E U R O P E A N J O U R N A L O F C A N C E R 4 2 ( 2 0 0 6 ) 1 3 4 4 – 1 3 5 0
`
`a v a i l a b l e a t w w w . s c i e n c e d i r e c t . c o m
`
`j o u r n a l h o m e p a g e : w w w . e j c o n l i n e . c o m
`
`Review
`
`Prostate-specific antigen (PSA) alone is not an appropriate
`surrogate marker of long-term therapeutic benefit in
`prostate cancer trials
`
`Laurence Collettea,*, Tomasz Burzykowskib, Fritz H. Schro¨ derc
`aEuropean Organisation for Research and Treatment of Cancer (EORTC) Data Center, Avenue E. Mounier 83/11, 1200 Brusells,
`Biostatistics Department, Brussels, Belgium
`bCenter for Statistics, Hasselt University, Diepenbeek, Belgium
`cDepartment of Urology, Erasmus MC, University Medical Center, Rotterdam, The Netherlands
`
`A R T I C L E I N F O
`
`A B S T R A C T
`
`Article history:
`Received 27 January 2006
`Received in revised form
`2 February 2006
`Accepted 6 February 2006
`Available online 30 May 2006
`
`Keywords:
`PSA
`Surrogate endpoint
`Clinical trials
`Prostate cancer
`Prognostic factor
`
`The prostate-specific antigen (PSA) is the most studied marker of prostate cancer. It is used
`for screening and as indicator of disease evolution for individual patients. PSA being a prog-
`nostic factor is however not sufficient to justify using PSA-derived endpoints as surrogate
`for definitive survival endpoint in phase III trials. First, we clarify the terminology and
`requirements for a marker to be a valid surrogate endpoint. We then review the published
`literature pertaining to the validation of PSA endpoints as surrogate in all disease stages.
`We discuss the limitations of these studies and conclude that so far, PSA is not a validated
`surrogate endpoint in any of the disease settings and treatment conditions considered. We
`give some recommendations for the planning of trials that would use PSA endpoints (in
`hormone refractory disease) and for the early stop of (endocrine treatment) trials on the
`basis of intermediate results based on PSA.
`
`Ó 2006 Elsevier Ltd. All rights reserved.
`
`1.
`
`Introduction
`
`Phase III cancer clinical trials that evaluate the clinical benefit
`of new treatment options often require large patient numbers
`and long follow-up. Recent advances in the understanding of
`the biological mechanisms of disease development have re-
`sulted in the emergence of a large number of potentially effec-
`tive new agents. There is also increasing public pressure for
`promising new drugs to receive marketing approval as rapidly
`as possible, in particular for life threatening diseases such as
`cancer. For these reasons, there is an urgent need to find ways
`of shortening the duration of cancer clinical trials. The dura-
`
`tion of phase III trials results from the use of long-term clinical
`endpoint (clinical progression, survival). Therefore, to replace
`this endpoint (the ‘‘true’’ endpoint) by another (a ‘‘surrogate’’
`endpoint), that could be measured earlier, more conveniently
`or more frequently, and that would adequately reflect the ben-
`efit of new treatments on the clinical endpoint(s), seems an
`attractive solution. In the field of prostate cancer, prostate-
`specific antigen (PSA) has probably been the most studied bio-
`marker.1 It has been investigated as a prognostic factor and as
`a potential surrogate endpoint across disease stages.
`It is a common misconception that established prog-
`nostic factors necessarily make valid surrogate endpoints. A
`
`* Corresponding author: Tel.: +32 2 774 16 69; fax: +32 2 771 38 10.
`E-mail address: laurence.collette@eortc.be (L. Collette).
`0959-8049/$ - see front matter Ó 2006 Elsevier Ltd. All rights reserved.
