Surrogacy Case Study: Prostate Specific Antigen (PSA ... - Isped

EORTC
Surrogacy Case Study:
Prostate Specific Antigen (PSA)-based
endpoints in Prostate Cancer Clinical
Trials
Laurence Collette, MSc, PhD
Associate Head, Statistics Department
EORTC Headquarters, Brussels, Belgium

Prostate cancer (PCa)
• PCa is a cancer that affects older men (above 50, with
peak incicence around age 70)
• It is a very frequent disease: one in every 6 men will
some day be diagnosed with prostate cancer (PCa)
(Jemal et al. Cancer statistics 2003)
• But it is a very slowly growing disease: PCa accounts
for only 4% of all male deaths (SEER 1973-95)
• PCa is an hormone sensitive disease, the mainstay of
its treatment consists in androgen ablation.

Prostate Specific Antigen (PSA)
• Most studied biomarker in cancer!
• It is a glycoprotein that is produced (mostly) by the
glandular tissue of the prostate and is normally secreted
into semen or lost in urine
• Normally, only very low levels are detectable in the
blood
• Elevated PSA levels are indicative of
• presence of prostate cancer (detection), or
• regarded as signs of disease activity after treatment
(progression). PSA increases antedate clinical disease
progression by several months to years.
• PSA is also a prognostic factor and
• Is used as indicator of response to treatment in some settings

Prostate Specific Antigen (PSA)
• PSA is prostate-specific but not cancer specific.
• PSA is androgen dependent.
• Different therapies may have different effects on PSA
expression.
• Prostate cancer is a heterogeneous disease: not all prostate
cancer express PSA at the same level (eg: lower expression
in poorly differentiated cancer)
• Increases of PSA levels may be induced by
• Prostatitis (infection)
• BPH
• Manipulations: prostate biopsy, TUR
• Stones
• Ejaculation
• Specificity is low!

Transition between clinical states of PCa
Prostatectomy Wait and See Hormones Chemotherapy
Irradiation Hormones (Taxotere)
Salvage RT
Clinically
Localized
Disease
Rising PSA
Clinical
Metastases:
Noncastrate
Clinical
Metastases:
Castrate
Death
median
8 years
12-25% over 15 years
Urology 55:323, 2000
median
3-5 years
median
1.5 years

Transition between clinical states of PCa
Clinical endpoints take too long
shorter endpoints are needed
to fasten treatment development
Prostatectomy Wait and See Hormones Chemotherapy
Irradiation Hormones (Taxotere)
Salvage RT
Clinically
Localized
Disease
Rising PSA
Clinical
Metastases:
Noncastrate
Clinical
Metastases:
Castrate
Death
median
8 years
12-25% over 15 years
Urology 55:323, 2000
median
3-5 years
median
1.5 years

Surrogate endpoint
• An outcome measure that is assessed earlier, more
frequently or more easily than the final endpoint
• That is “reasonably likely” to predict the eventual
treatment benefit on the definitive endpoint
(survival) (FDA definition)
• Requires correlation between treatment effects on the
marker and on the true endpoint, across groups of
patients
• Prostate-specific antigen (PSA)
• PSA changes are indicators of disease evolution
• 50% PSA decrease (“response”)
• PSA increase (“progression”)

How to demonstrate surrogacy ?
• Prentice Criteria: “A test of no treatment effect on
the surrogate is equivalent to a test of no treatment
effect on the true endpoint”
• The treatment has a significant effect on survival
• The treatment has a significant effect on PSA
• PSA is prognostic for survival
• Once PSA is adjusted for, there is no significant treatment
effect on survival
• Meta-analytic validation: The treatment effect on the
Surrogate is used to predict the treatment effect onto
the True endpoint. A valid surrogate is one that
enables a precise prediction.

