|Home | About | Journals | Submit | Contact Us | Français|
To explore whether progression-free survival (PFS) or biochemical PFS can be used as a predictor of overall survival (OS) and to investigate the dependence between PFS and OS in men with castrate-resistant prostate cancer.
Data from nine Cancer and Leukemia Group B trials that enrolled 1,296 men from 1991 to 2004 were pooled. Men were eligible if they had prostate cancer that had progressed during androgen deprivation therapy and did not receive prior treatment with chemotherapy, immunotherapy, or other nonhormonal therapy. Landmark analyses of PFS at 3 and 6 months from randomization/registration were performed to minimize lead time bias. The proportional hazards model was used to assess the significance effect of PFS rate at 3 and at 6 months in predicting OS. In addition, biochemical progression using the definitions of Prostate-Specific Antigen Working Group (PSAW) Criteria PSAWG1 and PSAWG2 were analyzed as time-dependent covariates in predicting OS.
The median survival time among men who experienced progression at 3 months was 9.2 months (95% CI, 8.0 to 10.0 months) compared with 17.8 months in men who did not experience progression at 3 months (95% CI, 16.2 to 20.4 months; P < .0001). Compared with men who did not progress at 3 and at 6 months, the adjusted hazard ratios for death were 2.0 (95% CI, 1.7 to 2.4; P < .001) and 1.9 (95% CI, 1.6 to 2.4; P < .001) for men who experienced progression at 3 and 6 months, respectively. In addition, biochemical progression at 3 months predicted OS. The association between PFS and OS was 0.30 (95% confidence limits = 0.26, 0.32).
PFS at 3 and 6 months and biochemical progression at 3 months predict OS. These observations require prospective validation.
The United States Food and Drug Administration approved docetaxel for first-line chemotherapy for men with progressive castrate-resistant prostate cancer (CRPC) based on two trials that demonstrated an improvement in overall survival (OS).1,2 The only other United States Food and Drug Administration–approved agents for the treatment of men with CRPC are mitoxantrone and estramustine. This paucity of approved agents for a disease that claims the lives of nearly 29,000 American men each year is striking.3 There are many reasons for this relative lack of agents, but one clearly is the difficulty in designing clinical trials for this group of patients.
A major challenge in designing trials in men with CRPC revolves around defining appropriate end points for these trials. Although OS remains the regulatory gold standard, the relatively long median survival of approximately 20 months slows trial completion. Other challenges in designing trials besides using end points other than survival (ie, radiographic or biochemical response) include the fact that only 30% of the patients have measurable disease,4 bone progression is not an accurate and reliable measure of progression, and prostate-specific antigen (PSA) progression, a biochemical marker, is the most common type of progression. Whereas consensus definitions of progression exist,5,6 it is worth noting that these are exactly that: criteria derived as a consensus process that have not been prospectively validated. Although improvement in survival and in patient symptoms are sufficient to demonstrate tangible clinical benefit, the former end point requires dauntingly large numbers of patients and the latter end point faces difficulties in terms of data collection and analysis if there is missing information. Therefore, there exists a great unmet need for validated intermediate end points in clinical trials in men with CRPC.
In an effort to address this question, we undertook a pooled analysis of 1,296 men with metastatic CRPC who participated in multicenter trials conducted by the Cancer and Leukemia Group B (CALGB). In particular, we sought to1 evaluate the effect of progression-free survival (PFS) in predicting OS,2 explore whether biochemical PFS as defined by consensus criteria was in fact a predictor of OS, and3 investigate the dependence between PFS and OS and biochemical PFS and OS.
