Ability of C-reactive protein to complement multiple prognostic classifiers in men with metastatic castration resistant prostate cancer receiving docetaxel-based chemotherapy


  • G.R. Pond and A. Armstrong contributed equally to this article.

Guru Sonpavde, 501 Medical Center Boulevard, Webster, TX 77598, USA. e-mail: guru.sonpavde@usoncology.com


Study Type – Retrospective analysis of clinical trial

Level of Evidence 3

What's known on the subject? and What does the study add?

Serum C-reactive protein (C-reactive protein) is emerging as a potential novel prognostic factor in metastatic castration-resistant prostate cancer (mCRPC).

In the present study, a prospective trial was investigated retrospectively and a significant prognostic impact for C-reactive protein that was independent of multiple published prognostic models was identified in men receiving docetaxel-based chemotherapy for mCRPC. Prospective validation is warranted.


  • • Given the recent emergence of C-reactive protein levels as a novel prognostic factor in men with metastatic castration-resistant prostate cancer (mCRPC), we sought to evaluate the independent prognostic ability of C-reactive protein in the context of published prognostic nomograms, risk grouping and disease state models in men receiving docetaxel-based chemotherapy for mCRPC.


  • • A large randomized phase II trial (CS-205) of mCRPC patients who received docetaxel-prednisone + AT-101 (Bcl-2 inhibitor) or docetaxel-prednisone + placebo was analyzed retrospectively (n= 220).
  • • Overall survival (OS), progression-free survival (PFS) and measures of discriminatory ability were assessed in a hypothesis-generating analysis using Cox regression and concordance probabilities.
  • • Patients from both treatment groups were combined for this analysis because no significant differences in outcomes were observed.
  • • Because some factors used in nomograms were not collected or defined differently, risk was estimated based on slightly modified versions of nomograms.


  • • C-reactive protein was independently prognostic for OS and PFS (P≤ 0.002) after adjusting for all modeled risk estimates and classifiers.
  • • C-reactive protein showed a concordance probability of 0.65 for both OS and PFS.
  • • A 10-factor modified prognostic model based on the TAX327 trial had the greatest observed discrimination ability for OS and PFS (concordance probability = 0.623 and 0.603, respectively) among the modified nomograms or classifiers.
  • • Adding the TAX327 model risk estimates to C-reactive protein did not substantially increase discrimination ability over C-reactive protein alone.


  • • Current prognostic classifications provide modest discrimination of outcomes in mCRPC receiving docetaxel-based chemotherapy, highlighting the need for improved risk-based models.
  • • Baseline C-reactive protein appears to be an useful, independent prognostic factor and prospective external validation is warranted.



Eastern Cooperative Oncology Group


lactate dehydrogenase


metastatic castration-resistant prostate cancer


Prostate Cancer Working Group


progression-free survival


Docetaxel every 3 weeks plus prednisone (DP) yields a median overall survival (OS) of ≈19 months in men with metastatic castration-resistant prostate cancer (mCRPC) [1–3]. However, patients exhibit a range of outcomes that are associated with multiple readily available baseline clinical prognostic factors. These factors are components of multiple nomograms and include the presence of visceral or liver metastases, the number of metastatic sites, significant pain requiring narcotic analgesia, Karnofsky performance status, type of progression (i.e. measurable disease or bone scan progression), pretreatment PSA (Prostate Specific Antigen) levels and PSA doubling time, tumour grade, lactate dehydrogenase (LDH), albumin, alkaline phosphatase and haemoglobin (Table 1) [4–6]. Similarly, four risk factors derived from the TAX327 nomogram (clinically significant pain, presence of visceral metastatic disease, anaemia and bone scan progression at entry) were employed to construct three risk groups (good risk/intermediate risk/poor risk) based on the presence of 0–1, 2 or 3–4 factors. These risk groups were prognostic and showed moderate discriminatory ability for a ≥30% decline in PSA levels within 3 months and OS in patients with mCRPC starting docetaxel-based chemotherapy [7]. More simply, the Prostate Cancer Working Group (PCWG)-2 defined clinical subtypes of mCRPC according to patterns of metastatic disease (visceral, bone ± nodes, node-only), which may facilitate the interpretation of results across trials (Table 1) [8]. The discriminatory abilities of these prognostic models are all moderate, with the highest concordance index being 0.69 for the nomogram derived from the TAX327 trial [4]. Thus, novel prognostic factors including biomarkers are needed to provide additional discriminatory value for more accurately classifying the risk of death over time. This information is directly relevant to patients and providers, as well as for clinical research strategies for risk stratification.