Meta-analytic validation
First stage: a joint model for
individual observations on surrogate
and true endpoints (Tij,Sij)
Second stage: a linear model for the
trial-specific trt effects ⎛αi
⎞ ⎛α
⎞ ⎛a
i
⎜
⎟ = ⎜ ⎟ +
⎜
⎝ βi
⎠ ⎝ β ⎠ ⎝b
i
⎟ ⎞
⎠
Trial-unit 1
Trial-unit 2
Trial-unit 3
Trial-unit 4
.
.
Trial-unit k
(α PSA
, β OS
)
(α PSA
, β OS
)
(α PSA
, β OS
)
(α PSA
, β OS
)
.
.
(α PSA
, β OS
)
β OS
R 2 trial
close to 1
Slope = R² trial
α PSA
To allow precise prediction of the survival
treatment effect from the PSA treatment
effect
Ref: Buyse et al, Biostatistics 2000; Burzykowski et al, Applied Statistics 2001

Validation Criteria
R² trial =1: surrogate “perfect at the trial-level”
R² trial ≈1: surrogate “valid at the trial-level”
To be meaningful, a good surrogate must also
demonstrate high correlation at the individual
patient level (ie: be a strong prognostic factor)
Ref: Buyse et al, Biostatistics 2000;1:49.
• “how close” to 1 ?
• Computationally intensive
• Requires replicate trials of similar treatments

PSA – defined endpoints in prostate cancer
• PSA endpoints
• “Static” measures
PSA response: 50% decline from baseline (confirmed by 2 nd
observation)
Time to biochemical relapse: time from entry to increase of the
PSA above a certain thrshold (>0.2 ng/ml after RP, 2 increases
after RT, increase by 50% confirmed (M1), increase back to
50% of baseline..)
• “Dynamic measures”
PSA doubling time (PSA-DT) – exponential linear increase
PSA velocity (PSA-V) – linear increase

PSA response and time to PSA
progression (different definitions)
Baseline (B)
Response:
50% below B
C: back
to 50% of
baseline
Nadir (N)
D> N+50%
25
B>N
H. Scher JCO, 2004 (adapted)

PSA “surrogacy” studies in the
literature
• Many papers titled “PSA surrogate” in fact only
demonstrate that PSA changes are prognostic for
clinical relapse or death from the disease
• Since 2000, only 12 publications truly dealing
with PSA surrogacy

Published PSA surrogacy studies
D’Amico
JNCI ’03
Dehnam
Lancet ‘08
Sandler
JCO ’03
Valicenti
IJROBP ’06
D’Amico
J Urol ’05
Newling
JCO ’04
After RP or RT
(cohorts)
After RT+0 or 3 or
6 m ADT (TROG
96.01)
After RT+4 or 24 m
ADT (RTOG 92-02)
After RT+4 or 24 m
ADT (RTOG 92-02)
ADT for PSA
relapse after RP or
RT (cohorts)
M0 on primary
ADT (EPC trial)
PSA-DT PC-surv Prentice
PSA-DT
TTBF
PC-surv
Prentice
TTBF PC-surv Prentice
PSA-DT PC-surv Prentice
PSA-V PC-surv Prentice
T to PSA
doubling
Clinical
relapse
Meta-analysis
RP=radical prostatectomy, RT=radiotherapy, HT=hormonal therapy, M0=non metastatic,
M1=metastatic, PC-surv=Prostate cancer specific survival

Published PSA surrogacy studies
Collette
JCO ’04
M1 on ADT
(Casodex CT’s)
T to PSA
relapse
Overall
survival
Metaanalysis
Crawford
(JCO ’04)
HRPC on TXT
(SWOG 99-16)
PSA
response
Overall
survival
Prentice
Petrylak
(JNCI ’06)
HRPC on TXT
(SWOG 99-16)
PSA-DT
Overall
survival
Prentice
Roessner
(ASCO ’04)
HRPC on TXT
(TAX-327)
PSA
response
Overall
survival
Prentice
Armstrong
(JCO ´07)
HRPC on TXT
(TAX-327)
PSA-DT
Overall
survival
Prentice
Buyse
(Springer ’03)
HRPC on Liarozole
or FLU/CPA
PSA
response
Overall
survival
Metaanalysis
HRPC=hormone refractory prostate cancer, TXT=taxotere (chemotherapy),
FLU=Flutamide (anti-androgen), CPA=cyproterone acetate (anti-androgen)