Data from 1,296 men with CRPC who were treated on nine CALGB multi-institutional clinical trials from 1991 to 2004 were pooled. Men were eligible if they had prostate cancer that had progressed during androgen deprivation therapy and had not received prior treatment with chemotherapy, immunotherapy, or other nonhormonal therapy. In addition, an Eastern Cooperative Oncology Group performance status of 0 to 2 and adequate hematologic, renal, and hepatic functions were required. Each participant signed an institutional review board–approved informed consent document in accordance with federal and institutional guidelines. Details regarding these trials have been published elsewhere.7–15
The primary end point was OS, which was defined as the time from randomization/study entry to date of death from any cause. We evaluated the secondary end points of PFS and biochemical progression in predicting OS. PFS was defined uniformly in all CALGB protocols as the time of randomization/study entry to any progression (bone progression defined as two or more new lesions on bone scan, PSA progression using the PSA Working Group consensus criteria of 1999 [PSAWG1],5 or objective progression in lung, liver, nodes, or soft tissue disease) or death, whichever occurred first. PFS rate at 3 and 6 months was defined as a binary variable: a patient experiencing any type of progression at or before 3 months (or 6 months) was considered to have experienced an event. Otherwise, the patient was censored. The protocols required radiologic assessments every 8 to 12 weeks and PSA assessments every 3 to 4 weeks, although the extent of adherence to these requirements is not known.
In addition, we considered two definitions of biochemical progression: the PSAWG15 and PSAWG2.6 For both criteria, biochemical failure depended on whether patients experienced PSA declines of at least 50% or more. If they did, using PSAWG1, the PSA post-treatment measurement should be increased by 50% from nadir and PSA ≥ 5 ng/mL. Otherwise, progression was defined based on a 25% increase from nadir/post-treatment. Conversely, using the PSAWG2, a patient was considered as experiencing biochemical failure if their PSA post-treatment determination increased by 50% and PSA measurement ≥ 2 ng/mL. Both biochemical progression criteria required a second PSA confirmation at least 1 week later.
Landmark analyses of PFS at 3 and 6 months from randomization/registration were performed to minimize lead time bias.16 We were interested in analyzing the PFS rates at 3 and 6 months for the following two reasons. First, the median time to progression in the CALGB database was approximately 3 months. Second, the two trials demonstrating a survival advantage with docetaxel-based chemotherapy reported that the median time to progression was approximately 6 months in men with CRPC. Ninety-five patients died before 3 months and were excluded from the landmark analysis. The sample of 1,201 men was randomly allocated into training and testing data sets, with a roughly 2:1 ratio: 781 patients (65%) and 420 patients (35%), respectively.
The Kaplan-Meier product-limit method was used to estimate the OS distribution by the PFS rate at 3 and at 6 months and by the biochemical progression at 3 months.17 The proportional hazards model was used to assess the significance effect of PFS rate at 3 and at 6 months in predicting OS.18 The estimates of the PFS rates at 3 and 6 months were applied to the testing data set, and the overall misclassification error rates were computed. In addition, biochemical progression (PSAWG1 and PSAWG2) were analyzed at 3 and 6 months and as time-dependent covariates in predicting OS in the proportional hazards model. In these models, PSA progressions were considered as binary variables. Known prognostic variables4,19 including age, race, performance status, Gleason score (was assigned by local pathologists and was not centrally reviewed), hemoglobin, testosterone, PSA, alkaline phosphatase, lactate dehydrogenase (LDH), presence of visceral disease, prior treatment with radiotherapy, and years since diagnosis were included in the multivariable proportional hazards models. LDH, years since diagnosis, alkaline phosphatase, and PSA were modeled using the restricted cubic spline function as they had skewed distributions. Because not all of the protocols considered included docetaxel, the only known agent to prolong survival in CRPC, subgroup analyses of men who had received docetaxel every 3 weeks were performed.
The associations between OS and PFS and between OS and time to PSA progression were investigated using a statistic that estimates Kendall's τ measure of association for bivariate time to event outcomes subject to censoring.20 The null sampling distribution of the test statistic was approximated using 2,000 permutation replicates. Along with the bootstrap bias and SE, an interval estimate (confidence limit [CL]) of the association parameter was constructed using adjusted bootstrap percentile CLs based on 2,000 permutation replicates.21 S-plus statistical software (version 8.0, Insightful Corp, Seattle, WA) and R were used for data analyses and all statistical tests were two-sided.
