V O L U M E 2 3 䡠 N U M B E R 2 5 䡠 S E P T E M B E R 1 2 0 0 5
`
`JOURNAL OF CLINICAL ONCOLOGY
`
`O R I G I N A L R E P O R T
`
`From the European Organisation for
`
`Research and Treatment of Cancer
`
`Data Center, Brussels; Limburgs
`
`Universitair Centrum, Diepenbeek,
`
`Belgium; AstraZeneca Pharmaceuticals,
`
`Macclesfield, United Kingdom; and
`
`Erasmus Medical Centrum, Rotterdam,
`
`the Netherlands.
`
`Submitted August 27, 2004; accepted
`
`May 11, 2005.
`
`Authors’ disclosures of potential con-
`
`flicts of interest are found at the end of
`
`this article.
`
`Address reprint requests to Laurence
`
`Collette, MSc, European Organization for
`
`Research and Treatment of Cancer Data
`
`Center, Avenue Emmanuel Mounier
`
`83/11, B-1200 Brussels, Belgium; e-mail:
`
`laurence.collette@eortc.be.
`
`© 2005 by American Society of Clinical
`
`Oncology
`
`0732-183X/05/2325-6139/$20.00
`
`DOI: 10.1200/JCO.2005.08.156
`
`Is Prostate-Specific Antigen a Valid Surrogate End Point
`for Survival in Hormonally Treated Patients With
`Metastatic Prostate Cancer? Joint Research of the
`European Organisation for Research and Treatment of
`Cancer, the Limburgs Universitair Centrum, and
`AstraZeneca Pharmaceuticals
`Laurence Collette, Tomasz Burzykowski, Kevin J. Carroll, Don Newling, Tom Morris,
`and Fritz H. Schröder
`
`A
`
`B
`
`S
`
`T
`
`R
`
`A
`
`C
`
`T
`
`Purpose
`(OS) slows down the
`The long duration of phase III clinical trials of overall survival
`treatment-development process. It could be shortened by using surrogate end points.
`Prostate-specific antigen (PSA) is the most studied biomarker in prostate cancer (PCa). This
`study attempts to validate PSA end points as surrogates for OS in advanced PCa.
`
`Patients and Methods
`Individual data from 2,161 advanced PCa patients treated in studies comparing bicalutamide
`to castration were used in a meta-analytic approach to surrogate end-point validation. PSA
`response, PSA normalization, time to PSA progression, and longitudinal PSA measurements
`were considered.
`
`Results
`The known association between PSA and OS at the individual patient level was confirmed.
`The association between the effect of intervention on any PSA end point and on OS was
`generally low (determination coefficient, ⬍ 0.69).
`
`Conclusion
`It is a common misconception that high correlation between biomarkers and true end point
`justify the use of the former as surrogates. To statistically validate surrogate end points, a
`high correlation between the treatment effects on the surrogate and true end point needs to
`be established across groups of patients treated with two alternative interventions. The
`levels of association observed in this study indicate that the effect of hormonal treatment on
`OS cannot be predicted with a high degree of precision from observed treatment effects on
`PSA end points, and thus statistical validity is unproven. In practice, non-null treatment
`effects on OS can be predicted only from precisely estimated large effects on time to PSA
`progression (TTPP; hazard ratio, ⬍ 0.50).
`
`J Clin Oncol 23:6139-6148. © 2005 by American Society of Clinical Oncology
`
`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 un-
`
`derstanding of the biologic mechanisms of
`disease development have resulted in the
`emergence of a large number of potentially
`effective new agents. There is also increasing
`public pressure for promising new drugs to
`receive marketing approval as rapidly as
`
`6139
`
`Downloaded from ascopubs.org by Univ of Chicago Library on October 25, 2016 from 205.208.121.028
`Copyright © 2016 American Society of Clinical Oncology. All rights reserved.
`
`MYLAN PHARMS. INC. EXHIBIT 1121 PAGE 1
`
`

`

`Collette et al
`
`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
`duration of phase III trials results from the use of long-term
`clinical end points (clinical progression and survival).
`Therefore, to replace this end point (the “true” end point)
`by another (“surrogate”) end point that could be measured
`earlier, more conveniently, or more frequently and would
`adequately reflect the benefit of new treatments on the
`clinical end point(s) seems to be an attractive solution.
