHO response categories are defined: complete response (CR), partial response (PR), stable disease (SD), and progressive disease (PD). CR requires disappearance of all known disease, PR is a ≥50% decrease in tumor area, and PD is ≥25% increase in area or the appearance of ≥1 new lesion, each maintained for at least 2 successive evaluations. A PSA response was defined as a reduction from baseline of ≥50% that was maintained for ≥3 weeks, a pain response was defined as a 2-point reduction in the present pain index (PPI) score from baseline without an increase in the analgesic score or as a reduction of ≥50% in the analgesic score without an increase in the PPI score, either of which was maintained for ≥3 weeks. To be eligible for the current retrospective analysis, patients were required to have measurable tumors at baseline according to WHO criteria. The present analysis includes updated survival data that includes information on 867 deaths known to have occurred by March 2007.17

Statistical Considerations

The Kaplan-Meier method was used to estimate OS. As it is possible that there may be a time bias related to objective response status, and because date of response was not recorded (ie, only best objective response while on study was recorded), a landmark analysis was performed at 60, 90, 120, and 180 days after baseline.19 In essence, this resets the baseline date to the landmark date, and data analysis is performed from that point forward. Responses by WHO criteria (CR, PR, SD, PD, or not evaluable [NE]) were assessed as categorical variables because degree of tumor shrinkage was not available. The chi-square test was used to test for associations between tumor response and other outcomes of interest, such as PSA decline and pain response, as well as baseline prognostic factors of interest, such as risk groups and presence of metastases in bone, liver, and lymph nodes.8, 20, 21 Best overall response was assessed as a prognostic factor for OS by using Cox proportional hazards regression. For the multivariate model, only patients who were evaluable for pain response, ≥30% PSA decline, and objective tumor response were included (ie, n = 168 for OS from baseline and n = 120 for OS from day 180). All models included stratification factors for baseline pain and Karnofsky performance status in accordance with the initial analysis. Assessment of objective tumor response as a potential surrogate for OS was performed by using the likelihood reduction factor (LRF) on the full model.22 The LRF is a measure of the effect of the prognostic factor (ie, tumor objective response status) on the outcome (ie, OS) and is measured on a scale from 0 to 1. Although no definitive guidelines exist, an optimal surrogate will have an LRF close to 1.


Patient Characteristics

Of 1006 patients enrolled in the TAX327 trial, 412 (41%) patients were evaluable at baseline for objective tumor response assessment (Table 1), 369 of whom had died as of the updated dataset in March 2007. These patients were evenly distributed among the 3 treatment groups (Table 2). Three (0.7%) patients had a CR and were combined with the 34 (8.3%) patients with PR for all statistical analyses, with a numerically greater number of CR/PR and fewer PD results observed among D3W-treated patients when compared with the other treatment arms. SD was observed in 116 (28.2%) men and PD in 99 (24%) men. Response of measurable disease was not evaluable (NE) in 160 (38.8%) men because of removal from trial before objective evaluation or lack of confirmatory radiologic evaluation.

Table 1. Characteristics of Patients With Measurable Tumors Versus No Measurable Tumors
ParameterTumor Response EvaluableTumor Response Not Evaluable
  • PSA indicates prostate-specific antigen; CI, confidence interval.

  • a

    Patients evaluable for PSA response were those with a baseline PSA level of ≥20 ng/mL.

  • b

    ≥50% reduction in PSA from baseline maintained for at least 3 weeks.

  • c

    Patients evaluable for pain response were those with a score of ≥2 on the Present Pain Intensity scale or an analgesic score of ≥10 averaged over the previous week at baseline.