First remarks
• Many of the papers (especially in the US) still rely
on the Prentice Criteria for validation
• Several papers are based on non randomized
cohorts
• Validity specific to
• Disease stage
• Form of summary statistic (eg: PSA-DT vs PSA
response for HRPC on Taxotere)
• Treatment mechanism of action (HR vs Taxotere vs
Other drugs)

Review
• 2 meta-analytic validations in trials of long term
ADT in M0 or M1 hormone sensitive
• 2 Prentice validations in randomized trials of
durations of ADT adjuvant to irradiation for
locally advanced PCa
• Trials in HRPC – chemotherapy vs cytostatic
agents

PSA surrogate endpoint in M1
prostate cancer?
Bicalutamide (‘Casodex’)
50 mg versus castration 1
(n=530)
Bicalutamide 100/150 mg
versus castration 2,3
(n=870)
N=2161
M1 (D2)
WHO PS: 0-2
Age: 71 years
PSA: 200 ng/ml
META
ANALYTIC
VALIDATION
Bicalutamide 50 mg +
castration versus
flutamide + castration 4
(n=761)
Unit: by study and country
1
Iversen et al. Scand J Urol Nephrol 1996; 30: 93-98
2
Kaisary et al. Eur Urol 1995; 28: 215-222; 3 Tyrrell et al. Eur Urol 1998; 33: 447-456
4
Schellhammer et al. Urology 1995; 45: 745-752

PSA endpoints
PSA normalization
A PSA decline from baseline level (>20 ng/mL)
to ≤ 4 ng/ml at 2 subsequent observations
>4 weeks apart
Time to PSA progression-1
Time to PSA progression-2
Time to a >20% increase above the nadir
and which exceeded 4 ng/ml (UNL)
Time to an increase >50% above the
(3-)moving average nadir and which exceeded
10 ng/ml (2.5xUNL). This increase had to be
either the last observed value or be sustained for
>4 weeks
Longitudinal PSA profile
The complete series of PSA measurements in
each patient

Pooled results
100%
Time to PSA progression
(50% increase)
100%
Overall survival
75%
50%
HR=1.6
75%
50%
HR=1.08
25%
Castration
Casodex
0%
0 1 2 3 4
Time (years)
665/2094 (31.8%) progressed
Median : 19 and 30 months (24.9)
25%
0%
Castration
Casodex
0 1 2 3 4
Time (years)
1143 / 2161 (52.9%) died
Median: 26 and 28 months
5

PSA data (M1)
Mean log(PSA)
-2 0 2 4 6
Association between PSA longitudinal
changes and hazard of death
R²(t) >= 0.90 at any time from m7
0 12 24 36 48 60 72 84 96 108 120 132 144 156 168 180 192
Week

Treatment effects on OS against treatment effects
on Time to PSA Progression (20% rise)
Effect on
survival
(log HR)
1.0
-0.5 0.0 0.5 1.0 1.5
-1.0
2.0
-2.0
-3.0
-4.0
R 2 trial
= 0.21 (CI: 0 – 0.56)
Low trial-unit correlation
-5.0
Effect on TTPP-1 (log HR)
Within patient: concordance coefficient τ = 0.52 (SE=0.004)

Treatment effects on OS against treatment effects
on Time to PSA Progression (50% rise)
-0.5
Effect on
survival
(log HR)
-0.5
0.0
0.5
1.0
-0.5
R 2 trial
= 0.66 (CI: 0.3 – 0.85)
Moderate trial-unit correlation
Effect on TTPP-2 (log HR)
Within patient: concordance coefficient τ = 0.61 (SE=0.02)