`“Biomarkers” (ie, physical signs or laboratory mea-
`surements that occur in association with a pathological
`process or that have putative diagnostic and/or prognostic
`utility1) are generally regarded as the best candidate surro-
`gate end points. A biomarker is an intermediate outcome
`that is correlated with the true clinical outcome for an
`individual patient. It may be useful for diagnostic or prog-
`nostic information on a particular patient. It is a common
`misconception that established biomarkers necessarily
`make valid surrogate end points. To this aim, it is required
`that “the effect of treatment on a surrogate end point must
`be reasonably likely to predict clinical benefit.”2 Thus, “sur-
`rogacy” is a concept that relates to groups of patients. To
`demonstrate surrogacy, a strong association between the
`treatment effects on the surrogate and on the true end point
`needs to be established across groups of patients treated
`with the new and standard interventions.
`The validation of a candidate surrogate end point is not
`straightforward. Until recently, the statistical methods de-
`veloped for this purpose used the data from a single trial.3-5
`These methods suffer from numerous drawbacks: some of
`them are too stringent to be of practical value, whereas
`others are based on nontestable assumptions.6,7 To over-
`come these limitations, a new methodology, known as
`the “meta-analytic” validation approach, was developed
`recently.8-10 This method uses large databases from multi-
`ple randomized clinical trials and aims at measuring di-
`rectly the association between the treatment effects on the
`surrogate and the true end point.
`In the field of prostate cancer (PCa), prostate-specific
`antigen (PSA) has probably been the most studied biomar-
`ker. It has been investigated as a potential surrogate end
`point across disease stages,11-14 and in hormone-refractory
`patients in particular.15-18 In a recent article, Buyse et al19
`considered several PSA-based end points in androgen-
`independent patients treated with liarozole (an imidazole-
`like compound that causes elevation of retinoic acid,
`postulated to have antitumor activity), cyproterone acetate,
`or flutamide. They showed that despite a strong association
`at the individual patient level, none of the end points qual-
`ified as a surrogate for overall survival (OS). In early PCa,
`Newling et al20 found a modest correlation between the
`effect of Casodex on time to PSA progression (TTPP) and
`on objectively confirmed progression. In primary meta-
`
`static PCa, several studies demonstrated some level of asso-
`ciation between a post-therapy fall in PSA or a PSA relapse
`on treatment and long-term survival prognosis.21-25 How-
`ever, this merely qualifies PSA as a biomarker. In trial NCI-
`INT-105, treatment differences in post-therapy PSA levels
`did not translate into survival differences.26 Thus, whether
`PSA is a valid surrogate for survival in hormonally treated
`PCa remains an open question. This question is of impor-
`tance, because the use of PSA could shorten the time to the
`end point from between several months in advanced dis-
`ease27 to several years in early disease.28
`The objective of the present research is to assess PSA-
`based end points as surrogates for OS in hormone-naı¨ve
`metastatic PCa using the meta-analytic approach. The data
`from ⬎ 2,000 patients treated with bicalutamide (Casodex)
`that were made available by AstraZeneca Pharmaceuticals
`were used for this purpose.
`
`PATIENTS AND METHODS
`
`Individual data from three large international randomized trials of
`AstraZeneca’s Casodex Development Program were used (301/
`302,29,30 306/307,31 and US trial 000132,33; Table 1). In studies
`301/302 and 306/307, Casodex monotherapy (50 and 150 mg/day,
`respectively) was compared to medical or surgical castration. In
`the US trial, Casodex (50 mg/day) in combination with goserelin
`or leuprolide acetate was compared to the combination of flut-
`amides (750 mg/day) and castration in a 2 ⫻ 2 factorial design.
`All patients were newly diagnosed with metastatic PCa. Four hun-
`dred eighty patients with T3-4 M0 disease and elevated PSA from
`trial 306/307 were excluded. Survival was an end point in all studies,
`although time to treatment failure (Table 1) was the primary end
`point in most. PSA was monitored at months 1, 2 (except US trial),
`and 3 and then every 3 months until month 18 (trial 301/302) or
`death (other trials). For the analysis, the PSA test date was assumed
`to be the visit date.
`
`End Points
`We considered OS calculated from randomization to the date
`of death or last visit as the true end point. PCa-specific survival was
`defined similarly but with deaths unrelated to PCa or treatment
`censored at the last visit. PSA response, PSA normalization, TTPP,
`and the complete series of PSA measurements (“PSA profile”)
`were successively assessed as potential surrogate end points for OS.