`doi:10.1016/j.ejca.2006.02.011
`
`MYLAN PHARMS. INC. EXHIBIT 1122 PAGE 1
`
`

`

`E U R O P E A N J O U R N A L O F C A N C E R 4 2 ( 2 0 0 6 ) 1 3 4 4 – 1 3 5 0
`
`1345
`
`prognostic factor is an intermediate outcome that is corre-
`lated with the true clinical outcome (T) for an individual pa-
`tient.2
`Its knowledge may be useful
`for diagnostic or
`prognostic assessment of an individual patient. For a prog-
`nostic factor to be a surrogate endpoint (S), it is further re-
`quired that ‘‘the effect of treatment on a surrogate endpoint
`must be ‘‘reasonably likely’’ to predict clinical benefit’’.2,3 In
`other words, a biomarker S will be a good surrogate for the
`true endpoint T if the results of a trial using outcome S can
`be used to infer the results of the trial if T had been observed
`and used as endpoint and this with sufficient precision. To
`demonstrate surrogacy, a high association between the treat-
`ment effects on the surrogate and on the true endpoint thus
`needs to be established across groups of patients treated with
`a new versus a standard intervention.
`Fig. 1 shows a schematic of two situations: a) where post-
`treatment PSA level (S) is prognostic for mortality risk (T) (as
`shown by the diagonal orientation of the ovals representing
`the individual patient data in two treatment groups), but is
`not surrogate, as is indicated by the line linking the two
`group’s averages being horizontal; and b) where post-treat-
`ment PSA level (S) is weakly prognostic for mortality risk (T)
`(as shown by the more horizontal orientation and more circu-
`lar shape of the ovals representing the individual patient
`data), but is a strong surrogate of the treatment difference
`on the mortality risk, as shown by the bar linking the group’s
`averages being diagonal, so that differences in average post-
`treatment PSA level between the treatment groups correlate
`with difference in average mortality risk.
`To illustrate this, let us consider the recently published
`secondary results of the Tax-327 study.4 This study compared
`a weekly and a three-weekly schedule of docetaxel plus pred-
`nisone to mitoxantrone and prednisone in hormone refrac-
`tory prostate cancer (HRPC). In this study, like often in this
`disease state, patients who achieved a PSA response had a
`60% reduction in mortality risk compared with non-responders
`(hazard ratio (HR) = 0.40, 95% CI: 0.31–0.51). The reduction of
`the PSA by 50% or more from baseline value, which was de-
`fined as a PSA response, was a strong prognostic factor for
`survival. Now considering PSA response as an endpoint and
`as a putative surrogate for overall survival, we observe with
`the authors that the weekly docetaxel arms resulted in a
`
`Median survival
`
`17.4 m
`
`16.5 m
`
`P=0.362
`
`Weekly Docetaxel
`
`Mitoxantrone + Prednisone
`
`P<0.0001
`
`0%
`
`32%
`
`48%
`
`PSA response rate
`
`Fig. 2 – A prognostic factor does not make a surrogate
`endpoint–the Tax 327 trial.
`
`response rate of 48% which was significantly different from
`the 32% response rate that was obtained with standard arm
`mitoxantrone plus prednisone (P < 0.0001). However, the med-
`ian overall survival on the weekly docetaxel arm amounted
`17.4 months and did not differ statistically significantly from
`the 16.5 months median survival achieved with the standard
`treatment (P = 0.362, Fig. 2). The benefit amounting less than a
`month was also not medically relevant, contrary to the differ-
`ence in response rates. Thus in this study, PSA response
`although it was a strong prognostic factor for survival at the
`patient level, did not appear to be reliable as a surrogate for
`survival when comparing the weekly docetaxel treatment to
`mitoxantrone plus prednisone.
`
`2.
`
`Statistical validation of surrogate endpoints
`
`Traditionally, the ‘‘Prentice Criteria’’5 were used for the pur-
`pose of demonstrating surrogacy on the basis of data from a
`single trial. The Prentice criteria require that four conditions
`be shown to be true in order to demonstrate the validity of
`a putative surrogate endpoint (here PSA), as a replacement
`endpoint for a true endpoint T (here survival):
`
`(a) There must be a statistically significant treatment effect
`on the PSA endpoint (in univariate analysis)
`(b) There must be a statistically significant treatment effect
`on survival (univariate analysis)
`
`Fig. 1 – Prognostic factor versus Surrogate endpoint. Schematic of two situations where (a) the surrogate S (PSA) is a good
`prognostic factor for the true endpoint T (mortality risk) in both treatment groups but is a not a surrogate for T and (b) the
`surrogate S (PSA) is only a weak prognostic indicator of the endpoint T (mortality) at the individual level and is a good
`surrogate endpoint for replacing the true endpoint T (mortality) in phase III clinical trials.