Treatment group (%)  
 D3W141 (34.2)194 (32.7)
 DW134 (32.5)200 (33.7)
 MP137 (33.3)200 (33.7)
Gleason score (%)  
 ≤7155 (37.6)263 (44.3)
 8-10137 (33.3)163 (27.4)
 Not available120 (29.1)168 (28.3)
Karnofsky performance-status ≤70 (%)70 (17.0)60 (10.1)
Serum PSA, median, ng/mL129106
Extent of disease (%)  
 Bone metastases342 (83.0)574 (96.6)
 Visceral disease154 (37.4)75 (12.6)
Baseline alkaline phosphatase  
Baseline hemoglobin  
Evaluable for PSA response (%)a361 (87.6)512 (86.2)
PSA response (%)  
 Per protocolb143 (39.6)219 (42.8)
 ≥30% reduction170 (47.1)263 (51.4)
Evaluable for pain response (%)c186 (45.1)278 (46.8)
Pain response (%)52 (28.0)83 (29.9)
Overall survival  
 Median (95% CI)15.3 (14.1-16.5)19.2 (18.4-20.7)
 Number censored (%)43 (10.4)96 (16.2)
Table 2. Patient Characteristics of Eligible Population With Measurable Tumors
Best Objective Tumor Response StatusAll Patients No. (%)D3W No. (%)DW No. (%)MP No. (%)
  1. D3W indicates docetaxel-prednisone every 3 weeks; DW, weekly docetaxel-prednisone; MP, mitoxantrone-prednisone; CR, complete response; PR, partial response; SD, stable disease; PD, progressive disease; NE, not evaluable.

 CR3 (0.7)2 (1.4)0 (0.0)1 (0.7)
 PR34 (8.3)15 (10.6)11 (8.2)8 (5.8)
 SD116 (28.2)40 (28.4)46 (34.3)30 (21.9)
 PD99 (24.0)26 (18.4)31 (23.1)42 (30.7)
 NE160 (38.8)58 (41.1)46 (34.3)56 (40.9)

Association of Objective Responses With Survival

The 594 patients who were inevaluable (IE) for response had longer median OS (19.2 months vs 15.3 months) and were more likely to have censored survival data (16.2% vs 10.4%) than response evaluable patients. Of patients evaluable for tumor response, those with an objective response had longer median survival from baseline (29.0 months) than those with stable disease (22.1 months), who in turn had longer survival than those patients with either PD (10.8 months) or NE disease (12.7 months; Fig. 1a and b). For D3W patients only, median survival from baseline was 29.0, 23.4, 12.9, 13.1 months for CR/PR (n = 17), SD (n = 40), PD (n = 26), and NE (n = 58) patients, respectively. This corresponds to hazards ratios of 0.46 (95% confidence interval [CI], 0.24-0.86; P = .016), 0.48 (95% CI, 0.30-0.77; P = .003), and 1.45 (95% CI, 0.89-2.38; P = .14) for D3W patients with CR/PR, SD, and PD, respectively compared with NE patients. There was no significant difference in OS among patients having CR/PR between the different treatment groups (P = .96). Results were very similar at each landmark time; thus, only OS from baseline and at 180 days are discussed.

Figure 1.

(a) Median overall survival from baseline, and (b) from day 180 landmark, and 95% confidence intervals by tumor response and treatment group are shown. Survival according to objective tumor response refers to the evaluable population (N = 412) with D3W (n = 141), DW (n = 134), and MP (n = 137) patients. The IE population refers to those without objectively measurable tumors (N = 594). CR indicates complete response; PR, partial response; SD, stable disease; NE, not evaluable; IE, inevaluable for objective tumor response.

Association of Objective Response With PSA Response and Pain Response

There was a statistically significant association (P < .001) between objective tumor response and ≥30% and ≥50% PSA declines (Table 3). There were 361 patients evaluable for both PSA and objective tumor measurements. Of these patients, all 32 patients who had an objective tumor response also had a confirmed ≥30% PSA reduction within 3 months (all 32 patients also had PSA reductions of ≥50%; however, a confirmatory ≥50% PSA measurement occurred in only 31 of these 32 patients). In addition, confirmatory ≥30% PSA reductions were observed in 79 of 99 (79.8%) patients with objective SD, 31 of 90 (34.4%) patients with objective PD, and 28 of 140 (20.0%) of NE patients. Similar results were observed when employing confirmed ≥50% reduction in PSA, with 31 of 32 (96.9%) of CR/PR patients, 69 of 99 (69.7%) of SD patients, 22 of 90 (24.4%) of PD patients, and 21 of 140 (15.0%) of NE patients having a PSA response. The sole patient with an objective tumor response without a confirmed PSA response attained a ≥50% reduction in PSA, but it was not confirmed until more than 3 weeks later. Objective tumor response was also associated (P = .007) with occurrence of a pain response (Table 3), with 4 of 8 (50%) of CR/PR patients, 20 of 43 (46.5%) of SD patients, 11 of 55 (20.0%) of PD patients, and 17 of 80 (21.3%) of NE patients having a pain response. Of 168 patients who were evaluable for all 3 categories of response (pain, PSA, and objective), only 4 attained all of these responses, indicating the relative rarity of this event.