`Patients who had a baseline PSA level at least five times above
`the normal range (⬎ 20 ng/mL) were included in the analyses of
`PSA response and PSA normalization. Patients qualified for PSA
`response if their PSA declined by at least 50% from baseline level
`at two subsequent observations at least 4 weeks apart. Patients
`in whom the decline reached a value below or equal to normal
`(4 ng/mL) qualified for PSA normalization.25
`Two definitions of TTPP were assessed: (1) For TTPP-1, PSA
`progression was defined as a PSA value above normal (4 ng/mL),
`representing a first increase ⱖ 20% above the nadir25 (eg, with a PSA
`nadir of 2 ng/mL, a minimum increase to 4 ng/mL [100% increase] is
`required, whereas with a PSA nadir of 3.5 ng/mL, a 20% increase to
`4.2 ng/mL is enough). (2) For TTPP-2, PSA progression was defined
`as a PSA value ⬎ 2.5 times the normal range (10 ng/mL), representing
`a first increase ⱖ 50% above the moving average (based on three
`
`6140
`
`JOURNAL OF CLINICAL ONCOLOGY
`
`Downloaded from ascopubs.org by Univ of Chicago Library on October 25, 2016 from 205.208.121.028
`Copyright © 2016 American Society of Clinical Oncology. All rights reserved.
`
`MYLAN PHARMS. INC. EXHIBIT 1121 PAGE 2
`
`

`

`Is PSA a Valid Surrogate End Point?
`
`Table 1. Trials Used in the Analysis
`
`Trial
`
`301/30228,29
`Patients
`
`Treatments
`
`Design
`
`Objective
`
`Stage D2, fit for orchidectomy; ECOG performance status 0-2; no prior systemic therapy for prostate cancer, no previous
`radiotherapy to the prostate within 3 months of entry
`
`Bicalutamide (50 mg/d) v castration (orchidectomy in trial 301, orchidectomy or goserelin 3.6 mg monthly injection in trial
`302)
`
`Open two-arm randomization
`
`To compare bicalutamide to castration in a pooled analysis
`
`Efficacy end points
`
`Time to treatment failure (objective progression, change of treatment, death as a result of any cause)ⴱ; overall survival
`
`Results
`
`306/30730
`Patients
`
`Treatments
`
`Design
`
`Objective
`
`Efficacy end points
`
`Results (in M1)
`
`US trial31,32
`Patients
`
`Treatments
`
`Design
`
`Objective
`
`Efficacy end points
`
`Results
`
`Bicalutamide (50 mg/d) demonstrated significantly worse time to progression and survival in trial 301; the trend was not
`significant in trial 302; by pooled analysis, both end points were significantly worse with bicalutamide than with
`castration
`
`Metastatic (M1) or locally advanced with PSA five-fold in excess of the upper normal limit (T3-4 M0); only the M1
`patients were included in the presently reported analyses; fit for orchidectomy; ECOG performance status 0-2; no prior
`systemic therapy for prostate cancer, no previous radiotherapy to the prostate within 3 months of entry
`
`Bicalutamide (100 or 150 mg/d) or castration (medical or surgical at the patient’s discretion)
`
`Initially 2 (Casodex 100 mg):2 (Casodex 150 mg):1 (castration) then changed to 2:1 randomization between Casodex
`150 mg and castration
`
`To demonstrate noninferiority of Casodex 150 mg in comparison to castration by excluding a risk increase of 25%
`
`Time to treatment failure (addition of systemic therapy or withdrawal from therapy, objective progression, or death)ⴱ;
`overall survival; objective response
`
`Significant differences in favor of castration were found for time to treatment failure (HR, 1.43; 95% CI, 1.20 to 1.71 in
`favor of castration) and overall survival (HR, 1.30; 95% CI, 1.04 to 1.64)
`
`Stage D2 only; ECOG performance status 0-2; no prior systemic therapy for prostate cancer
`
`Bicalutamide (50 mg/d) v flutamide (250 mg tid) in combination with goserelin acetate (3.6-mg monthly injection) or
`leuprolide acetate (7.5-mg monthly injection)
`
`2 ⫻ 2 factorial design, blinding for the LHRH-A randomization
`
`To demonstrate noninferiority of bicalutamide ⫹ LHRH-A relative flutamide ⫹ LHRH-A by excluding a relative-risk
`increase of 25%
`
`Time to treatment failure (addition of systemic therapy or withdrawal from therapy, objective progression, or death)ⴱ;
`overall survival
`
`Noninferior time to treatment failure (HR, 0.93 in favor of bicalutamide; 95% CI, 0.79 to 1.10) Noninferior overall survival
`(HR, 0.87 in favor of bicalutamide; 95% CI, 0.72 to 1.05)
`
`Abbreviations: ECOG, Eastern Cooperative Oncology Group; PSA, prostate-specific antigen; HR, hazard ratio; LHRH-A, luteinizing hormone-releasing
`hormone agonist.