Table 1. Variables in reported nomograms, risk grouping and Prostate Cancer Working Group (PCWG)-2 subtypes for metastatic castration-resistant prostate cancer
Prognostic classification factorArmstrong nomogram (TAX327)Armstrong risk groups (TAX327)Halabi nomogram (Cancer and Leukemia Group-B)Smaletz nomogram (Memorial Sloan Kettering Cancer Center)PCWG-2 clinical subtypes
Liver metastasisX    
≥2 Metastatic organ sitesX    
Visceral metastasis XX X
Performance statusX XX 
Bone progressionXX   
Grade (Gleason or WHO)X X  
PSA levelX X  
PSA doubling timeX    
Alkaline phosphataseX XX 
Albumin   X 
Lactate dehydrogense  XX 
Lymph node/soft tissue metastasis only    X
Bone metastasis ± soft tissue/node metastasis    X

In addition to PSA levels and serum alkaline phosphatase, C-reactive protein is a readily measurable and affordable plasma-based marker of inflammation, which has shown early promise as a potential biomarker of prognosis. In a retrospective study of 160 patients from the ASCENT (Androgen-Independent Prostate Cancer Study of Calcitriol Enhancing Taxotere) trial evaluating docetaxel-based chemotherapy with or without calcitriol, higher plasma C-reactive protein levels appeared to be a predictor of poor survival and a poor PSA response to docetaxel-based therapy [9]. This finding was also shown in another independent dataset of 119 mCRPC patients (of whom 57 received docetaxel) enrolled in six phase II clinical trials [10]. These initial findings are limited, however, by the relatively small number of prognostic factors that were considered and adjusted for in the multivariate analysis.

Recently, a large randomized phase II trial (CS-205) enrolled men with mCRPC and compared DP combined with placebo or AT-101, a Bcl-2 family antagonist [11]. Although this trial did not show an improvement in outcomes by the addition of AT-101, the combined dataset including both arms of the trial provides an excellent resource for retrospective hypothesis-generating analyses in the setting of DP-based therapy. This trial also performed baseline C-reactive protein measurements in a subset of patients. Therefore, a retrospective analysis of the CS-205 trial was conducted aiming to evaluate and compare the prognostic abilities of the aforementioned prognostic classifiers (i.e. the three reported nomograms, risk grouping and PCWG-2 subtypes) and to investigate the ability of C-reactive protein to enhance their prognostic abilities.


The CS-205 phase II trial was approved by local institutional review boards and conducted at 41 centres in Russia and the USA [11]. The stratification factors were pain and Eastern Cooperative Oncology Group (ECOG) performance status (0–1 vs 2). A single patient did not receive any treatment as a result of disease progression and was excluded from all analyses. The remaining 220 men received a maximum of 17 cycles of DP treatment, unless unacceptable toxicity, progression by PCWG-2 criteria (symptomatic, Response Evaluation Criteria in Solid Tumors, bone scan but not PSA progression alone) or death occurred [8]. AT-101 was not continued after discontinuation of DP. Imaging was obtained every three cycles or at symptomatic progression. Men in both arms of the CS-205 trial were combined for analysis because no significant differences in outcomes were observed [11].

