The hazards of intermediate endpoints


  • Tomasz M. Beer MD,

    Corresponding author
    1. Department of Medicine, Oregon Health and Science University Cancer Institute, Portland, Oregon
    • Department of Medicine, Oregon Health and Science University, 3303 SW Bond, CH14R, Portland, OR 97239; Fax: (503) 494-6197
    Search for more papers by this author
    • Fax: (503) 494-6197

  • Christopher W. Ryan MD

    1. Department of Medicine, Oregon Health and Science University Cancer Institute, Portland, Oregon
    Search for more papers by this author

  • See referenced original article on pages 000–000, this issue.


Progression-free survival in men with advanced prostate cancer is a composite endpoint that has been neither standardized nor validated as a measure of clinical benefit in advanced prostate cancer. The absence of a progression-free survival benefit in this randomized trial of atrasentan leaves the authors uncertain with regard to the potential value of this agent in prostate cancer.

Clinical benefit of treatment for advanced prostate cancer is not hard to define. Extension of survival, improved quality of life, and reduced disease complications are concrete examples over which there is little controversy. Clinical trials that rely on alternate endpoints risk producing confusing, if not uninterpretable, results.

The article by Carducci et al. in this issue of Cancer presents such a dilemma.1 Atrasentan, an agent thought promising for prostate cancer based on several prior studies, including randomized Phase 2 data,2 was investigated in a placebo-controlled, Phase 3 trial in metastatic, androgen-independent patients using progression-free survival as the primary endpoint. At first glance, this concept would seem clear enough. However, progression in advanced prostate cancer is not a straight-forward endpoint. It requires a composite construct of clinical and radiographic measures and has not been validated. Consequently, we are left with uncertainty when interpreting the results of this clinical trial.

The authors should be congratulated for successfully conducting such a large study without reliance on prostate-specific antigen (PSA) as a component of the progression endpoint. This is quite an accomplishment in a PSA-driven era, in which both patients and clinicians feel compelled to make and break treatment decisions based on fluctuations in this marker. The temptation of choosing PSA decline as an endpoint in prostate cancer clinical trial design remains strong, but the meaning of such results remains uncertain. Reduction in serum PSA in response to chemotherapy for androgen-independent disease, although it is correlated with survival,3 is not a direct or universally accepted measure of patient benefit. Furthermore, it is not clear that biologic agents that may benefit patients by stabilizing their disease without classic responses can be evaluated using PSA metrics. Despite the decision to exclude PSA from the progression endpoint of their study, its influence remains: The rigorous bone imaging schedule that may have been the study's undoing was put in place as an effort to override PSA-influenced progression decisions.

In theory, progression-free survival could prove a versatile metric applicable to both cytotoxic chemotherapy and cytostatic agents. However, as evidenced by this trial, progression-free survival is difficult to define and doubly hard to measure in advanced prostate cancer. Progression in this disease has many potential manifestations: a modest increase in the size of asymptomatic—but radiographically detectable—pelvic lymph nodes; the development of asymptomatic lesions on a radionuclide scan of the skeleton; the appearance of important clinical consequences, such as ureteral obstruction or painful bone lesions; or the relatively mundane rise in serum PSA. Such events are not created equal. Although delaying progression that is associated with symptoms surely is beneficial, delaying progression of which patients would be unaware had it not been for an arbitrary imaging study is another matter. In the absence of evidence tightly linking such events to subsequent cancer symptoms or death, delaying such events loses all practical importance. The results reported by the atrasentan investigators are illuminating in this regard: Eighty-seven percent of radiographic progression events were not associated with any other clinical evidence of progression. Were these events clinically meaningful?