`ⴱA rising PSA was not considered a sign of progression in any of the studies.
`
`consecutive measurements) nadir. This increase had to be either the
`last observed value or be sustained for at least 4 weeks19 (eg, with a
`nadir of 2 ng/mL at three consecutive occasions, a 500% increase
`to 10 ng/mL is needed to reach the end point, whereas after a nadir
`of 7 ng/mL, a 50% increase to 10.5 ng/mL is enough).
`Patients who died or are alive without PSA progression were
`censored at the time of death or last visit, respectively.
`
`Statistical Methods
`The meta-analytic approach to surrogate end-point valida-
`tion has been detailed extensively elsewhere.6,9,34-36 Thus, we shall
`only summarize the key features. The method is rooted in the
`concept that a valid surrogate end point must enable one to predict
`with sufficient precision the treatment effect on the true clinical
`end point (OS) from the observed treatment effect on the surro-
`gate (PSA-based) end point. Unlike traditional validation meth-
`ods such as the Prentice criteria,3 this new methodology does not
`require that any of those effects be statistically significant. Indeed,
`when data from several trials are available, the method consists of
`simultaneously estimating the relative treatment effects on the
`survival end point and on the PSA end point (log odds ratio of PSA
`response or normalization, log hazard ratio [HR] of PSA progres-
`sion, treatment effect on the longitudinal PSA measurements) in
`
`each trial. A model that estimates the association between the
`treatment effects on the true end point and the corresponding
`effects on the PSA end points (PSA response,34 TTPP,35 or longi-
`tudinal PSA measurements36) in a way similar to standard linear
`regression (although mathematically more sophisticated) is then
`adjusted. As in linear regression, the strength of the association is
`measured by the squared correlation coefficient that we shall de-
`2
`. This coefficient also indicates the precision with which
`note Rtrial
`the treatment effect on the survival end point can be predicted
`from the observed treatment effect on the surrogate. The maximal
`2
`is 1, which indicates a perfect prediction. In
`possible value of Rtrial
`2 ⫽ 1 is not possible, and one rather seeks a
`practice, observing Rtrial
`value close to 1, which indicates a strong association between the
`treatment effects and thus a relatively precise prediction.9,35 Addi-
`tionally, the model quantifies the association between the PSA-based
`end point and the survival end point at the individual patient level.
`Parameters quantifying the strength of the association at this level will
`be denoted by the subscript “patient.” They can be regarded as mea-
`sures of validity of the PSA end point as a biomarker for predicting
`duration of survival.
`Only three trials were available, which is too few to allow a
`2
`. Therefore, the patients were grouped
`precise estimation of Rtrial
`
`www.jco.org
`
`6141
`
`Downloaded from ascopubs.org by Univ of Chicago Library on October 25, 2016 from 205.208.121.028
`Copyright © 2016 American Society of Clinical Oncology. All rights reserved.
`
`MYLAN PHARMS. INC. EXHIBIT 1121 PAGE 3
`
`

`

`Collette et al
`
`by the trial they entered and their country of residence, as done
`by Buyse et al.19 These groups will be henceforth referred to as
`“trial units.”
`
`RESULTS
`
`After excluding nonmetastatic patients and those with no
`baseline or follow-up PSA measurements, the individual
`data from 2,161 patients classified into 21 trial units were
`available for the analysis (Table 2). Their baseline and treat-
`ment characteristics are listed in Table 3. More than half of
`the patients presented with six or more bone metastases.
`After a median follow-up of 3.25 years, 1,018 patients
`(52.9%) had died, 815 (71.3%) as a result of PCa (Table 4).