The present analysis aimed to retrospectively evaluate and compare the prognostic abilities of previously published risk assessment classification systems (Table 1) in the CS-205 database and the ability of C-reactive protein to enhance their prognostic abilities. Tools to be evaluated were published nomograms by Halabi, Smaletz and Armstrong; the TAX-327 risk factor grouping (good risk/intermediate risk/poor risk, as described previously), and the PCWG-2 clinical subtype classification (Table 1) [4–8]. Descriptive statistics were used to summarize patient baseline characteristics. OS and progression-free survival (PFS) estimates were calculated using the Kaplan–Meier method. Cox proportional hazards regression was used to test for factors prognostic for outcome. Risk groups and nomogram estimates were defined as a continuous variable, whereas PCWG-2 subtypes were defined as categorical factors.

All laboratory assessments in the CS-205 trial, including C-reactive protein, were performed in a central laboratory. C-reactive protein was measured on American patients only (n= 112) on cycle 1, day 1; however, two patients had screening values substituted as a result of missing day 1. Logarithmic transformations for C-reactive protein, PSA levels and alkaline phosphatase were performed to normalize the data where appropriate, and the results are presented on a logarithmic scale. Performance status was defined using the ECOG scale; where necessary, ECOG 0, 1 and 2 performance status was defined as being equivalent to a Karnofsky performance status of 100%, 80–90% and 60–70%, respectively. LDH (which was a factor in both the Halabi and Smaletz nomograms) was not collected as part of the CS-205 trial and was excluded from the estimates for these nomograms (Table 1). Models including estimated LDH values were simulated with similar results being obtained; thus, only the non-simulated results are presented for brevity. The Armstrong nomogram defined high tumour grade (Gleason score ≥8 or WHO grade 3–4) vs low tumour grade (Gleason score ≤7 or WHO grade 2–3). WHO tumour grade was not collected in the CS-205 trial and almost 20% of patients had unknown Gleason score at baseline. Thus, tumour grade was defined as high (Gleason score ≥8) vs low (Gleason score ≤7 or unknown). The robustness of combining the unknown Gleason score patients with low Gleason score patients was assessed by: (i) scoring patients with unknown Gleason score as having a Gleason score ≥8 and (ii) by excluding these patients, with similar results being obtained. Only results using the initial imputation are presented for brevity. All of the other variables were available for estimating outcomes with prognostic classification systems (Table 1). Given the missing data and revised definitions of some factors within each of the three nomogram models, only reduced nomogram estimates are available. The reduced nomograms are labelled as the Halabi, Smaletz and Armstrong nomograms for simplicity.

Concordance probabilities were calculated to estimate the level of discrimination of different factors for selected outcomes. For OS and PFS outcomes, the method of Gonen and Heller [12] was used because the endpoint is a time-to-event outcome and also because of the frequency of censored observations (using the CPE package in R; http://www.r-project.org). The concordance probability is interpreted as the probability of observing a longer survival for a patient in one group compared to a patient in the other group. P≤ 0.05 (two-sided) was considered statistically significant.


The primary results of the CS-205 trial have been reported previously [11]. Briefly, a total of 108 patient deaths was observed; 153 patients had progressed after a median (maximum) follow-up in survivors of 18.0 (28.8) months at the time of data analysis. The treatment arms were balanced and outcomes were similar, with median OS of 18.1 vs 17.8 months (hazard ratio, HR, 1.07; 95% CI, 0.72–1.55; P= 0.63) for the AT-101–DP and placebo–DP arms respectively. Secondary endpoints including median PFS (11.0 and 10.3 months), a decline in PSA levels ≥50% (54% and 46%), a decline in PSA levels ≥30% (66% and 54%) and measurable disease control rates (93% and 80%) were also similar. Baseline characteristics are shown in the Supporting information (Table S1). All 220 evaluable patients were classifiable by PCWG-2 subtype, with most (148; 67.3%) having bone ± nodal disease. Fifty-six (25.5%) had visceral disease, whereas 16 (7.3%) had nodal disease only. There were two patients who did not have their baseline haemoglobin measured and six patients did not have baseline pain assessed, resulting in 212 patients being evaluable for risk groups classification. In total, 93 (43.9%) patients were good risk, 82 (38.7%) patients were intermediate risk and 37 (17.5%) patients were poor risk (see Supporting information, Table S1). In addition to the eight patients who could not be assigned a risk group status, two patients had missing baseline PSA levels, leaving 210 patients with an estimate of outcome using the Armstrong nomogram. The two patients with missing baseline PSA levels and the two patients with missing baseline haemoglobin were also excluded from both the Halabi and Smaletz nomograms. There were 42 patients without available Gleason scores who were assigned a low grade status (0 points).