Table 3. Relation Between Best Objective Tumor Response and Outcomes
 No. EvaluableCR or PRSDPDNE
  • CR indicates complete response; PR, partial response; SD, stable disease; PD, progressive disease; NE, not evaluable; PCWG2, Prostate Cancer Clinical Trials Working Group.

  • a

    Risk group based on baseline pain, visceral metastases, anemia, and progression by new bone scan lesion at baseline.

Measurable tumors41237 (9.0%)116 (28.2%)99 (24.0%)160 (38.8%)
≥30% reduction in PSA36132 (100.0%)79 (79.8%)31 (34.4%)28 (20.0%)
≥50% reduction in PSA36131 (96.9%)69 (69.7%)22 (24.4%)21 (15.0%)
Pain-response1864 (50.0%)20 (46.5%)11 (20.0%)17 (21.3%)
Bone mets at baseline34224 (7.0%)86 (25.1%)92 (26.9%)140 (40.9%)
No bone mets at baseline7013 (18.6%)30 (42.9%)7 (10.0%)20 (28.6%)
Visceral disease at baseline1548 (5.2%)44 (28.6%)45 (29.2%)57 (37.0%)
No visceral disease at baseline25829 (11.2%)72 (27.9%)54 (20.9%)103 (39.9%)
Liver metastases at baseline513 (5.9%)7 (13.7%)22 (43.1%)19 (37.3%)
Lymph node metastases at baseline30931 (10.0%)94 (30.4%)78 (25.2%)106 (34.3%)
PCWG2 risk groups     
 Node only disease4210 (23.8%)18 (42.9%)4 (9.5%)10 (23.8%)
 Bone±Nodes without visceral disease21319 (8.9%)52 (24.4%)50 (23.5%)92 (43.2%)
 Visceral disease1548 (5.2%)44 (28.6%)45 (29.2%)57 (37.0%)
Risk groupa     
 Good (0-1 factor)12823 (18.0%)42 (32.8%)15 (11.7%)48 (37.5%)
 Intermediate (2 factors)1369 (6.6%)37 (27.2%)37 (27.2%)53 (39.0%)
 Poor (3-4 factors)1485 (3.4%)37 (25.0%)47 (31.8%)59 (39.9%)

Predictive Factors for Objective Response

A recent analysis of the same TAX327 trial identified the following 4 factors as predictive for ≥30% PSA declines and OS: baseline pain, visceral metastases, anemia, and progression by new bone-scan lesions at baseline.23 Based on these factors, 3 risk groups were identified with different outcomes: good (0-1 factors), intermediate (2 factors), and poor (3-4 factors). These risk groups were also significantly associated with objective tumor response and tumor nonprogression (P < .001, for both). Objective response rate and tumor nonprogression rate were 23 of 128 (18.0%) and 65 of 128 (50.8%) for the good prognosis group, 9 of 136 (6.6%) and 46 of 136 (33.8%) for the intermediate prognosis group, and 5 of 148 (3.4%) and 42 of 148 (28.4%) in the poor prognosis group. Among D3W patients only, the objective response and tumor nonprogression rates were 10 of 46 (21.7%) and 25 of 46 (54.4%), 6 of 48 (12.5%) and 17 of 48 (35.4%), and 1 of 47 (6.7%) and 15 of 47 (31.9%), for the good, intermediate, and poor prognosis groups, respectively. The median OS demonstrated a substantial divergence for patients based on tumor response status and risk group (Fig. 2a).

Figure 2.

(a) Median overall survival and 95% confidence intervals (CI) from baseline based on tumor response and risk groups and (b) presence of bone metastases at baseline are shown. CR indicates complete response; PR, partial response; SD, stable disease; NE, not evaluable; IE, inevaluable for objective tumor response.