`The median OS was 2.2 years (95% CI, 2.1 to 2.5) for the
`Casodex-treated patients and 2.3 years (95% CI, 2.1 to 2.6)
`in the pooled control groups (Fig 1). The average number of
`PSA assessments per patient was 6.9 (range, 1 to 23)
`
`PSA Response (> 50% Decline From Baseline)
`and PSA Normalization
`PSA response could be assessed for 1,853 patients. A total
`of 974 (89.4%) and 687 (90.0%) assessable patients on the
`Casodex and control groups, respectively, achieved a PSA re-
`sponse (Table 4). Only thirteen trial units representing 1,606
`patients were used in the analysis: two trial units were removed
`because no deaths were observed in the castration group, and
`six were removed because all patients responded in one or both
`treatment arms. At the individual level, PSA response was a
`strong predictor of prolonged survival with a survival odds
`
`Table 2. Trial Units Available for the Analysis (N ⫽ 2,161)
`
`Country
`
`Canada
`
`United States
`
`Denmark
`
`Norway
`
`Sweden
`
`Austria
`
`The Netherlands
`
`United Kingdom
`
`Denmark
`
`Finland
`
`Norway
`
`Sweden
`
`Australia
`
`Austria
`
`Belgium
`
`Germany
`
`The Netherlands
`
`Italy
`
`Republic of South Africa
`
`Spain
`
`United Kingdom
`
`N
`
`114
`
`647
`
`158
`
`75
`
`63
`
`46
`
`29
`
`159
`
`83
`
`69
`
`83
`
`86
`
`35
`
`14
`
`95
`
`47
`
`35
`
`11
`
`48
`
`22
`
`242
`
`Trial
`
`US
`
`US
`
`301
`
`301
`
`301
`
`302
`
`302
`
`302
`
`306
`
`306
`
`306
`
`306
`
`307
`
`307
`
`307
`
`307
`
`307
`
`307
`
`307
`
`307
`
`307
`
`6142
`
`patient of 1.94 (SE, 0.33), representing a two-fold increase
`ratio ␪
`in the odds of surviving beyond any specified time t for the PSA
`responders compared to the nonresponders. At the trial level,
`the effects of hormonal intervention on PSA response and on
`2 ⫽ 0.08 (SE, 0.14; 95% CI,
`OS were poorly correlated with Rtrial
`0.0 to 0.49). Figure 2A presents the estimated treatment effects
`on the response (log odds ratio) and OS (log HR).
`One should be careful in interpreting these results,
`because eight trial units with extreme results were excluded
`from the analysis.
`In 399 (36.6%) and 380 (49.8%) of the assessable pa-
`tients, the PSA declined to a value ⱕ 4 ng/mL. Seventeen
`trial units representing 1,778 patients could be used for this
`analysis: four were excluded for same reasons as above. At
`patient for pa-
`the individual level, the survival odds ratio ␪
`tients with PSA normalization compared to those without
`was 4.90 (SE, 0.52), indicating a 4.9-fold greater odds
`of surviving any specified time t for the patients whose
`PSA normalized. At the trial level, the treatment effects on PSA
`2 ⫽ 0.41 (SE,
`and on OS were moderately correlated with Rtrial
`0.18; 95% CI, 0.05 to 0.72; Table 5). Figure 2B presents the
`estimated treatment effects on PSA normalization and OS.
`
`PSA Progression
`Nineteen trial units (2,070 patients) and 18 trial units
`(2,043 patients) could be used for the analysis of TTPP-1 and
`TTPP-2, respectively (two trial units were excluded from both
`analyses because of absence of deaths in the castration arm and
`one from the TTPP-2 analysis because of the absence of PSA
`progressions in both treatment arms).
`The TTPP-1 is presented in Figure 3A: 54.6% of the
`patients progressed according to this definition (Table 4)
`within a median time of 11.1 months after being randomly
`assigned. TTPP-1 was somewhat shorter for the pooled
`Casodex group than for the control group. TTPP-1 was
`moderately associated with OS at the individual patient
`⫽ 0.52 (SE, 0.004)
`level: the concordance coefficient ␶
`patient
`indicates that for each individual patient there is an approx-
`imately 50% chance to observe a long (short) OS given a
`long (short) TTPP. At the trial-unit level, the association
`between the effects of Casodex on TTPP-1 and on OS was
`2 ⫽ 0.21 (SE, 0.17; 95% CI, 0.0 to 0.56; Table
`low, with Rtrial
`5). This analysis is depicted in Figure 4A, where the treat-
`ment effect on survival is regressed against the treatment
`effect on TTPP-1: the size of the circles represents the trial-
`unit size. The low trial-level association may be partly be-
`cause of the outlying data from one trial unit. Excluding this
`unit from the analysis leaves the individual-level association
`2
`to
`⫽ 0.52; SE, 0.004) but increases Rtrial
`unchanged (␶
`patient
`0.58 (SE, 0.15; 95% CI, 0.20 to 0.81).
`Only 31.8% of the patients met the more stringent
`criterion TTPP-2 (Table 4) at a median time of 24.9 months
`(Fig 3B). At the patient level, the association of TTPP-2
`and OS was somewhat stronger than for TTPP-1, with a
`
`JOURNAL OF CLINICAL ONCOLOGY
`
`Downloaded from ascopubs.org by Univ of Chicago Library on October 25, 2016 from 205.208.121.028
`Copyright © 2016 American Society of Clinical Oncology. All rights reserved.
`
`MYLAN PHARMS. INC. EXHIBIT 1121 PAGE 4
`
`

`

`Is PSA a Valid Surrogate End Point?