Univariately, risk groups and each nomogram calculation were statistically significantly prognostic for PFS and OS (P≤ 0.012 for each) (Table 2). Discrimination was modest regardless of the risk assessment method used. The Armstrong TAX327 nomogram estimator had the highest discrimination ability, with concordance probabilities of 0.623 for OS and 0.603 for PFS. When patients were separated by quartiles of the Armstrong nomogram risk estimates, the 1-year survival was 84.4% (95% CI, 71.3–91.9%), 81.3% (95% CI, 67.0–89.8%), 67.7% (95% CI, 52.2–79.1%) and 57.3% (95% CI, 41.9–70.1%). Discrimination ability for PCWG-2 subtypes, risk groups, and reduced Halabi and Smaletz nomograms was poor to moderate, with concordance probabilities of 0.527, 0.553, 0.595 and 0.590 for PFS and 0.530, 0.593, 0.617 and 0.616 for OS, respectively (see Supporting information, Table S2). These results are comparable to the discrimination ability of the two major stratification factors (baseline pain and ECOG performance status) employed in the trial (concordance probabilities of 0.590 for OS and 0.564 for PFS). Additionally, combining risk groups and stratification factors did not lead to a substantially better concordance probabilities, although pain, known to be prognostic factor, is a component of both of these sets of variables [13].

Table 2. Univariate analysis
VariableAll patientsPatients with C-reactive protein measured
n Overall survivalProgression-free survival n Overall survivalProgression-free survival
Hazards ratio (95% CI)Hazards ratio (95% CI)Hazards ratio (95% CI)Hazards ratio (95% CI)
  • *

    Lactate dehydrogenase was not collected and could not be included in either the Halabi or Smaletz nomograms.

  • Multivariate estimates were not given by Smaletz et al. [5]; hence, nomogram points are estimated values. Eastern Cooperative Oncology Group (ECOG) performance status = 0, 1, 2 was assigned Karnofsky performance status = 100, 80–90, 60–70 and extrapolation of points for Karnofsky = 100 occurred.

  • Tumour grade was not captured; hence, high vs low tumour grade was defined as Gleason 8–10 vs <7 or unknown. Karnofsky ≥80 vs <80 was defined as ECOG = 0–1 vs ECOG = 2.

  • For nomogram calculations where applicable, Gleason unknown patients were grouped with Gleason <8 patients.

  • NC, not calculated because no patient with liver disease had an observed death date; PCWG, Prostate Cancer Working Group.