Treatment effects on Prostate Cancer Survival
against treatment effects on Time to PSA
Progression (50% rise)
1
R 2 trial
= 0.49 (CI: 0.16 – 0.82)
Moderate trial-unit correlation
0
-1 -.5 0 .5 1
-1
Within patient: concordance coefficient τ = 0.64 (SE=0.02)

Prediction for a new trial – M1
• R² trial for time to PSA progression (50% rise) is 0.66
(95% CI:0.30 – 0.85)
• Assuming a new study of 400 events leading to a HR
for PSA of 0.50 (Z=-6.9, P=

PSA doubling-free survival (PSA-D) as a surrogate for
progression-free survival in M0 hormone sensitive
patients (EPC program)
• EPC program: 3 randomized, double-blind, placebocontrolled
trials of bicalutamide 150 mg/day vs placebo in
addition to either RP, RTx or wawa
• 8113 patients with T1b-4 M0 any N, from 23 countries
• PSA doubling-free survival (PSA-D) = time to first of
• a PSA doubling (value no lower than 0.2 ng/ml to a
value of at least 0.4 ng/ml) or PFS event
• Progression confirmed by bone scan, biopsy, CT scan
or other imaging technique
• Death of any cause
Carroll K. et al, FDA, June ’04
Newling, JCO ‘04

PSA doubling-free survival (PSA-D) as a surrogate
for progression-free survival in M0 hormone
sensitive patients (EPC program)
1.0
Log HR for PFS
0.5
0
-0.5
-1.0
◦ = region
-1.5
-2.0
-2.0 -1.5 -1.0 -0.5 0 0.5 1.0
Log HR for PSA-D free survival
R²=0.69, p=0.001
Newling et al, ASCO 05

Recommendation, M0 and M1 hormone
sensitive patients & hormonal treatments
• PSA does not fully validate as a surrogate endpoint
(R²

Recommendation, M1 hormone sensitive
patients
• Conversely, early marketing authorization might be
sought on the basis of documentation very large effects
on PSA progression (HR

4m vs 24m ADT adjuvant to irradiation of
locally advanced PCa (RTOG 92-02)
T2c-T4
PreRx PSA
< 150 ng/ml
R
A
N
D
O
M
I
Z
E
Arm 1: goserelin and flutamide 2 mos
before and during standard RT (STAD)
1554 patients
448 deaths
145 of PC
Arm 2: goserelin and flutamide 2 mos before
and during standard RT, followed by goserelin
alone for 24 mos (LTAD)
Ref: Valicenti RK et al. Int J Rad Oncol Biol Phys 2006

4m vs 24m ADT adjuvant to irradiation of locally
advanced PCa (RTOG 92-02)
• Biochemical failure (ASTRO) failed the Prentice validation
• LADT was superior to SADT for PC-survival (P=0.006) and
for Biochemical failure (P

4m vs 24m ADT adjuvant to irradiation of
locally advanced PCa (RTOG 92-02)
• PSA-DT estimated based on minimum 3
measurements (ASTRO definition of PSA failure)
• Median follow-up 5.9 years
• PSA-DT

PSA-DT and TTBF surrogate for PCa-specific
survival after RT+ADT? TROG 96.01
Locally
advanced
PCa
R
RT alone
RT + 3 months ADT
RT + 6 months ADT
PC-
Death
• PSADT ≤ 12 m and ≤ 15 m satisfied all Prentice
criteria (PTE of 0.36 a 0.56 resp)
• TTBF ≤ 1.5, 2 or 2.5 y satisfied all Prentice criteria
(PTE between 0.45 & 0.64)
?!??
Ref: Denham JW et al. Lancet 2008