`
`MYLAN PHARMS. INC. EXHIBIT 1122 PAGE 2
`
`

`

`1346
`
`E U R O P E A N J O U R N A L O F C A N C E R 4 2 ( 2 0 0 6 ) 1 3 4 4 – 1 3 5 0
`
`(c) The PSA endpoint must be a statistically significant prog-
`nostic factor for survival (univariate analysis)
`(d) The treatment effect on survival must completely van-
`ish in a survival model with both the treatment and the
`PSA endpoint as explanatory variables (multivariate
`analysis).
`
`Although often used, the criteria are not a proper tool to
`check the validity of a surrogate. They do not aim at verifying
`the quality of prediction of clinical benefit. Condition (b) lim-
`its the applicability of the criteria to trials that showed a sta-
`tistically significant treatment effect on the true endpoint, a
`condition which is rarely fulfilled by clinical trials in prostate
`cancer. Condition (d) is impossible to verify in practice, as it
`amounts to ‘‘proving a null hypotheses’’, i.e., showing that
`the treatment effect is zero. Usually, it is checked by requiring
`that a statistical test shows the treatment effect to be statis-
`tically not significant in a model adjusted for the surrogate
`endpoint. Statistical tests however are designed to reject a
`null hypothesis and non-rejection of the null hypothesis
`never stands as definitive proof that the null hypothesis is
`true.6 In fact, one can obtain a non-significant test result sim-
`ply by having an inadequate sample size. Finally, it was
`shown that for time-to-event endpoints, the Prentice criteria
`are neither necessary nor sufficient to demonstrate that sur-
`rogacy holds true.7 Thus, failure to demonstrate that the four
`criteria hold does mean a biomarker should be disregarded as
`a surrogate endpoint, while successful demonstration that
`the four criteria hold true is not sufficient to actually demon-
`strate that a biomarker is a surrogate for a time-to-event
`endpoint.
`More recently, a new methodology known as the ‘‘meta-
`analytic validation’’ was developed.8,9 Using data from several
`trials, this method consists in deriving a model that can pre-
`dict the magnitude of the treatment effect on the true end-
`point,
`from the treatment difference observed on the
`surrogate (PSA) endpoint. A surrogate is valid if the prediction
`is sufficiently precise. This new methodology aims directly at
`verifying whether ‘‘the effect of treatment on a surrogate end-
`point is reasonably likely to predict clinical benefit’’. Further-
`more, it does not require that any of the treatment effects in
`
`Treatment effect
`observed in the
`trials
`
`R²trial indicates t he
`quality of the
`regression
`
`1
`
`.5
`
`0
`
`Treatment
`Effect
`on
`True
`Endpoint
`(β)
`
`-.5
`
`-1
`
`0
`1
`-1
`Treatment Effecton Surrogate Endpoint (α)
`
`Fig. 3 – Prediction using data from several trials: the
`meta-analytic validation method.
`
`It does
`the individual studies be statistically significant.
`necessitate, however, large databases from multiple random-
`ized clinical trials with similar design and treatments. Using
`data from several trials, the method consists of simulta-
`neously estimating the treatment effect (e.g., hazard ratio)
`for the true (survival) endpoint and for the surrogate (PSA)
`endpoint in each trial. The association between the treatment
`effects on the true endpoint and the corresponding effects on
`the surrogate endpoint is then modelled in a way similar to
`standard linear regression (Fig. 3), although mathematically
`more sophisticated. Alike in linear regression, the strength
`of the association is measured by the squared correlation
`coefficient (R2
`trial) that also indicates the precision with which
`the treatment effect on the true (survival) endpoint can be
`predicted from the observed effects on the surrogate (PSA).
`The maximal possible value of R2
`trial is 1 which indicates a per-
`fect prediction. In practice, observing R2
`trial ¼ 1 is not possible
`and one rather seeks a value close to one which indicates a
`strong association between the treatment effects and thus a
`relatively precise prediction.8,10
`
`Published results on PSA surrogacy
`3.