PCWG-2 subtypes220  112  
 Visceral disease1.43 (0.62–3.26)1.18 (0.57–2.42)2.17 (0.63–7.51)1.41 (0.48–4.16)
 Bone ± nodal disease1.07 (0.49–2.32)0.91 (0.46–1.80)1.90 (0.58–6.15)1.24 (0.44–3.45)
 Nodal disease onlyReferenceReferenceReferenceReference
P value0.390.370.470.79
Risk groups2121.55 (1.21–1.99)1.34 (1.07–1.67)1061.42 (1.01–1.98)1.35 (0.99–1.85)
P value<0.0010.0120.0420.058
Armstrong nomogram (/10 points)2101.15 (1.09–1.22)1.13 (1.07–1.18)1061.12 (1.03–1.22)1.12 (1.04–1.21)
P value<0.001<0.0010.0060.003
Halabi nomogram (/10 points)*2161.21 (1.11–1.31)1.16 (1.08–1.25)1101.12 (1.00–1.24)1.19 (1.07–1.32)
P value<0.001<0.0010.0450.002
Smaletz nomogram (/10 points)*2160.87 (0.81–0.92)0.90 (0.85–0.95)1100.90 (0.84–0.97)0.91 (0.84–0.97)
P value<0.001<0.0010.0060.006
Baseline log(C-reactive protein)1121.44 (1.21–1.72)1.44 (1.23–1.68)
Age2201.02 (1.00–1.05)1.00 (0.98–1.02)1121.04 (1.01–1.08)1.01 (0.99–1.04)
ECOG performance status >02201.76 (1.17–2.65)1.37 (0.97–1.92)1121.39 (0.82–2.36)1.35 (0.84–2.17)
Log(baseline PSA level)2181.22 (1.09–1.37)1.13 (1.02–1.26)1121.21 (1.03–1.42)1.18 (1.01–1.38)
Log (baseline alkaline phosphatase)2201.43 (1.15–1.79)1.42 (1.17–1.74)1121.13 (0.81–1.58)1.53 (1.11–2.10)
Rising PSA progression2201.21 (0.73–2.01)0.96 (0.64–1.44)1121.36 (0.73–2.53)1.14 (0.68–1.92)
Measurable disease2201.38 (0.90–2.11)1.17 (0.81–1.70)1121.10 (0.63–1.93)0.98 (0.60–1.61)
Previous surgery2200.83 (0.56–1.22)0.75 (0.54–1.03)1120.74 (0.44–1.26)0.52 (0.33–0.84)
Previous radiotherapy2201.35 (0.92–1.99)1.24 (0.90–1.71)1122.03 (1.14–3.63)1.45 (0.89–2.37)
Disease stage2160.94 (0.76–1.16)0.93 (0.78–1.11)1080.87 (0.68–1.13)0.86 (0.68–1.09)
Gleason score (8–10 vs others)2200.81 (0.55–1.19)0.91 (0.66–1.26)1120.88 (0.52–1.49)1.25 (0.78–2.01)
Gleason score (8–10 vs 1–7)1770.74 (0.49–1.12)0.94 (0.66–1.33)1020.81 (0.47–1.40)1.23 (0.75–2.02)
Visceral disease2201.35 (0.88–2.06)1.29 (0.90–1.84)1121.22 (0.67–2.20)1.16 (0.69–1.98)
Liver disease2200.82 (0.34–2.02)1.10 (0.56–2.17)112NC0.30 (0.04–2.17)
Baseline pain2141.72 (1.17–2.53)1.55 (1.11–2.16)1081.61 (0.93–2.80)1.51 (0.91–2.50)
Bone scan progression2201.07 (0.73–1.56)0.81 (0.59–1.12)1121.33 (0.78–2.24)1.10 (0.69–1.75)
Baseline anaemia2182.23 (1.48–3.36)1.74 (1.24–2.43)1102.24 (1.26–3.99)1.96 (1.18–3.25)
Baseline haemoglobin2180.75 (0.66–0.84)0.80 (0.73–0.89)1100.79 (0.67–0.93)0.82 (0.71–0.95)
Baseline lactate dehydrogenase00
Baseline albumin2200.67 (0.50–0.90)0.93 (0.73–1.18)1120.20 (0.11–0.37)0.46 (0.26–0.79)
PSA doubling time (<55 days)2201.43 (0.97–2.10)1.09 (0.79–1.50)1121.55 (0.91–2.61)1.07 (0.66–1.72)
Number of metastatic sites (>2)2201.43 (0.96–2.14)1.51 (1.08–2.11)1121.81 (0.99–3.33)1.47 (0.84–2.57)