Explanation?
• Time to testosterone recovery relates to duration of
ADT and strongly influences the time to biochemical
relapse
Results obtained on trials with only short term ADT
may not generalize to studies randomizing between
longer durations of ADT
Other comments (from Denham et al.):
1. PSADT strongly dependent on #PSA values available
2. Cut-point on PSA-DT or TTBF not clearly defined
3. Need a meta-analysis validation

PSA response or PSA-V
surrogate for survival in HRPC?
• Meta-analytic validation in 2 studies of liarozole versus
CPA/FLU (1)
• PSA response (50% decline) not a surrogate for OS
• SWOG-9916: Docetaxel+Estramustine vs M+P (2)
• PSA response (first 3 m) not a surrogate for OS
• PSA velocity (first 3 m) is a surrogate for OS by Prentice
criteria and should be further investigated
• Tax 327: Docetaxel(D)(q1s)+Prednisone(P) vs D(q3s)+P vs
Mitoxantrone(M)+P (3)
• PSA response (50% decline from entry) not validated as
surrogate for OS
• Other changes in PSA or PSA kinetics are weaker
surrogates
1. Buyse M. et al In Biomarkers in Clinical Drug Development. Springer-Verlag, 2003
2. Crawford et al. ASCO 2004, abs. 4505, Petrylak DP. J Natl Cancer Inst 2006.
3. Roessner M et al. ASCO 2005, abs. 391s.; Armstrong AJ et al. JCO 2007

Ref.: Petrylak et al. JNCI 2006
Of interest… in SWOG trial

Of interest…
The search for the golden cut-point…
Testing as much as 10…
you’re certain to find something…

Tax 327 mimicking SWOG

Conclusions for HRPC on
chemotherapy
• Bubley et al (PCWG1, JCO ‘1995) proposed to use
a 50% PSA decline as primary endpoints in phase
II trials in HRPC Since then most phase II trials
used it as primary endpoint in phase II and
secondary endpoint in phase III trials
• However, despite an apparent strong intraindividual
correlation, PSA-response is NOT a
surrogate in HRPC, even treated with Docetaxel.
• PSA-V or PSA-DT may be better endpoints

ASCENT trial
High dose Calcitriol (DN-101) + Docetaxel
vs Placebo + Docetaxel
• 250 men with M1 HRPC randomized to
• Primary endpoint: PSA response (first 6 months
defined as 50% decline)
• Secondary endpoint: Overall survival, PSA
progression-free survival
R
Docetaxel + Placebo
Docetaxel + DN-101
Ref. Beer TM et al, JCO 2007

ASCENT trial
High dose Calcitriol (DN-101) + Docetaxel
vs Placebo + Docetaxel
• Results:
DN-101 Placebo P-value
PSA response 58% 49% 0.16
PSA-PFS 7.9 m 7.6 m 0.7
Overall survival 24.5m 16.4 m P=0.04
HR=0.67
Ref. Beer TM et al, JCO 2007

Kaplan-Meier curve of overall survival in ASCENT trial
Beer, T. M. et al. J Clin Oncol; 25:669-674 2007

Be aware of your drug’s effect on the marker
Broken Arrow
(Immuno, APC8501)
No effect
(anti-angiogenic drug as sorafeni
Start of Therapy
Lead In Post-therapy
Delayed effect
(Differentiating agent)
Rapid decline
(HT, CT, Suramin)
Scher et al.

Proposed strategy for preparing the use of a marker
in phase II
• Solid preclinical and clinical data are needed about the
drug’s effect on PSA, it’s shape, it’s timing (remember
Calcitriol, APC 8501, Suramin..)
• Neo-adjuvant studies are required to document
• The effect of the new drug (alone or in the intended
combination) on the marker
• Correlation between changes in the marker and observed
changes on the tumor
• Without such information, even phase II trials with a markerbased
endpoint, be it response or PFS, are a very risky
gamble
• New guidelines (PCWG2, JCO 08) move away from the PSA
response endpoint
• Research is being done to make bone metastases changes
measurable to use them as phase II endpoint