`
`Table 3. Patient Characteristics
`
`Age
`
`Baseline PSA
`
`Mean
`
`SD Median
`
`First and Third
`Quartiles
`
`Performance Status
`0/1/2/3/4, %
`
`Mean
`
`SD
`
`Median
`
`First and Third
`Quartiles
`
`301/302
`
`Total (N ⫽ 530)
`
`Casodex 50 mg (N ⫽ 262)
`
`Castration (N ⫽ 268)
`
`306/307 (UICC M1 pts.)
`
`Total (N ⫽ 870)
`
`Casodex 100/150 mg (N ⫽ 617)
`
`Castration (N ⫽ 253)
`
`US (D2 pts.)
`
`Total (761)
`
`Casodex ⫹ castration (N ⫽ 377)
`
`Data not available
`
`Data not available
`
`Data not available
`
`71.6
`
`71.2
`
`72.7
`
`70.2
`
`69.8
`
`8.2
`
`8.2
`
`8.1
`
`8.7
`
`8.2
`
`72
`
`72
`
`73
`
`70
`
`70
`
`66, 78
`
`66, 77
`
`67, 78
`
`65, 76
`
`65, 75
`
`Data not available
`
`Data not available
`
`Data not available
`
`53.8/32.8/13.3/0/0.1
`
`54.0/31.9/14.1/0/0
`
`53.4/34.8/11.5/0/0.4
`
`839.1
`
`811.2
`
`866.3
`
`747.3
`
`772.6
`
`685.6
`
`1,551.3
`
`1,477.8
`
`1,622.2
`
`1,657.2
`
`1,772.5
`
`1,336.0
`
`51.4/37.2/11.4/0/0
`
`53.8/36.1/10.1/0/0
`
`694.2
`
`650.4
`
`1,444.2
`
`1,382.8
`
`267.9
`
`273.2
`
`266.7
`
`179.1
`
`189.8
`
`156.0
`
`174.3
`
`170.0
`
`178.3
`
`98.6, 784.7
`
`98.3, 840.0
`
`99.4, 713.3
`
`65.7, 634.7
`
`64.5, 658.4
`
`67.0, 587.3
`
`45.6, 580.6
`
`53.8, 588.1
`
`38.7, 576.5
`
`Flutamide ⫹ castration (N ⫽ 384)
`
`70.5
`
`9.2
`
`71
`
`65, 77
`
`49.0/38.3/12.8/0/0
`
`737.3
`
`1,502.6
`
`Abbreviations: PSA, prostate-specific antigen; SD, standard deviation.
`
`⫽ 0.61 (SE, 0.02). The asso-
`concordance coefficient ␶
`patient
`ciation between the treatment effects on TTPP-2 and OS
`2 ⫽ 0.66
`was somewhat higher than for TTPP-1, with Rtrial
`(SE, 0.13; 95% CI, 0.30 to 0.85; Fig 4B and Table 5).
`
`Longitudinal Measurements of PSA
`All previously considered PSA-based end points are sum-
`mary measures derived from the longitudinal PSA measure-
`ments and use only a limited amount of the available
`information. It thus seemed logical to investigate if the longi-
`tudinal series of PSA measurements would not be a better
`surrogate end point for OS. Figure 5A presents the mean
`profiles of log-transformed PSA measurements for groups of
`patients with similar observation time: all profiles eventually
`end with a PSA increase (progression), and patients with an
`early progression tend to have a higher initial PSA that does not
`decrease as much early on.
`
`Figure 5B displays the mean PSA profiles per treatment
`group: starting from week 52 the curves show a relatively stable
`linear decrease rather than the increasing curvature observed
`in Figure 4A. This distortion results from attrition: progressive
`patients, in whom PSA increases, tend to leave the study, and
`thus the curve in Figure 5B reflects only those with stable PSA.
`In view of Figure 5A, the treatment effect on the log-
`transformed PSA levels was expressed as a function of time and
`its square root in a joint model of PSA measurements and
`survival times. In that model, the individual patient-level asso-
`ciation between the PSA process and the hazard of dying is a
`function of time and cannot be easily summarized into a single
`measure.35 The results indicated that the correlation between
`the individual PSA and mortality hazard processes was ⬎ 0.90
`at any time ⬎ 7 months, which suggests a strong association
`between the PSA profile and the hazard of dying for individual
`
`Table 4. Survival and Prostate-Specific Antigen Outcome
`
`Casodex (n ⫽ 1,256)
`
`Control (n ⫽ 905)
`
`Total (N ⫽ 2,161)
`
`Alive
`
`Dead
`
`Because of prostate cancer
`
`Because of another cause
`
`PSA response
`
`Evaluable
`
`Decline to ⱕ 4 ng/mL
`
`Decline by ⱖ 50% of baseline
`
`No response
`
`Not evaluable
`
`PSA progression (TTPP-1)
`
`PSA progression (TTPP-2)
`
`Not evaluable for PSA progression
`
`No.