In total, 112 of the 116 (96.6%) patients recruited in the USA had C-reactive protein measured, with a median (range) level of 0.496 (0.028–25.2) ng/mL. There was no statistical difference in C-reactive protein between the two arms of the trial (median of 0.69 ng/mL in the AT-101 group and 0.42 ng/mL in the placebo group), treatment group was not predictive for OS or PFS after adjusting for baseline C-reactive protein (P= 0.27 and 0.76) and no interaction effect between treatment group and C-reactive protein was observed (P= 0.19 and 0.31). When combining both arms of the trial, men with a higher C-reactive protein at baseline had significantly worse OS (HR, 1.44; 95% CI, 1.21–1.72 on a logarithmic scale; P < 0.001) and PFS (HR, 1.44; 95% CI, 1.23–1.68; P < 0.001). A log unit increase is equivalent to a change from 0.37 to 1 ng/mL, from 1 to 2.72 ng/mL or from 2.72 to 7.39 ng/mL. The 56 men who had C-reactive protein less than the median had a median OS of 24.5 months (95% CI, 21.4 to not reached) compared to 15.4 months (95% CI, 12.2 to 17.9 months) among 56 men with C-reactive protein equal to or more than the median (P < 0.001) (Fig. 1A); a prognostic difference appeared within the good-risk group (n= 47; HR, 3.56; P= 0.012) and the intermediate risk group (n= 39; HR, 2.42; P= 0.044), with a similar, statistically non-significant trend in the poor risk group (n= 20; HR, 2.80; P= 0.18) (Fig. 1B). Only seven patients had an abnormal C-reactive protein (≥8 ng/mL) and they had a median OS of 5.5 months (95% CI, 4.2–13.7), which was significantly worse than men who had normal C-reactive protein (HR, 4.20; 95% CI, 1.66–10.63; P= 0.002) (Fig. 1C).

Figure 1.

Survival based on baseline median C-reactive protein (CRP) for all evaluable patients (A), by risk group (B) and for normal or higher CRP levels (C).

Patients who had C-reactive protein measured had a lower probablility of having ≥2 metastatic sites, significant pain, ECOG 1 or 2 performance status, previous PSA progression, disease stage 3 or 4 and PSA doubling time <55 days, and a higher probablility of having previous radiotherapy and objective tumour progression. Of the patients with C-reactive protein measured (see Supporting information, Table S1), those with C-reactive protein above the median had a higher probablility of poor performance status (ECOG 1–2), pain, bone scan progression, no previous radical prostatectomy and high Gleason sum (8–10), and also higher median PSA and alkaline phosphatase levels and lower haemoglobin and albumin levels. Men with C-reactive protein levels above the median had a higher probablility of belonging to the poor risk classification, although they had a lower probablility of visceral/liver metastases and a higher probablility of node-only metastatic disease. Despite the fact that these patients were different from the population as a whole, the prognostic ability of the risk grouping, the PCWG-2 classifier and nomograms on OS and PFS, as measured by the HRs, were generally similar within this smaller cohort (Table 2).

As a continuous value on the logarithmic scale, C-reactive protein remained statistically significantly prognostic for OS after adjusting for the estimates using the Armstrong (P= 0.002), Halabi (P < 0.001) or Smaletz (P < 0.001) nomograms, risk groups (P < 0.001) and PCWG-2 subtypes (P < 0.001) (Table 3). Similarly, C-reactive protein remained statistically prognostic for PFS (P≤ 0.001 for all models) after adjustment. Alternatively, the nomogram estimators and the risk group or PCWG-2 classification were not statistically significant after adjusting for log(C-reactive protein). A sensitivity analysis was performed excluding all patients with abnormally high C-reactive protein levels (≥8 ng/mL) with similar results being obtained (data not shown). Discrimination ability (concordance probability) using log(C-reactive protein) as an univariate prognostic factor was 0.65 for both OS and PFS. Adding the nomogram estimates to log(C-reactive protein) did not substantially increase the discrimination ability over log(C-reactive protein) alone.

