Originally published as JCO Early Release 10.1200/JCO.2008.21.4726 on April 20 2009

This Article Full Text (PDF) Purchase Article View Shopping Cart Alert me when this article is cited Alert me if a correction is posted Services Email this article to a colleague Similar articles in this journal Similar articles in PubMed Alert me to new issues of the journal Save to my personal folders Download to citation manager Rights & Permissions Citing Articles Citing Articles via Google Scholar Google Scholar Articles by Pollock, B. H. Search for Related Content PubMed PubMed Citation Articles by Pollock, B. H. Related Articles Related Article Social Bookmarking                            What's this?

Cheaper, Faster, Better: Chasing the Elusive Clinical Trial End Point

Department of Epidemiology and Biostatistics, and Cancer Therapy and Research Center, University of Texas Health Science Center at San Antonio, San Antonio, TX

Cancer clinical trialists perennially seek to use end points in trials that are objective, easily measured and, if not the definitive end point of interest, are at least highly predictive of a definitive outcome. Researchers have been on a persistent quest to identify shorter-term, easier to measure, more frequent event-driven end points such as tumor response, biochemical failure, or other markers of disease progression as surrogate measures for definitive outcomes such as overall survival (OS). Evidence for this long-term ongoing pursuit can be traced back to the subject of an article that appeared in the first issue of Journal of Clinical Oncology (JCO) by Anderson et al¹ The authors illustrated the bias-ridden and then more common reporting of tumor response–specific survival. They presented the landmark method that corrects for biases by fixing the time point for the determination of tumor status and enumerating cumulative person-time and events from that point forward. An alternative was to use tumor status as a time-dependent covariate and to count person-time from the beginning of treatment, and then to switch patients from a nonresponder designation to responder designation at the time of tumor response. Simon and Makuch² proposed a method that combines the landmark and time-dependent covariate approaches. Without appropriate adjustment, any early event that is used to predict a definitive outcome will be biased. A recent review can be found by Anderson et al.³

For the design of a clinical trial, surrogate end points may be the only practical measures that can be used when definitive events occur infrequently. In oncology, these situations arise when the focus is on the evaluation of rare malignancies, small clinically defined subgroups, or when the end point of interest occurs at a low event rate.

In this issue of JCO, Halabi et al⁴ present an analysis of the relationship between progression-free survival and overall survival for men with castrate-resistant prostate cancer. Establishing a historical cohort of 1,296 men treated on nine multi-institutional Cancer and Leukemia Group B protocols during the years 1991 to 2004, the cohort included men who had progressed while being treated on androgen deprivation therapy alone but were otherwise not exposed to treatments such as chemotherapy, immunotherapy, or nonhormonal therapy. This cohort had a relatively long median survival of approximately 20 months. From a planning perspective, such a long median survival reduces the feasibility of using OS as a primary definitive clinical trial end point for this clinically defined subset of men with castrate-resistant prostate cancer. The authors evaluated the relationship between OS as the primary end point and secondary end points, including progression-free survival (PFS) and biochemical progression of prostate-specific antigen (PSA). Simple failure measures such as biochemical failure plus death, as well as more complex indicators of treatment failure (a combination of signs of progression such as radiographic evidence of bone, lung, liver, nodes, or soft tissue progression, or PSA progression) plus death, occur with greater frequency than death alone and therefore may be more suitable to reduce the time required to conduct a clinical trial. The authors assessed the relationship between OS and alternative measures of disease progression. Appropriate landmark analyses were conducted comparing OS across groups defined as having experienced progression at 3- and 6-month time points, or by groups defined by biochemical failure at 3- and 6-month time points. Covariate adjustment for known prognostic factors was performed using Cox proportional hazards regression. A subgroup analysis was performed for the men who received docetaxel on their treatment protocol with similar findings.

Halabi et al found that PFS was significantly associated with OS. The magnitude of the association between 3-month PFS and OS, measured by the proportional hazard ratio (PHR) after covariate adjustment was 2.0 (95% CI, 1.7 to 2.4). Thus, men who had evidence of disease progression were twice as likely to die. Using the 6-month PFS criterion, the PHR was 1.9 (95% CI, 1.6 to 2.4). The association between PFS and OS was 0.30 (Kendall's , a measure of association which ranges between 1.0 for perfect agreement, 0.0 for no agreement, to –1.0 for perfect disagreement).

There are some practical questions that arise in interpreting the findings. What is the clinical impact of demonstrating that patients who have experienced progression at 3 or 6 months have roughly double the mortality rate (PHR = 2.0)? Although statistically significant, early failure measured by PFS is a much less than perfect indicator of subsequent death. What does the association between PFS and OS of 0.30 (Kendall's ) mean? The magnitude of agreement between PFS and OS did not even approach a value of 1.0.

Scher et al⁵ examined the association between PFS and OS in a castrate-metastatic prostate cancer cohort enrolled on three consecutive trials of microtubule-targeted cytotoxic agents from a single institution. The major results of Scher et al are consistent with those reported by Halabi et al. For the association with OS, Scher et al reported Kendall's = 0.40 for radiographic imaging-determined PFS, and Kendall's = 0.33 for PSA-determined PFS.