A total of 1,201 men enrolled onto CALGB clinical trials were included in this analysis. Their baseline clinical and laboratory characteristics are presented inTable 1. The median age at diagnosis was 71 years, and approximately 15% of the men were African American. Forty-four percent of men had a Gleason sum of 8 or higher, and 36% had measurable disease. Eighty-nine percent of men had bone metastases. The median LDH, PSA, and alkaline phosphatase levels were 221 ng/mL, 107 U/L, and 148 ng/dL, respectively.
With the exception of LDH, there were no statistical differences in baseline clinical and laboratory variables in the training data set compared with the testing data set. A total of 771 and 419 progression events were observed in the training and testing data sets, respectively. The median survival times were similar between the training and testing data sets and were equal to 13.3 months.
Progression by PSA is observed as the first progression event in 60% of the patients, bone progression in 18%, measurable disease progression in 7%, and death as the first event in the remaining 15%. The median survival time among men who experienced any PFS at 3 months was 9.2 months (95% CI, 8.0 to 10.0 months) compared with 17.8 months in those men who did not experience any PFS at 3 months (95% CI, 16.2 to 20.4 months; P < .0001). A similar pattern is observed in the testing data set. The median survival times were 8.9 months (95% CI, 7.6 to 10.5 months) and 17.9 months (95% CI, 16.0 to 20.6 months; P < .0001) in men who experienced and did not experience PFS at 3 months, respectively. The Kaplan-Meier product-limit survival curves by PFS at 3 months are presented in Figure 1.
In multivariable analysis, PFS at 3 months predicted OS (Table 2). Compared with men who did not experience disease progression at 3 months, the adjusted hazard ratio (HR) for death of men was 2.0 (95% CI, 1.7 to 2.4; P < .001). Other statistically significant factors of OS were age, performance status, race, body mass index, hemoglobin, LDH, PSA, alkaline phosphatase, Gleason score, prior radiotherapy, and years since diagnosis.
In a subgroup analysis of 232 men who received docetaxel every 3 weeks, PFS at 3 months also predicted OS. The median survival times were 7.6 months (95% CI, 6.3 to 9.5 months) and 20.4 months (95% CI, 17.4 to 22.2 months) for men who experienced and did not experience disease progression at 3 months, respectively (P < .0001). The adjusted HR for PFS at 3 months was 2.4 (95% CI, 1.3 to 4.2; P < .001).
The median survival time among men who experienced any type of progression at 6 months was 10.1 months (95% CI, 9.0 to 11.2 months) compared with 19.6 months (95% CI, 16.6 to 21.7 months; P < .0001). In the testing data set, the median survival times were 9.3 months (95% CI, 7.6 to 11.6 months) and 19.2 months (95% CI, 16.0 to 23.4 months; P < .0001; Appendix Fig A1, online only) in men who experienced and did not experience progression at 6 months, respectively.
PFS at 6 months also predicted OS. Compared with men who did not experience progression at 6 months, the adjusted HR for death was 1.9 (95% CI, 1.6 to 2.4; P < .001). The predicted PFS probability at 3 and 6 months was compared with the observed survival probability. The estimates of the PFS rates at 3 and 6 months were applied to the testing data set, and the misclassification error rates were estimated to be 0.27 and 0.25, respectively.
Furthermore, in men who had received docetaxel, the median survival times were 9.8 months (95% CI, 6.2 to 12.5 months) and 20.3 months (95% CI, 17.7 to 22.7; P < .0001) for men who did and did not experience progression at 6 months, respectively. The adjusted HR for OS for PFS at 6 months was 2.6 (95% CI, 1.7 to 4.0; P < .001).