The possibility of differential objective response based on the presence of concurrent bone metastasis was studied. There were 342 patients eligible for objective response assessment with bone metastases present at baseline, whereas 70 patients did not have bone metastases. No patient with bone metastasis at baseline had a CR, 24 (7.0%) patients had a PR, and 86 (25.2%) patients had SD. Seventy patients evaluable for objective response did not have bone metastases at baseline, of whom 3 (4.3%) had a CR, 10 (14.3%) had a PR, and 30 (42.9%) had SD. Thus, both a favorable risk group and lack of bone metastases at baseline appeared to predict objective tumor response, and all 3 of these variables were independently prognostic for OS (Fig. 2b). In addition, the Prostate Cancer Working Group (PCWG)-2 recommended prognostic groups also appeared to predict for objective responses; those with lymph node only metastasis, bone plus lymph node, and those with visceral metastasis demonstrated responses in 23.8%, 8.9%, and 5.2%, respectively (Table 3).8

Combinations of Objective, PSA, and Pain Responses as Prognostic Factors for Survival

The final multivariate model included only patients who were evaluable for pain response, ≥30% PSA decline, and objective tumor response (ie, n = 168 for OS from baseline and n = 120 for OS from day 180). Objective tumor response (P = .018), pain response (P = .004), and ≥30% PSA declines (P = .0012) were all independently significant prognostic factors for OS (Table 4). Objective tumor response remained a significant but modest post-treatment prognostic factor for OS after adjusting for treatment, pain response, and ≥30% PSA decline (P = .009). However, when studying the 168 patients eligible at baseline for evaluating all 3 types of response (∼15% of the total population studied), the LRF for the impact of objective tumor response on OS improvement ranged from 0.08-0.17.

Table 4. Best Overall Response as a Predictor of Overall Survival in Men Evaluable for Pain-Response, ≥30% PSA Decline and Objective Tumor Response
Baseline Model, n=168
CovariateHR (95% CI)P
Docetaxel Q3Wb1.10 (0.73-1.65)0.65
Docetaxel weeklyb1.07 (0.72-1.59)0.75
≥30% PSA decline0.62 (0.43-0.90)0.012
Pain response0.58 (0.40-0.84)0.004
Tumor PR or CRc0.31 (0.11-0.94)0.038
Tumor SDc0.65 (0.40-1.04)0.073
Tumor PDc1.23 (0.84-1.80)0.3
Tumor NEReference
180 Day Model, n=120
CovariateHR (95% CI)P
  • PSA indicates prostate-specific antigen; HR, hazard ratio; CI, confidence interval; PR, partial response; CR, complete response; SD, stable disease; PD, progressive disease; NE, not evaluable.

  • aAll tests included stratification for baseline Karnofsky Performance Score (KPS) and baseline pain.

  • b

    Hazard ratio compared with mitoxantrone (adjusted for PSA, pain and radiographic response).

  • c

    Hazard ratio compared with tumor status not evaluable.

Docetaxel Q3Wb0.86 (0.52-1.42)0.56
Docetaxel weeklyb1.01 (0.62-1.63)0.98
≥30% PSA decline0.74 (0.48-1.15)0.18
Pain response0.76 (0.50-1.16)0.21
Tumor PR or CRc0.28 (0.09-0.85)0.024
Tumor SDc0.56 (0.33-0.97)0.039
Tumor PDc1.01 (0.63-1.65)0.95
Tumor NEReference

Patients who did not achieve an objective tumor response, ≥30% PSA decline, or pain response had the shortest median survival of only 6.8 months from baseline (Table 5). In comparison, the median survival from baseline of patients with both objective tumor response and ≥30% PSA declines was the longest at 29.0 months (95% CI, 19.7-39.5). When examining OS based on objective tumor PD versus non-PD, those with non-PD (n = 153) demonstrated a median OS from baseline of 22.4 months (Table 5). Patients with both non-PD and ≥30% PSA declines (n = 111) demonstrated a median OS of 23.0 months. Pain response did not appear to be associated with an extension of survival in patients with non-PD. There were 31 patients who had a best objective response of PD, yet achieved a ≥30% reduction in PSA. All 31 had bone metastases at baseline, 29 had lymph node metastases, and 30 had soft tissue disease, while only 3 and 10 patients, respectively, had liver and visceral metastases. Of these 31 men, 29 are known to have died, and median survival from baseline was 14.1 months (95% CI, 11.0-15.8). Of note, bone-scan progression was not evaluated separately from the overall WHO criteria in this trial to accurately assess for bone-scan flare. In contrast, 59 men (58 known to have died) had a best objective response of PD and failed to achieve a 30% PSA response. The median survival from baseline of these 59 men was 9.4 (95% CI, 5.6-11.6) months.