`in prostate cancer
`
`Although the literature concerning the association between
`PSA and long-term outcome with prostate cancer is exten-
`sive, there are relatively few reports of true validation studies
`of this endpoint. We shall critically review the published evi-
`dence assessing PSA endpoints (PSA response, time to PSA
`progression, PSA velocity, PSA doubling time) as potential sur-
`rogate endpoint for overall or progression-free survival, for
`each stage of prostate cancer.
`
`3.1.
`
`Non-metastatic disease
`
`D’Amico and colleagues11 studied the surrogacy of a PSA dou-
`bling time less than 3 months, as a potential surrogate for
`prostate cancer mortality, in a non-randomized cohort of
`5918 men treated with surgery and 2751 with radiation. They
`showed that the Prentice criteria were fulfilled, however the
`fourth condition was demonstrated by showing no effect of
`the initial treatment on the cancer specific survival after
`PSA relapse, in the subset of 1551 patients with PSA relapse.
`The value of the study is limited by the non-randomized nat-
`ure of the series, the fact that the three-month cut-point is
`data driven and the fact that the timing of salvage hormonal
`treatment was not accounted for. The applicability of the re-
`sults is limited by the fact that few patients actually have a
`PSA doubling time shorter than three months (74 of 611 cases
`with PSA relapse after radical prostatectomy, 12%).
`Sandler and colleagues12 showed that in the Radiation
`Therapy Oncology Group (RTOG) trial 92-02 that compared
`short-term versus long-term androgen deprivation in addi-
`tion to irradiation for T2c-T4 prostate cancer, time to PSA fail-
`ure (defined using the American Society for Therapeutic
`Radiation Oncology (ASTRO) definition) was not a surrogate
`for cancer-specific survival: the PSA endpoint failed the
`fourth Prentice criteria. In that study, time to PSA failure
`was longer on the long-term androgen deprivation arm but
`the survival time after PSA failure was shorter. The authors
`
`MYLAN PHARMS. INC. EXHIBIT 1122 PAGE 3
`
`

`

`E U R O P E A N J O U R N A L O F C A N C E R 4 2 ( 2 0 0 6 ) 1 3 4 4 – 1 3 5 0
`
`1347
`
`postulated that on the long-term androgen deprivation arm,
`some patients may have already had hormone insensitive
`disease at the time of PSA relapse and thus decreased respon-
`siveness to salvage treatment. They concluded that time to
`PSA failure should not be used as a surrogate endpoint in tri-
`als that test endocrine treatment of differing duration. Two
`years later, Valicenti and colleagues13 reported from the same
`study, showing that post-treatment PSA doubling time (calcu-
`lated using first-order kinetics on the basis of minimum three
`post-treatment measurements) of less than 12 months ful-
`filled all Prentice’s criteria in respect to the endpoint of pros-
`tate cancer mortality. In their study, 142 of the 1514 eligible
`patients had died of prostate cancer. These two reports sug-
`gest that dynamic measures of PSA might be stronger surro-
`gates than static measures such as the PSA increase above a
`threshold value.
`Newling and colleagues14 carried out a meta-analytic val-
`idation of PSA-doubling free survival (BPFS) as potential sur-
`rogate for clinical progression free survival (PFS) in over 8000
`patients with localized or locally advanced M0 disease, who
`were randomized within AstraZeneca’s Early Prostate Cancer
`Program between treatment with bicalutamide (Casodex) 150
`mg daily versus placebo in addition to standard care (radical
`prostatectomy, radiotherapy or watchful waiting). PFS was
`defined as the time to objectively confirmed disease progres-
`sion or death from any cause. They report an R2
`trial of 0.65
`(95% CI: 0.55–0.92) for the whole group and of 0.52 (95% CI:
`0.37–0.89) on only the European patients, and a lower associ-
`ation in prostatectomy patients (R2
`trial ¼ 0:46) than in irradi-
`ated patients (R2
`trial ¼ 0:65). They concluded that
`large
`positive treatment effects on BPFS are likely to reflect a clin-
`ically important benefit of bicalutamide as regards clinical
`PFS. They estimated the minimum reduction in the risk of
`a PSA doubling to yield a significant reduction (P < 0.05) in
`the risk of a PFS event to P20% in all patients, P30% in rad-
`ical prostatectomy patients and P50% in irradiated patients.