The median number of PSA measurements was seven (interquartile range, three to 12). The median survival time among men who experienced biochemical failure progression using PSAWG1 at 3 months was 10.0 months (95% CI, 8.9 to 11.1 months) compared with 15.1 months (95% CI, 13.9 to 16.1 months; P < .0001;Fig 2) in men who did not experience biochemical failure progression at 3 months. Similar results were observed using PSAWG2 definition at 3 months.
In multivariable analysis, biochemical PFS predicted OS. Compared with men who did not experience biochemical progression at 3 months, the adjusted HR for death of men was 1.5 (95% CI, 1.3 to 1.7; P < .001). Furthermore, the HR was 1.44 for biochemical progression using the PSAWG1 definition as a time-dependent covariate (95% CI, 1.28 to 1.62; P < .001,Table 3). The HR was 1.43 (95% CI, 1.27 to 1.61; P < .001;Table 3) when using the PSAWG2 biochemical progression criteria as a time-dependent covariate.
Similar statistically significant results were observed in men who received docetaxel every 3 weeks. The HRs for death were 1.94 (95% CI, 1.38 to 2.72; P < .0001) and 1.95 (95% CI, 1.39 to 2.74; P < .0001) for biochemical progression using PSAWG1 and PSAWG2 definitions, respectively.
The estimated association parameter for OS and PFS was 0.30 (bootstrap SE = 0.0172, 95% CL = 0.26, 0.32). There is strong evidence suggesting that OS and PFS are statistically associated (P < .00001). In addition, the estimated association parameter for biochemical PFS and OS was 0.30 (bootstrap SE = 0.017, 95% CL = 0.27, 0.33; P < .00001). The results based on a complete-case analysis (uncensored cases) yielded similar results.
In this pooled analysis of 1,201 men with CRPC, we observed that PFS at 3 months predicts OS. The median survival time was statistically shorter among men who experienced any type of progression at 3 months (9.2 months) compared with men who did not experience progression (17.8 months; P < .0001). Furthermore, in multivariable analysis, the HR for death for men who experienced any type of progression was 2.0 compared with men who did not experience progression. Similar results were observed in men who experienced progression by 6 months compared with men who did not experience progression by 6 months. Importantly, these observations were validated in a testing data set.
In this data set, progression by PSA is observed as the first progression event in 60% of the patients, bone progression in 18%, measurable disease progression in 7%, and death as the first event in the remaining 15%. Although PSA progression is by no means the only type of progression in these patients, biochemical progression was experienced first by a majority of patients. Consequently, we further investigated whether biochemical progression at 3 months predicted OS. The median survival times among men who experienced or did not experience biochemical progression at 3 months were statistically different (median, 10.0v15.1 months, respectively).
Furthermore, using the PSAWG definitions of biochemical progression as time-dependent covariates, we have shown that biochemical progression is a statistically significant predicator of OS. The hazard for death for men who had experienced biochemical progression at any time during the trial represented a 50% increase compared with men who did not experience biochemical progression. It is gratifying that at least in this case, a consensus process yielded definitions of progression that could subsequently be validated.
CRPC is heterogeneous disease, and using PFS as a composite end point can be justified in study design to obtain higher event rates and therefore require smaller sample sizes. There are, however, many challenges to using PFS as an end point. The PFS end point requires careful review of its components. The inclusion of one component should be expected to be affected by the therapy, the component should be objective and should be measured accurately and reliably on all patients, it should be clearly defined and prespecified, and, finally, each component should be proven to be an important outcome and demonstrate clinical benefit to the patients.
PFS is usually defined as time to first biochemical, bone, or objective progression, or death. Although biochemical response (30% decline in PSA at 3 months) has been shown in retrospective analysis to be a surrogate of OS,22,23 prospective validation is still needed. In addition, it is more difficult to establish surrogacy with other components of PFS. There is little agreement on the definition of bone progression. Some investigators assume one new lesion,24,25 whereas others consider two lesions26,27 as an indication of disease progression. Furthermore, there have been reports of variability in reading radiographic bone scans.28 There are no studies that have demonstrated that bone progression predicts OS. Among the challenging factors is the lack of complete data on the individual components of the PFS end point once the patient has experienced progression. Consequently, it is important to collect all types of progression failures so that appropriate statistical methods for analysis can be implemented.