`
`571
`
`685
`
`496
`
`189
`
`1,090
`
`399
`
`575
`
`116
`
`142
`
`415
`
`432
`
`35
`
`%
`
`36.6
`
`52.8
`
`10.6
`
`No.
`
`447
`
`458
`
`319
`
`139
`
`763
`
`380
`
`307
`
`76
`
`166
`
`729
`
`233
`
`32
`
`%
`
`49.8
`
`40.2
`
`10.0
`
`No.
`
`1,018
`
`1,143
`
`815
`
`328
`
`1,853
`
`779
`
`882
`
`192
`
`308
`
`1,144
`
`665
`
`67
`
`Abbreviations: PSA, prostate-specific antigen; TTPP, time to PSA progression.
`
`www.jco.org
`
`Downloaded from ascopubs.org by Univ of Chicago Library on October 25, 2016 from 205.208.121.028
`Copyright © 2016 American Society of Clinical Oncology. All rights reserved.
`
`%
`
`52.9
`
`71.3
`
`28.7
`
`42.0
`
`47.6
`
`10.4
`
`6143
`
`MYLAN PHARMS. INC. EXHIBIT 1121 PAGE 5
`
`

`

`Collette et al
`
`DISCUSSION
`
`Using data from the Casodex Development Program, we
`investigated whether the biomarker PSA could be used to
`define a valid surrogate for OS in patients with metastatic
`PCa. The analyses confirm the known value of PSA as a
`biomarker of prognosis and disease activity (individual-
`level association). When comparing groups of patients
`treated with Casodex-based or control treatment, however,
`the association between the treatment effect on any PSA-
`based end point and the treatment effect on OS was low in
`2 ⬍ 0.69 with wide confidence intervals).
`general (Rtrial
`2
`The choice of the threshold for Rtrial
`required for a valid
`surrogate is still a matter of debate.6 Nevertheless, one can
`argue that the precision of the prediction of the treatment
`effect on OS from the effect on the PSA-based end points,
`2
`indicated by the Rtrial
`values observed in the present
`study, is insufficient to claim any of the assessed PSA-
`based end points as a statistically valid surrogate end
`point for OS in phase III clinical trials of hormonal
`treatment in metastatic PCa.
`To illustrate the problem, let us consider a new trial
`with TTPP as the primary end point (defined as TTPP-2),
`where data analysis occurs after 400 events and yields an HR of
`0.75 for PSA progression (with 400 events, SE [log{HR}]
`would be of the order of 0.10, resulting in P ⬍ .01). Without
`adjusting for the estimation error in the parameters of the
`prediction model, one could predict with approximately 95%
`confidence that the corresponding survival HR would lie
`within the interval 0.48 to 1.12. Adjustment for the estimation
`error would widen the confidence interval even further; thus,
`non-null treatment effects on survival would potentially be
`identifiable only in large new trials showing a large effect on the
`PSA end point (eg, HR approximately 0.50 with SE ⫽ 0.10).
`Buyse et al19 assessed similar PSA-based end points as
`candidate surrogates for OS in patients with androgen-
`independent PCa treated with liarozole versus antiandro-
`gen monotherapy. In their study, the association between
`treatment effects at the trial level were generally low, with
`2 ⬍ 0.45 for all tested PSA end points. They concluded
`Rtrial
`that PSA end points could not be regarded as valid surro-
`gates for OS. The reasons for the lack of association in their
`study may be different than ours; the disease was more
`advanced, and treatment mode of action differed. In early
`disease, for which the time savings of using PSA could be
`greater than in advanced disease, Newling at al20 also found
`only moderate correlation between the effect of Casodex on
`PSA progression and objective clinical progression.
`Unfortunately, in cancer and other diseases, biomark-
`ers that are strong predictors of the clinical end point for the
`individual patient often proved to be poor surrogate end
`points.37-43 Several authors have discussed biologic and
`medical reasons why biomarkers often fail to validate as
`
`Fig 1. Overall survival by randomized treatment.
`
`patients. At the trial-unit level, the association between the
`effect of Casodex on the longitudinal PSA and OS was slightly
`2 ⫽ 0.68; SE, 0.12; Table 5).