`However, we must note that part of the association observed
`in their study may be induced by the overlap between PFS
`and BPFS for the patients in whom the first event is death
`in absence of PSA doubling.
`
`3.2.
`
`Metastatic disease
`
`As a counter example to time to PSA being a surrogate for sur-
`vival in metastatic disease, we can already mention trial Na-
`tional Cancer Institute (NCI) INT-10515 that randomized 1382
`eligible patients undergoing bilateral to additional flutamide
`or nil. In that study, the treatment differences in post-therapy
`PSA response; defined as a PSA level 64 ng/mL at any time
`after randomization did not translate into survival differ-
`ences; the PSA response rates on treatment and control were
`74.0% versus 61.5% (P < 0.0001) but there was no significant
`difference in overall survival (P = 0.14). The latter may be re-
`lated to a lack of statistical power for survival in this study.
`However, a meta-analysis of 8275 patients later confirmed
`the absence of benefit of maximal androgen blockade over
`castration.16
`More recently, Collette and colleagues17 reported a meta-
`analytic validation of several PSA endpoints (PSA response
`defined as P50% decline from baseline PSA level, PSA nor-
`
`malization, time to PSA progression) as potential surrogates
`for overall survival in a database of 2161 patients with pri-
`mary diagnosis of metastatic prostate cancer who had been
`treated within AstraZeneca’s Casodex (bicalutamide) develop-
`ment program. The patients were randomized between bica-
`lutamide monotherapy and castration or between combined
`androgen blockade with bicalutamide or with flutamide.
`The study showed that the association between the treatment
`effect on any PSA based endpoint and the treatment effect on
`overall survival was in general low (R2
`trial < 0:69 with wide con-
`fidence intervals). The association between the time to PSA
`progression defined as a confirmed 50% relative increase
`above the previously observed nadir yielded R2
`trial ¼ 0:66 (stan-
`dard error = 0.13) with a corresponding 95% confidence inter-
`val ranging from 0.30 to 0.85. Sensitivity analyses using
`prostate-cancer survival as the true endpoint led to similar
`results. Similar to Newling and colleagues,14 they concluded
`that non-null treatment effects on survival would potentially
`be identifiable only in new trials showing a very large effect
`on the PSA endpoint (e.g., HR around 0.50 with standard er-
`ror = 0.10) on the basis of large patient numbers. Moreover,
`irrespective of the size of the effect on the PSA endpoint,
`the prediction of the treatment effect on overall survival
`could not be precise, due to the large unexplained variability
`in the estimated prediction model (as indicated by low R2
`trial
`values). Thus, with the information at hand, a trial based on
`the PSA endpoint would not require fewer patients than sur-
`vival trial.
`
`3.3.
`
`Hormone refractory disease
`
`D’Amico and colleagues18 assessed whether PSA velocity (cal-
`culated by linear regression of all PSA values within one year
`of initially detectable and increasing PSA level) can serve as
`surrogate endpoint for prostate cancer specific mortality
`(PCSM) in 919 patients with non-metastatic hormone refrac-
`tory prostate cancer (HRPC) treated with salvage hormonal
`treatment for PSA failure after initial radical prostatectomy
`or radiation therapy. They demonstrated that a PSA velocity
`>1.5 ng/mL yearly fulfilled the Prentice conditions of surro-
`gacy for the endpoint PCSM. However, only 26 patients died
`of prostate cancer in their study, and their demonstration
`(in particular Prentice’s fourth criteria) is therefore potentially
`affected by lack of statistical power. In addition, the cut-point
`of 1.5 ng/mL yearly was data driven and needs further valida-
`tion in an independent dataset, the study is non-randomized
`and the models used did not control for the timing of the sal-
`vage hormonal treatment. In view of these limitations, the
`authors themselves conclude that they cannot claim that
`they have completely demonstrated surrogacy.