Table 3. Bivariate analysis
Variable n Overall survivalProgression-free survival
Hazards ratio (95% CI) P Hazards ratio (95% CI) P
  1. PCWG, Prostate Cancer Working Group.

PCWG-2 subtypes     
 Visceral disease1122.62 (0.76–9.07)0.151.83 (0.62–5.41)0.27
 Bone ± nodal disease2.39 (0.74–7.79)0.131.67 (0.60–4.67)0.33
 Nodal disease onlyReferenceReference
 Log(C-reactive protein)1.46 (1.23–1.73)<0.0011.45 (1.24–1.69)<0.001
Risk groups1061.20 (0.85–1.69)0.311.07 (0.77–1.49)0.69
Log(C-reactive protein)1.39 (1.16–1.68)<0.0011.41 (1.19–1.66)<0.001
Armstrong nomogram (/10 points)1061.06 (0.97–1.16)0.181.05 (0.96–1.15)0.26
Log(C-reactive protein)1.37 (1.13–1.66)0.0021.36 (1.14–1.62)<0.001
Halabi nomogram (/10 points)1101.02 (0.90–1.15)0.811.08 (0.95–1.22)0.25
Log(C-reactive protein)1.42 (1.17–1.73)<0.0011.35 (1.14–1.61)<0.001
Smaletz nomogram (/10 points)1100.97 (0.89–1.06)0.461.01 (0.92–1.11)0.88
Log(C-reactive protein)1.38 (1.12–1.70)0.0031.44 (1.17–1.77)<0.001


In the present retrospective study, we analyzed the prognostic ability of C-reactive protein levels in the context of established multivariate models incorporating most of the known prognostic factors in men with mCRPC based on published nomograms (Armstrong, Halabi and Smaletz models), as well as other more simple CRPC classifications, including the PCWG-2 CRPC subtypes and TAX327 risk groups. In addition to substantiating a modest prognostic ability for each model and classification system, we found that C-reactive protein levels provided independent prognostic significance, with a discriminatory ability by itself that rivalled the entire nomogram-based estimates. C-reactive protein, when staffed by the median value in this dataset (0.5 ng/mL), showed a favourable concordance probability of 0.65 for OS and PFS and 0.64 for a decline in PSA levels ≥30% by week 12. Indeed, adding the best prognostic classifier (i.e. the Armstrong nomogram estimates) to log(C-reactive protein) did not substantially increase the discrimination ability over log(C-reactive protein) alone. To confirm the integrity of our findings, we performed a sub-analysis excluding all seven patients with abnormally high C-reactive protein levels (≥8 ng/mL) possibly as a result of ongoing non-CRPC-related inflammation (e.g. infections), which also yielded similar results.

Although the present evaluation of the Halabi and Smaletz nomograms estimates, which were constructed in the pre-docetaxel era, could not incorporate LDH (because this was not collected in the CS-205 trial) and therefore comprised reduced nomogram estimates, C-reactive protein levels provided a stronger discriminatory ability than these reduced nomogram estimates. A recent retrospective analysis by Scher et al. [14] did identify LDH as a significant independent prognostic biomarker, although the trial evaluated a different agent and was conducted in a different setting (i.e. abiraterone acetate as post-docetaxel therapy). Notably, risk factor groupings and PCWG-2 subtypes were not initially constructed as prognostic groupings for OS. Risk groups were constructed to predict a decline in PSA levels ≥30% and PCWG-2 subtypes were suggested to enhance the clinical trial design (i.e. enroll clinically homogeneous patients) and enable a comparison across trials. Given that nomograms require computation and are difficult to utilize at the bedside for making clinical decisions, and also that these computation difficulties are compounded by missing information or changes in definitions over time, a simpler biomarker driven prognostic model using a simple, cheap and easily obtained blood test represents an unmet medical need.