There were several potential explanations for low Kendall's values. Evidence for progression is determined at fixed time points (eg, 3-month intervals for imaging or 1-month intervals for PSA) leading to interval censorship. This is in contrast to the assessment of OS, which is measured daily. The strength of association is also dependent on the need to observe both failure events (PFS and OS) in each case. Thus, a high proportion of censored observations can bias the determination of Kendall's . Scher et al concluded that the relatively low association between PFS and OS implicates other physiologic pathways that are not captured by PSA or radiographic imaging. Although the Halabi et al study was conducted in a much larger multi-institutional population, even with more precise estimates of association and a slightly more generalizable multi-institution study population, the overall conclusion is that the predictive utility of PFS for OS is limited regardless of the definition of progression that is used.

Clinical outcomes may be affected by disease processes that are not mediated through a surrogate measure and by mechanisms that are independent of the disease process.⁶ Ideally, surrogate end points should be definitive, precise, and not infrequent. Validation of their use is often difficult, especially for uncommon diseases or in populations expected to have a low event rate for the definitive outcome of interest. Often that means pooling patients across multiple studies during lengthy periods of time. The historical cohort analyzed by Halabi et al spanned 14 years, 1991 to 2004. In addition to changes in therapy (which can be measured and accounted for statistically), in that 14-year period it is likely that significant secular changes in diagnostic imaging technologies have taken place, as well as changes in supportive care practices that may affect outcome.

Studies that rely on the use of end points that are weakly associated with survival will be underpowered, having a greater potential to miss a truly beneficial therapy. Also, resources that are spent to conduct trials with little ability to affect survival will be wasted. It has been well established that assessing correlation alone does not suffice for an intermediate end point to replace, or serve as a surrogate end point for, the definitive outcome of interest in a clinical trial comparing treatments.^7–9 The measures proposed in this article must be proven to meet the stringent criteria required to substitute them for the true end point of relevance. Until then, they remain only as secondary observations in the assessment of treatment efficacy.

AUTHOR'S DISCLOSURES OF POTENTIAL CONFLICTS OF INTEREST

The author(s) indicated no potential conflicts of interest.

REFERENCES

1. Anderson JR, Cain KC, Gelber RD: Analysis of survival by tumor response. J Clin Oncol 1:710–719, 1983.[Abstract]

2. Simon R, Makuch RW: A non-parametric graphical representation of the relationship between survival and the occurrence of an event: Application to responder versus non-responder bias. Stat Med 3:35–44, 1984.[Medline]

3. Anderson JR, Cain KC, Gelber RD: Analysis of survival by tumor response and other comparisons of time-to-event by outcome variables. J Clin Oncol 26:3913–3915, 2008.[Free Full Text]

4. Halabi S, Vogelzang NJ, Ou SS, et al: Progression-free survival as a predictor of overall survival in men with castrate-resistant prostate cancer. J Clin Oncol 27:2766–2771, 2009.[Abstract/Free Full Text]

5. Scher HI, Warren M, Heller G: The association between measures of progression and survival in castrate-metastatic prostate cancer. Clin Cancer Res 13:1488–1492, 2007.[Abstract/Free Full Text]

6. Fleming TR, DeMets DL: Surrogate end points in clinical trials: Are we being misled? Ann Intern Med 125:605–613, 1996.[Abstract/Free Full Text]

7. Freedman LS, Graubard BI, Schatzkin A: Statistical validation of intermediate endpoints for chronic diseases. Stat Med 11:167–178, 1992.[Medline]

8. Lin DY, Fleming TR, De Gruttola V: Estimating the proportion of treatment effect explained by a surrogate marker. Stat Med 16:1515–1527, 1997.[CrossRef][Medline]

9. Prentice RL: Surrogate endpoints in clinical trials: Definition and operational criteria. Stat Med 8:431–440, 1989.[Medline]

CiteULike    Complore    Connotea    Del.icio.us    Digg    Facebook    Reddit    Technorati    Twitter    What's this?

Related Article

• Progression-Free Survival as a Predictor of Overall Survival in Men With Castrate-Resistant Prostate Cancer
  Susan Halabi, Nicholas J. Vogelzang, San-San Ou, Kouros Owzar, Laura Archer, and Eric J. Small
  JCO 2009 27: 2766-2771 [Abstract] [Full Text]

This Article Full Text (PDF) Purchase Article View Shopping Cart Alert me when this article is cited Alert me if a correction is posted Services Email this article to a colleague Similar articles in this journal Similar articles in PubMed Alert me to new issues of the journal Save to my personal folders Download to citation manager Rights & Permissions Citing Articles Citing Articles via Google Scholar Google Scholar Articles by Pollock, B. H. Search for Related Content PubMed PubMed Citation Articles by Pollock, B. H. Related Articles Related Article Social Bookmarking                            What's this?