Table 5. Overall Survival From Baseline by Response Statusa
Response StatusNo. EvaluableNo. with OutcomesMedian OS (95% CI)
TumorPSA Decline Within 3 MonthsPain Response
  • PSA indicates prostate-specific antigen; OS, overall survival; CI, confidence interval; NC, not calculated due to small numbers.

  • a

    For 412 patients with measurable tumors at baseline according to WHO criteria.

Nonresponse<30%No168846.8 (5.4-8.9)
ProgressionAll patientsAll patients412509.8 (5.6-11.8)
All patients≥30%All patients36117020.4 (17.7-22.6)
All patientsAll patientsYes1865216.8 (13.8-20.5)
All patients≥30%Yes1682619.1 (15.9-23.9)
NonprogressionAll patientsAll patients41215322.4 (20.6-24.4)
Nonprogression≥30%All patients36111123.0 (21.1-25.9)
NonprogressionAll patientsYes1862418.9 (15.9-23.9)
Nonprogression≥30%Yes1681918.9 (15.2-23.9)
ResponseAll patientsAll patients4123729.0 (19.6-39.5)
Response≥30%All patients3613229.0 (19.7-39.5)
ResponseAll patientsYes1864NC


To our knowledge, this is the first study demonstrating an association between objective tumor response and OS in the setting of modern chemotherapy for metastatic CRPC. This retrospective analysis of the TAX327 trial identified an association of objective tumor responses by WHO criteria and favorable survival. Those exhibiting both WHO responses and ≥30% PSA declines demonstrated the best median OS (29 months), whereas those with ≥30% PSA decline without objective responses demonstrated a median OS of 20.4 months. We also noted improved survival among patients who achieved SD radiologically as their best response, which has important prognostic and treatment implications for the clinic. The proportion of patients with measurable tumors may be higher with improved computerized tomography (CT) technology and after administration of docetaxel, eg, in the recent trial comparing second-line cabazitaxel versus mitoxantrone, the proportion of patients with measurable disease was >50%.24 Therefore, the use of objective tumor response as a signal of activity may increase as a function of optimized radiographic imaging and in more heavily pretreated settings. Moreover, the evaluation of objective response entails no additional costs in addition to routine CT imaging, whereas CTCs or functional imaging incur substantial additional costs. However, we recognize that the lack of measurable disease in the majority of men with metastatic CRPC will limit the use of radiographic response as a surrogate and that the major clinical use of this assessment will be for prognostic purposes and in the early assessment of efficacy among men with measurable disease. A higher proportion of patients with objective responses in addition to PSA declines may provide a signal of better antitumor activity in the context of cytotoxic agents. Therefore, these data may assist in screening the activity of new cytotoxic agents in phase 2 trials through the use of composite measures of response (pain, PSA, radiologic, and circulating tumor cell alterations for example).11

These data are consistent with those for other metastatic solid tumors that have been used to evaluate the relation between objective response and either OS or PFS.25-27 Conversely, a retrospective study of 124 men with metastatic CRPC treated on 3 different protocols elicited the difficulties of categorizing changes in the individual sites of disease into a CR, PR, or SD, which were thought to render the best overall response parameter of limited value.15 For noncytotoxic agents (ie, antiangiogenic agents, immunologic agents, cytostatic targeted therapies), the relation between radiographic response, PFS, and OS is unknown at this time in CRPC. While all patients with measurable tumors and objective responses exhibited PSA declines, only a proportion of patients with PSA declines exhibited objective responses. Objective responses appeared to indicate a better quality of response translating to better OS. Intriguingly, a small subset of 31 patients that exhibited PSA declines ≥30% coupled with PD of measurable tumors demonstrated a median survival of 14.5 months. Notably, progression by composite criteria including objective progression has been found to predict OS.7 Future studies employing current Prostate Cancer Working Group (PCWG)-2 criteria may be better able to evaluate this issue.8 The clinical relevance of responses in small-volume measurable disease in the context of large-volume bony metastatic disease may be questioned. However, objective responses appear to be associated with better OS with or without underlying bone metastasis, although absolute survival is shorter in those with bone metastasis. Those with no bone metastasis had a higher chance of objective response, and such responders displayed the best median survival of approximately 41 months (Fig. 2b).