`Crawford and colleagues19 used Prentice’s criteria to dem-
`onstrate the surrogacy of the three-month PSA change (PSA
`velocity) as surrogate for mortality in the SouthWest Oncol-
`ogy Group (SWOG) trial S9916 that compared docetaxel/est-
`ramustine to mitoxantrone/prednisone in 770 patients with
`HRPC. The four Prentice criteria were fulfilled and they con-
`cluded that PSA velocity measured during the first three
`months on study should be further studied as surrogate end-
`point for mortality in future studies of chemotherapeutic reg-
`imens for HRPC.
`
`MYLAN PHARMS. INC. EXHIBIT 1122 PAGE 4
`
`

`

`1348
`
`E U R O P E A N J O U R N A L O F C A N C E R 4 2 ( 2 0 0 6 ) 1 3 4 4 – 1 3 5 0
`
`The findings in the Tax 327 study mentioned earlier4 con-
`flict to some extent with those of Crawford and colleagues
`since the use of PSA changes would have resulted in wrong
`conclusions regarding the weekly docetaxel arm. Therefore
`the question whether PSA endpoints should be used as surro-
`gate in chemotherapy trials or in trials involving docetaxel re-
`mains not fully answered.
`The only meta-analytic validation study in HRPC we
`know of is the study by Buyse and colleagues20 who as-
`sessed several PSA-based end points in androgen-indepen-
`dent patients treated with liarozole, cyproterone acetate or
`flutamide. They showed that despite a strong prognostic
`association neither PSA response (defined as a decline by
`50% or more from baseline level), nor time to PSA progres-
`sion (defined as a greater than 50% increase over nadir va-
`lue) qualified as a surrogate for overall survival (R2
`trial was
`<0.45 for all tested PSA endpoints). One of the reasons for
`the lack of association may relate to the mode of action
`of
`liarozole which is an imidazole-like compound that
`causes elevation of retinoic acid, postulated to have anti-tu-
`mour activity and which effect may not be mediated by
`PSA. Other reasons for the lack of association might be that
`the patient population was very advanced and that PSA
`expression might be affected by tumour de-differentiation.
`This suggests at least that surrogacy of PSA endpoints
`might not be generally applicable between treatments with
`very different modes of actions or which effect on PSA is
`expected to differ substantially.
`
`4.
`
`Discussion
`
`The literature on PSA surrogacy thus far failed to satisfacto-
`rily demonstrate the value of PSA as a surrogate endpoint in
`prostate cancer.
`From this review, one can broadly conclude that for the
`comparison of primary treatments, PSA is until now not pro-
`ven to be a suitable replacement for a final survival endpoint.
`The association between PSA changes after initial treatment
`and survival is likely to diminish in the future, as second
`and third line treatments may become increasingly effica-
`cious. As seen in the RTOG 92-02 trial,12,13 caution is espe-
`cially needed when only one of randomized treatments
`involves long-term hormonal manipulations because PSA will
`not reflect the development of hormone refractory disease,
`that carries poor prognosis for salvage. Vicini and col-
`leagues21 recently reviewed the value of monitoring PSA after
`initial treatment for prostate cancer. They concluded that PSA
`reading should not be used rapidly to judge difference in
`treatment efficacy in this setting.
`The studies on PSA velocity and other dynamic measures
`of PSA changes suggest that, these might be more powerful
`than classical definitions of PSA changes using threshold val-
`ues and the results by D’Amico and colleagues11 need further
`validation. PSA doubling time and PSA velocity have other-
`wise been mostly studied for testing chemotherapeutic
`agents against HRPC. However, it is well documented that
`all pharmacological agents do not affect PSA in the same
`way:22 drugs may decrease, increase or not change PSA, with
`or without a delay after treatment initiation. Therefore the
`PSA endpoint in phase II or phase III trials should be designed
`
`to match the anticipated effect of the tested drugs on PSA lev-
`els.23 In addition, it is also essential to understand and docu-
`ment
`the drug’s effect on tumour growth and how it
`correlates with PSA changes, since drugs, e.g., suramin, were
`shown to modify PSA production without having an impact
`on tumour growth.24 For this purpose, the algorithm proposed
`by Schro¨ der and colleagues25 is very interesting. It incorpo-
`rates an experimental ‘‘proof-of-concept’’ in vivo study before
`or in parallel with the phase II clinical trial. The design of
`phase II trials of targeted agents is further discussed by Sta-
`dler,26 who shows that the Bubley definition27 of PSA response
`in phase II of HRPC is not an appropriate endpoint when test-
`ing cytostatic drugs. Consequently to this and as seen in the
`review, further research is still needed before using measures
`of PSA change as the final endpoint in phase III studies in
`HRPC.