Remarkably little is known concerning the correlation between progression-free survival and measures of clinical benefit. In an analysis of 98 patients with androgen-independent, metastatic prostate cancer treated with microtubule-targeting chemotherapy, Scher et al. examined the association between radiographic (as well as PSA) progression-free survival and overall survival.4 Overall, the concordance was moderate for both progression measures. Expressed as the Kendall τ statistic, concordance was 0.4 for radiographic progression-free survival and 0.33 for PSA progression-free survival. It is noteworthy that little association was observed between radiographic progression events and subsequent death when the progression event occurred during the first 4 months of treatment. Why would radiographic changes early in the course of therapy be less informative than similar changes noted later? One possible explanation could be that new radiographic abnormalities early in the course of treatment may represent disease progression; however, they also may reflect changes that already were underway before therapy was started or before the therapy had its full impact on the disease process. Another possibility is that they represent a flare phenomenon. Such phenomena have been well described with initial hormone therapy5 and may occur with chemotherapy.6

In the study by Carducci et al.,1 the a priori use of the chosen intermediate endpoint was either the undoing of an active agent or the validation of an inactive one. Which it was, we may never know. Several aspects of the design of their study make it impossible to determine whether the events classified as progression were clinically significant. Patients were not followed for any additional evidence of progression after asymptomatic radiographic progression was noted. Early cross-over of placebo-treated patients to open-label treatment further compromised the ability to detect more long-term benefits, such as delay in symptom development or an impact on overall survival.

It is worth noting that, in some other cancers, radiographically determined progression-free survival has been embraced unquestioningly despite the lack of a known correlation with firm endpoints, such as overall survival. In a recent example, U.S. Food and Drug Administration approvals for several kinase inhibitors in renal carcinoma were granted based on Response Evaluation Criteria in Solid Tumor (RECIST) response rates or progression-free survival advantage. Lost in these assumptions of clinical relevancy is the historic footnote that the radiographic changes we use unquestioningly to define solid tumor “response” or “progression” are descended from what differences a small group of oncologists palpably could discern among spheres imbedded in a foam mattress.7 Moertel and Hanley reflected on this exercise in their seminal 1976 article, in which they noted that most experimental therapeutic trials for solid tumors culminate when a clinician places a ruler or caliper over a lump and tries to estimate its size. Thirty years later, calipers have been replaced by computed tomography, skeletal scintigraphy, quality-of-life instruments, and enzyme-linked immunoadsorbent assays, but the challenges we face remain the same: Let us measure our endpoint accurately, and let these measurements be useful.

So, is progression-free survival dead in prostate cancer? We do not think so, but it is in dire need of refinement and subsequent validation. Methods of assessment of progression, including types and frequency of imaging studies and their interpretation, must be refined, and the relation between radiographically detected, asymptomatic events and clinically meaningful outcomes must be examined. The revised guidelines developed by the PSA Working Group and initially presented at the 2007 Multidisciplinary Prostate Cancer Symposium are a down payment on this important work.8 These guidelines inject new rigor into the diagnosis of progression of bone metastases that are imaged using bone scintigraphy. Two unequivocally new lesions are required to be present; then, additional new lesions must be observed on a subsequent scan before disease progression is declared. This approach, which was developed through expert consensus, is logical; and, as it is adopted, it also should be validated by demonstrating an association between this refined definition of progression on bone imaging and unambiguous measures of clinical benefit. Further refinement of the progression-free survival endpoint will depend on better understanding of the relation between each of the elements of this composite construct and the ultimate development of clinically meaningful outcomes. Progress toward this goal may be possible when final results from the Satraplatin and Prednisone Against Refractory Cancer (SPARC) trial are reported. The initial results of that trial, which compares satraplatin plus prednisone with prednisone alone, recently were reported,9 and assessment of both progression-free and overall survival is expected when the final analysis is completed.

We are looking forward to a more refined, robust progression-free survival endpoint. We look forward to its use in Phase 2 studies that are designed to screen new drugs or combinations for promising activity. Would we recommend its use in pivotal Phase 3 trials? No. Not now, and not any time soon. Not until we are convinced that progression-free survival is an unambiguous surrogate for clinical benefit. We are not there today, and we may never get there. The job of Phase 3 studies is to tell us whether a new treatment helps patients more than the best available treatments of today. To accomplish this, Phase 3 trials should be designed to measure patient benefit. Today, that means overall survival, quality of life, or delay in skeletal or other important clinical complications of prostate cancer.