The analysis in the present study is limited by the relatively modest numbers of patients evaluable for PCWG-2 subtypes (n= 220), risk groups (n= 212), nomogram estimates (n= 210–216) and C-reactive protein (n= 112). Additionally, there were some differences in baseline characteristics between patients with and without C-reactive protein measurements. Despite these caveats, C-reactive protein consistently emerged as having the greatest discriminatory ability among the factors evaluated, which is similar to that reported in previous studies [9,10]. Similar to circulating tumour cells, changes in C-reactive protein may be worthy of evaluation when aiming to examine its value as an early surrogate for long-term outcomes. Unfortunately, post-baseline C-reactive protein measurements were not captured as part of the CS-205 trial.

Another caveat is that C-reactive protein is not specific for prostate cancer and may be considered as an acute-phase reactant, albeit a sensitive marker of tissue damage and inflammation [15]. Pro-inflammatory cytokines (e.g. interleukin-6, and TNF-α) are released from the tumour microenvironment, inducing C-reactive protein synthesis from the liver, which may eventually modify the host immune response [16]. Preclinical results suggest that C-reactive protein enhances the effective uptake and presentation of bacterial antigens through FcyR on dendritic cells and stimulates protective adaptive immunity [17]. Moreover, chronic inflammation has an established role in carcinogenesis, including prostate cancer [18,19]. Baseline C-reactive protein was shown to be the only plasma based biomarker associated with poor survival in mCRPC patients receiving docetaxel-based chemotherapy when examining a panel of 16 cytokines, suggesting that C-reactive protein may best capture the overall inflammatory state [9]. Similar to the findings of the present study, when C-reactive protein was entered into a multivariate model including clinical and laboratory factors, only higher C-reactive protein levels remained a significant predictor (HR = 1.41) of poorer OS [9]. Higher C-reactive protein levels have also been associated with poorer survival in multiple solid tumours in addition to mCRPC, including melanoma, colorectal cancer, non-Hodgkin's lymphoma, oesophageal carcinoma, cervical cancer, endometrial cancer, ovarian cancer and Renal Cell Carcinoma [20–27]. C-reactive protein may also promote the proliferation of myeloma tumour cells and inhibit apoptosis [28]. Interestingly, lipid-lowering statins appear to reduce major cardiovascular events in patients with elevations of C-reactive protein ≥2 ng/mL, and also lead to a lowering of C-reactive protein, possibly via an anti-inflammatory mechanism [29,30].

In conclusion, the present study supports the modest prognostic discrimination provided in mCRPC by multiple prognostic classifiers. Baseline C-reactive protein appears to be a potentially useful prognostic factor that moderately predicted OS and PFS. Moreover, C-reactive protein appeared to improve the prognostic capability of all the prognostic classifiers evaluated. Although our data are hypothesis-generating, they complement an emerging body of evidence indicating the prognostic impact of C-reactive protein in mCRPC [9,10]. Given the plasma half-life of C-reactive protein of ≈19 h, immediate sample freezing is not necessary, which could be advantageous compared to most other sophisticated biomarkers [31]. Hence, the prospective external validation of baseline and serial changes in C-reactive protein may be warranted in multiple therapeutic settings, including chemotherapy, immunotherapy and androgen inhibitors, and in the context of a larger cohort of men with a range of additional biomarkers such as circulating tumour cells [32–36]. Prognostic models that incorporate clinical and plasma and tumour tissue molecular profiles as well as host genomics may optimize discriminatory ability in men with mCRPC.


This study was partly presented as a Poster at the American Society of Clinical Oncology Genitourinary Cancer Symposium, FL, USA (February 2011).


Andrew J. Armstrong received research support from Sanofi-Aventis, speaking honoraria from Sanofi-Aventis; Brian A. Wood is employed by Ascenta Therapeutics; Lance Leopold is employed by Ascenta Therapeutics; Matthew D. Galsky received research support from Ascenta Therapeutics; Guru Sonpavde received research support from Ascenta Therapeutics and research support and speaking honoraria from Sanofi-Aventis.