Limitations of this retrospective analysis include response evaluation by WHO criteria, not RECIST.2829 The TAX327 trial employed WHO criteria (see methods) for the objective evaluation of responses and was conducted before the widespread adoption of RECIST criteria. The date of first objective response was not captured; hence, a time bias is still possible, even though extensive landmark analysis was performed and no time bias was detectable. Only grouped response was measured (ie, CR/PR/SD/PD) and not percentage of tumor shrinkage—hence some details may be lost. However, the measurement of percentage declines in dimensions (employed extensively in the setting of antiangiogenic agents for renal cell carcinoma) is still not widely adopted, especially in the setting of cytotoxic chemotherapy. It is unclear whether examining percentage declines in tumor size will better discriminate between outcomes in the setting of chemotherapy for metastatic CRPC, although this may be an appropriate issue to evaluate in a more modern trial where these parameters may be available and measurable disease more prevalent using modern CT imaging. The category of SD is defined according to older WHO criteria and consists of a heterogeneous group of patients because the absolute percentage increase or decrease in size and the duration of SD are not captured in this database. Although we have examined categories of objective response or progression, we were unable to evaluate the surrogate association of time to objective progression or radiographic progression-free survival with OS given the high degree of censoring in this trial postprogression and the definition of progression in TAX327, which included PSA measures. At the time of conduct of the TAX327 trial, no effective second-line agents were available, which probably enhanced our ability to detect an association between objective response and OS. In the current era, the availability of effective second-line chemotherapy (cabazitaxel) and immunotherapy (sipuleucel-T), which yields delayed survival benefits, may confound the association between response to a front-line cytotoxic agent and survival. A limited number of all men (412 of 1006, 41%) enrolled on the TAX327 trial had measurable disease and were eligible for this analysis, and 160 (38.8%) of these men were not evaluable because of removal or censorship from trial before objective evaluation or confirmatory radiologic evaluation. The censorship in this trial and in many other phase 3 studies as well as the lack of measurable disease at baseline in many men with CRPC reduce the performance and generalizability of radiographic response as an effective broad surrogate of OS. Furthermore, the 594 patients who were inevaluable at baseline for objective response had longer median OS (19.2 months vs 15.3 months; Table 1), suggesting that patients with measurable tumors may have biologically more aggressive disease. Thus, as a stand-alone endpoint, radiographic response cannot be recommended as a primary endpoint in phase 2 or 3 trials in CRPC. Finally, the total number of patients with an objective response was small (37 of 412 evaluable, or 9.0%, and only 3.7% of all 1006 patients entered on the TAX327 trial). The assessment of surrogacy is limited by the small number of patients (n = 168) evaluable for all 3 categories of response (pain, PSA, and objective), and because percentage tumor shrinkage was not captured, and WHO response criteria were used and not RECIST. However, its association with OS outcomes suggests the utility of radiologic response along with other criteria in the selection of cytotoxic agents.

To conclude, in men with metastatic CRPC and measurable disease receiving chemotherapy, objective tumor response was prognostic for OS and appears to complement PSA and pain responses in the assessment of patient benefit over time. The validation of our findings in a separate dataset in the frontline as well as second-line setting is warranted, particularly in a large sample of patients with measurable disease, given that measurable tumors are being detected more commonly with current radiologic technology and criteria. With validation, objective responses may help expedite drug development by assisting in identifying more promising agents in screening phase 2 trials.


This study was supported, in part, by Sanofi-Aventis through provision of trial data. Guru Sonpavde, MD, is on the speakers bureau and receives honoraria from Sanofi-Aventis. William R. Berry, MD, is on the speakers bureau for Sanofi-Aventis. Ronald de Wit, MD, is a consultant to, on the advisory board of, on the speakers bureau of, and receives honoraria and research support from Sanofi-Aventis. Mario A. Eisenberger, MD, is a consultant to, on the advisory board of, and receives research support from Sanofi-Aventis. Ian F. Tannock, MD, PhD, receives research funding from Sanofi-Aventis. Andrew J. Armstrong, MD, is on the speakers bureau of and receives research funding from Sanofi-Aventis.