`The meta-analytic validation studies of Newling and col-
`leagues14 and of Collette and colleagues17 in hormonally trea-
`ted patients, confirm only moderate correlation between
`effects of hormonal treatments on the final clinical endpoint
`and on the PSA endpoints considered. Thus, phase III trials in
`these settings should not be based on PSA endpoints.
`However, PSA could still be used to shorten, in two ways,
`the duration of a phase III trial testing a new treatment the ef-
`fect of which is known (from preclinical and early phase stud-
`ies) to be expressed or mediated at least in part by PSA. First,
`early registration on the basis of a PSA endpoint could be
`envisaged in trials with clinical progression or survival as pri-
`mary endpoint. For that purpose, the trial sample size should
`be determined to demonstrate a very large effect on the PSA
`endpoint with great precision: for example to demonstrate
`the presence of a hazard ratio of the order of 0.50 on the
`PSA endpoint, with a standard error of the order of 0.10.
`The trial sample size calculation ought not to be on power
`considerations, as these would necessarily result in very
`small sample size due to the large target effect, but should
`be based on the required precision of the estimation of the
`treatment effect on the PSA endpoint. An interim analysis
`plan should be set up with plan for one or several interim
`looks at the PSA endpoint as well as to the safety data and
`one longer term analysis on the survival endpoint. At the in-
`terim, the trial results on the PSA endpoint could be used to
`estimate a prediction of the survival treatment effect using
`the regression results from former meta-analytic validation.
`Whenever the prediction interval for the survival hazard ratio
`would exclude the null effect, the trial results could be sub-
`mitted for early registration on the basis of the PSA results.
`In the light of the fact PSA is unlikely to capture all the poten-
`tial (negative) effects of the treatment on survival and be-
`cause PSA did not qualify as a surrogate endpoint, we
`recommend that the follow-up should continue to later docu-
`ment long-term safety of the treatments and their impact on
`survival. Of note, this procedure would likely not reduce the
`patient number to enter in studies. Second, even if PSA is
`not a surrogate, a treatment effect on the PSA endpoint might
`be seen, for specific drugs, as a pre-requisite for an ultimate
`effect on survival. Thus along the lines proposed by Royston
`and Parmar28 one could design a study with survival as the
`final endpoint, but with planned interim looks at the PSA end-
`point and a decision to stop the study for futility if insufficient
`
`MYLAN PHARMS. INC. EXHIBIT 1122 PAGE 5
`
`

`

`E U R O P E A N J O U R N A L O F C A N C E R 4 2 ( 2 0 0 6 ) 1 3 4 4 – 1 3 5 0
`
`1349
`
`benefit or a negative effect of the experimental treatment (e.g.
`HR > 0.5) was seen on the PSA endpoint. The assumption
`underlying this design is that a significant treatment effect
`on the final endpoint (survival) would be very unlikely if no
`beneficial effect was seen on the intermediate endpoint
`(PSA). The information on the correlation between the treat-
`ment effects on the PSA endpoint and on survival that is nec-
`essary to the set up of stopping rules according to Royston
`and Parmar is available from the value of R2
`trial obtained from
`meta-analytic validations, whenever they exist.