`higher than that for TTPP-2 (Rtrial
`
`Fig 2. The treatment effects on survival and prostate-specific antigen (PSA)
`response. The circles represent the observations in the trial units, and their
`size is proportionate to the trial-unit sample size. The line represents the
`prediction from an estimated (weighted) regression line. (A) ⱖ 50% decline
`2 ⫽ 0.08. (B) PSA normalization (PSA ⱕ 4 ng/mL):
`from baseline level: Rtrial
`2 ⫽ 0.41. HR, hazard ratio; OR, odds ratio.
`Rtrial
`
`6144
`
`JOURNAL OF CLINICAL ONCOLOGY
`
`Downloaded from ascopubs.org by Univ of Chicago Library on October 25, 2016 from 205.208.121.028
`Copyright © 2016 American Society of Clinical Oncology. All rights reserved.
`
`MYLAN PHARMS. INC. EXHIBIT 1121 PAGE 6
`
`

`

`Is PSA a Valid Surrogate End Point?
`
`Table 5. Summary of the Results
`
`Patient-Level Association Between PSA
`and Survival
`
`Trial-Level Association Between PSA
`and Survival
`
`PSA End Point
`
`PSA response (decline by ⱖ 50% from baseline)
`
`PSA normalization (ⱕ 4 ng/mL)
`
`TTPP-1
`
`TTPP-2
`
`Longitudinal PSA measurements
`
`␪
`patient
`
`⫽ 1.94
`
`␪
`patient
`
`⫽ 4.90
`
`␶
`patient
`
`⫽ 0.52
`
`⫽ 0.61
`
`␶
`patient
`⬎ 0.9 at all times ⬎ 7 mo
`
`R
`
`2
`patient
`
`SE
`
`0.33
`
`0.52
`
`0.004
`
`0.02
`
`—
`
`2
`R
`trial
`
`0.08
`
`0.41
`
`0.21
`
`0.66
`
`0.68
`
`SE
`
`0.14
`
`0.18
`
`0.17
`
`0.13
`
`0.12
`
`95% CI
`
`0 to 0.49
`
`0.05 to 0.72
`
`0 to 0.56
`
`0.30 to 0.85
`
`Undetermined
`
`
`
`Abbreviations: PSA, prostate-specific antigen; TTPP, time to PSA progression; ␪patient, survival odds ratio; ␶patient, concordance coefficient between time to
`
`
`PSA progression and duration of survival.
`
`surrogate end points.2,37,39,44 The principal explanation is
`that only a part of the treatment effect on the true clinical
`end point will be reflected in the biomarker, which may lead
`to over- or underestimation of the treatment effect on the
`true end point from the observed effect on the biomarker.
`Baker and Kramer45 mention that perfect predictors of the
`true end point at the patient level do not necessarily make
`
`good surrogate end points, because the prediction function
`could differ between randomized treatments and thus
`would induce incorrect inference on the true end point.
`The inability thus far to demonstrate surrogacy for PSA
`can be explained by several biologic mechanisms. PSA is also
`produced by normal prostatic tissue, and the amount present
`may vary between patients. Poorly differentiated tumors
`
`Fig 3. Time to prostate-specific antigen (PSA) progression (TTPP) by
`randomized treatment. (A) TTPP-1: time to the first 20% increase of PSA
`over previously observed nadir to a value above the upper limit of the normal
`PSA range (4 ng/mL). (B) TTPP-2: time to the first 50% increase of PSA over
`previously observed moving average nadir to a value ⬎ 2.5 times the upper
`limit of the normal PSA range (10 ng/mL), sustained for at least 4 weeks.
`
`Fig 4. The treatment effects on time to prostate-specific antigen (PSA)
`progression (TTPP) and on overall survival. The circles represent the
`observations in the trial units, and their size is proportionate to the trial-unit
`sample size. The line represents the prediction from an estimated
`2 ⫽ 0.66.(weighted) regression line. (A) TTPP-1: Rtrial2 ⫽ 0.21. (B) TTPP-2: Rtrial
`
`
`HR, hazard ratio.
`
`www.jco.org
`
`6145
`
`Downloaded from ascopubs.org by Univ of Chicago Library on October 25, 2016 from 205.208.121.028
`Copyright © 2016 American Society of Clinical Oncology. All rights reserved.
`
`MYLAN PHARMS. INC. EXHIBIT 1121 PAGE 7
`
`

`

`Collette et al
`
`erations led Scher et al18 to conclude that PSA may not be an
`appropriate end point for clinical trials of first-line hor-
`mona