The lack of a standard definition for PFS end point in trials in men with CRPC has hampered the development and approval of novel drugs. Although there is a lack of consensus concerning the definition of progression, these CALGB studies have the advantage of using a standard definition of PFS over 13 years and across nine multicenter trials.
The relationship between PFS and OS and biochemical PFS and OS in men with CRPC remains unclear. However, in the majority of men with CRPC having elevated PSA at the time of study entry, CALGB investigators hypothesized that increasing PSA levels preceded and reflected growing cancer that would shortly be followed by symptomatic progression, progression in bone or soft tissue, or death. Assuming that the prostate cancer model is correct (Fig 3), CALGB investigators postulated that there is a strong dependence between PFS and OS end points.29 The results of this large retrospective analysis demonstrated strong evidence of dependence between PFS and OS and biochemical PFS and OS (P < .00001). These findings are similar to what was reported previously from a single institution.30 Our estimate, however, is more representative of the patient population (multi-institution) and is more precise (more events observed) than that of Scher et al.30
A measure of concordance rather than a linear correlation was used to investigate the association between OS and PFS. It is therefore important to avoid a numerical comparison between the two types of measures. It may be helpful to consider the case of a Gaussian copula with parameter of θ; a Kendall's τ of 0.3 corresponds to θ = 0.45 if the marginals are uniform on [0,1].
Although this retrospective analysis suggests that there is strong statistical association between PFS and OS and biochemical progression and OS, the clinical relevance of these associations has been questioned by regulatory bodies and agencies and needs to be prospectively validated.
While these data suggest that a composite PFS end point may serve as an intermediate end point for OS in phase II trials, this analysis has several limitations. First, it is not clear that the experience of men who enrolled on cooperative group trials is generalizable to the CRPC population, and these findings may not necessarily be applicable to noncytotoxic therapies. Nevertheless, there are some advantages to using these data in that uniform inclusion criteria, uniform monitoring, and uniform definitions of progression were used. A second limitation of the existing data set may be the lack of complete data that are available from each component of the PFS end point, making it difficult to draw inferences about the integrated composite end point. Finally, as noted, PSA progression alone clearly does not capture all progression events.
In summary, PFS and biochemical progression seem to be associated with OS. These data need to be validated prospectively before they can be used routinely as an intermediate end point in phase II trials in CRPC. Because there is no curative treatment for men with CRPC, it is vital to use standard definition of PFS and to standardize the frequency of assessment for all components of the PFS end point. A critical task remains, namely to develop and validate intermediate end points of OS to accelerate drug approval and to improve the survival of the 29,000 men who will die of prostate cancer in 2009.
We thank Chen Jiang for her help in programming in R.
Supported in part by grants from the United States Department of Defense Grants No. DAMD 17-03-1-0112 and W81XWH-06-1-0032 and the National Cancer Institute (Grant No. CA 36601).
Presented in part in abstract format at the American Society of Clinical Oncology Genitourinary Symposium, San Francisco, CA, February 14-16, 2008
Authors' disclosures of potential conflicts of interest and author contributions are found at the end of this article.
The author(s) indicated no potential conflicts of interest.
Conception and design: Susan Halabi, Nicholas J. Vogelzang, Eric J. Small
Provision of study materials or patients: Nicholas J. Vogelzang
Collection and assembly of data: Eric J. Small
Data analysis and interpretation: Susan Halabi, Nicholas J. Vogelzang, San-San Ou, Kouros Owzar, Laura Archer
Manuscript writing: Susan Halabi, Kouros Owzar, Eric J. Small
Final approval of manuscript: Susan Halabi, Nicholas J. Vogelzang, San-San Ou, Kouros Owzar, Laura Archer, Eric J. Small