`PSA is the most widely available marker for prostate can-
`cer. However, PSA is not tumour specific. Prognostic studies
`have also shown that, in hormone independent disease, only
`17% of the variability in survival is explained by PSA.29 There-
`fore, it is unlikely that endpoints based solely on the marker
`PSA can make valid surrogate endpoints for long-term clinical
`outcome with prostate cancer. New serum and urine markers
`in prostate cancer are currently being studied, noticeably
`within the European Community project P-MARK (http://
`www.p-mark.org). A large number of these markers show
`promise to overcome the limitations of PSA30 and may in
`the future offer more solid surrogate endpoints to shorten
`the duration of phase III trials in prostate cancer.
`
`Conflict of interest statement
`
`The authors have no conflict of interest to declare.
`
`R E F E R E N C E S
`
`1. Small EJ, Mack III Roach. Prostate-specific antigen in prostate
`cancer: a case study in the development of a tumour marker
`to monitor recurrence and assess response. Semin Oncol
`2002;19:264–73.
`2. Biomarkers Definitions Working Group. Biomarkers and
`surrogate endpoints: Preferred definitions and conceptual
`framework. Clin Pharmacol Ther 2001;69:89–95.
`3. Johnson JR, Williams G, Pazdur R. Endpoints and United States
`Food and Drugs Administration Approval of Oncology Drugs. J
`Clin Oncol 2003;21:1401–11.
`4. Roessner M, de Wit R, Tannock IF, et al. Prostate-specific
`antigen (PSA) response as surrogate endpoint for overall
`survival (OS): Analysis of the TAX 327 Study comparing
`decetaxel plus prednisone to mitoxantrone plus prednisone
`in advanced prostate cancer. J Clin Oncol 2005;23:391s.
`5. Prentice RL. Surrogate endpoints in clinical trials: definitions
`and operational criteria. Stat Med 1989;8:431–40.
`6. Altman DG, Bland JM. Absence of evidence is not evidence of
`absence. Br Med J 1995;311:485.
`7. Buyse M, Molenberghs G. Criteria for the validation of
`surrogate end-points in randomized experiments. Biometrics
`1998;54:1014–29.
`8. Daniels MJ, Hugues MD. Meta-analysis for the evaluation of
`potential surrogate markers. Stat Med 1997;16:1515–27.
`9. Buyse M, Molenberghs G, Burzykowski T, Renard D, Geys H.
`The validation of surrogate endpoints in meta-analyses of
`randomized experiments. Biostatistics 2000;1:49–68.
`10. Burzykowski T, Molenberghs G, Buyse M, Geys H, Renard D.
`Validation of surrogate endpoints in multiple randomized
`clinical trials with failure-time endpoints. J R Stat Soc Ser C
`Appl Stat 2001;50:405–22.
`
`11. D’Amico AV, Moul JW, Carroll PR, Sun L, Lubeck D, Chen MH.
`Surrogate end point for prostate cancer specific mortality
`after radical prostatectomy or radiation therapy. J Natl Canc
`Inst 2003;95:1376–83.
`12. Sandler HM, Pajak TF, Hanks GE, Porter AT, DeSilvio M, Shipley
`WU. Can biochemical failure (ASTRO definition) be used as a
`surrogate endpoint for prostate cancer survival in phase III
`localized prostate cancer clinical trials? Analysis of RTOG
`protocol 92–02. J Clin Oncol 2003;22:381.
`13. Valicenti R, Deslivio M, Hanks G, et al. Surrogate endpoint for
`prostate cancer-specific survival: Validation from an analysis
`of the Radiation Therapy Group Protocol 92-02. J Clin Oncol
`2005;23:4549.
`14. Newling D, Carroll K, Morris T. Is prostate-specific antigen
`progression a surrogate for objective clinical progression in
`early prostate cancer? J Clin Oncol 2004;22:4652.
`15. Eisenberger MA, Blumenstein BA, Crawford ED, et al. Bilateral
`orchiectomy with or without Flutamide for metastatic
`prostate cancer. N Engl J Med 1998;339:1036–42.
`16. Prostate Cancer Trialists’ Collaborative Group. Maximum
`androgen blockade in advanced prostate cancer: An
`overview of the randomized trials. Lancet 2000;355:
`1491–1498.
`17. Collette L, Burzykowski T, Caroll KJ, Newling D, Morris T,
`Schro¨ der FH. PSA is not a valid surrogate endpoint for
`overall survival in patients with metastatic prostate