Operating Characteristics of Tumor Kinetic Response
Assessments in Early Phase Oncology Trials
Dean Bottino1, Arijit Chakravarty2, Eric Westin3 Millennium Pharmaceuticals, Inc., a wholly owned
subsidiary of Takeda Pharmaceuticals Company Limited (1) Clinical Pharmacology, (2) Drug
Metabolism & Pharmacokinetics, (3) Oncology Clinical Research
Summary: While the RECIST criteria have been a
valuable tool in standardizing anticancer treatment
response assessment, they do not take into account the
tumor growth rate prior to treatment intervention, which
can be highly heterogeneous, particularly in phase 1 all-
comers trials. We describe methodology whereby
additional pre-study scans can be used to estimate each
patient's pre-treatment growth rate and therefore
treatment benefit, defined to be the observed deflection
from that initial rate. We show that this methodology
outperforms RECIST percent change from baseline in
terms of accuracy of estimation of antitumor effect, even
when RECIST percent change from baseline is enhanced
to account for ‘placebo’ tumor growth rates. We therefore
anticipate that this kinetic based measure of antitumor
effect will enable more precise dose, exposure, and
biomarker vs. response relationships, leading to more
informed decisions in early oncology development.
128 64 32 16 8
53.3
73.6
106
163
291
1
1
1
2
2
2
median doubling time (weeks) =ln(2)/g
100*s
qrt
(e
g2
-1)=
%C
V o
f gro
wth
rate
g (
k)
p(|2-| 0.1)/p(|
1-| 0.1)
BC
LC
1
2
3
4
5
6
7
89
10
11
12
min
max
0.25
0.354
0.5
0.707
1
1.41
2
2.83
4
Median doubling time(weeks)
%CV o
f gro
wth
rate
g
PTK >
CFB m
eth
od
PTK <
CFB m
eth
od
-8 -1 8 160
50
100
150
200
250
weeks of treatment
tum
or
burd
en (
mm
)
Spaghetti plot: gmedian
=8 wks, CV(g) = 291%
0,
0,)(
)(
0
0
tey
teyty
tkg
gt
obs
freq
g
Median doubling time = 8 weeks = log(2)/gmed
%CV of growth rate g = 291 %
~%CV
gmed
PD
SD PR
-0.8 -0.6 -0.4 -0.2 0 0.2 0.4 0.6 0.8-1
-0.8
-0.6
-0.4
-0.2
0
0.2
0.4
0.6
0.8
1
normalized kill rate
est -
using baseline only: p(|kest
-|<tol*)=0.1461
-0.8 -0.6 -0.4 -0.2 0 0.2 0.4 0.6 0.8-1
-0.8
-0.6
-0.4
-0.2
0
0.2
0.4
0.6
0.8
1
normalized kill rate
est -
using pre-study + baseline: p(|kest
-|<tol*)=0.5784
Log(
Y)
-8 -1 0 8 16w
g gnet
kest = g – gnet
Log(
Y)
-8 -1 0 8 16w
gnet
g
kest = g – gnet
6. Conclusion: Pretreatment Tumor Kinetic (PTK) method, which uses a pre-study scan, is more accurate than Change From Baseline (CFB) method in this polygon spanned by published tumor kinetic parameter values in Breast Cancer1, Lung Cancer2, and 12 all-comers trials.3 This accuracy advantage is maintained over a range of pre-study scan times (4-16 weeks before treatment, results for 8w shown). 1. Heuser et al, Cancer (1979) 43:1888-1894. 2. Usuda et al, Cancer (1994) 74:8. 3. Ferte et al, Clin Cancer Res (2014) 20:46.
PTK method: 57.8%
of GRI estimates GRIPTK are within tolerance of
true growth rate inhibition GRITRUE
CFB method: 14.6%
of GRI estimates GRICFB are within tolerance of
true growth rate inhibition GRITRUE
Therefore PTK method is
times more accurate than CFB method for this choice of
growth rate median & %CV
Pretreatment Tumor Kinetic (PTK) method
Change From Baseline (CFB) method
3.96
1. Each point on the axes below represents a different patient population described by median and spread (%CV) of untreated tumor growth rates.
Tum
or
burd
en (
mm
)
Tumor burden (mm)
2. Next, we draw 10000 patients from each population distribution and simulate their tumor burden time courses, including ~8.5% random measurement error and assuming gtreated/g uniformly distributed between -1 and 1 to generate RECIST response rates typical of targeted ph1 monotherapy trials.
2b. For reference purposes, we report the RECIST response rates from the simulated tumor burden trajectories as a pie chart overlaid on the axes on the right.
3. For each patient, we estimate the study drug’s antitumor Growth Rate Inhibition in two different ways: • The Pretreatment Tumor Kinetic method (PTK, left) uses the patient’s historical pre-study scan to estimate the ‘placebo’ growth rate g. This is the ‘beyond RECIST’ estimate. • The Change From Baseline method (CFB, right) instead uses the median on-treatment growth rate of patients with progressive disease to estimate the ‘placebo’ growth rate g. This was intended to represent the best possible estimate obtainable without a pre-study scan. In both methods, we estimate the net growth rate on treatment gnet and report the ‘Growth Rate Inhibition’ as GRI = 1 - gnet/g.
4. We then compare the PTK and CFB methods to address the key question: what is the incremental benefit of obtaining a pre-study scan in addition to the baseline and on-treatment assessments we typically collect? We do this by summarizing each method’s accuracy as the percent of growth rate inhibition (GRI) estimates falling within a given tolerance of the true GRI originally used to simulate the tumor trajectories. For the particular case shown (8w doubling time, 291% CV), the PTK method, which uses the pre-study scan, is 57.8% accurate while the CFB method, which does not, is only 14.6% accurate.
5. Finally, we report the PTK/CFB ‘accuracy ratio’ for this choice of population parameters as a heat-mapped pixel in the axes above, where green represents population parameters for which the pre-study scan provides additional accuracy and red where it does not.
1-GRITRUE 1-GRITRUE
GRI C
FB-G
RI t
rue
GRI P
TK-G
RI t
rue
Objectives: To investigate
and compare the operating
characteristics of kinetic-
based methods for
quantifying antitumor
effects of investigational
anticancer agents, in
particular the Change From
Baseline (CFB) method and
the Pre-treatment Tumor
Kinetics (PTK) method
which requires an
additional pre-baseline
tumor burden assessment
[1-2].
Methods: Simulated data was generated from a (N=10^5) virtual
patient population having log-normally distributed (exponential) tumor
growth rates (TGRs), with median (mTGR) and %CV (cvTGR) of
TGR tested over ranges encompassing clinically observed values
[1,3,4]. Normalized growth rate inhibition GRI= (growth-kill)/growth
were uniformly distributed from -1 to 1, resulting in simulated
RECIST response frequencies similar to those observed in early phase
oncology trials. Exponential tumor burden measurement error (8.5%)
as fitted from a recent scan-to-scan variability study [5] was also
simulated. For each virtual patient, the CFB estimate of GRICFB was
calculated via log linear regression to assessments at -1 (baseline), 8
and 16 weeks after start of treatment. The PTK estimate GRIPTK for
each patient was calculated via piecewise log linear regression to
assessments at -8 (pre-study), -1 (baseline), 8 and 16 weeks after start
of treatment. Accuracy of each method for a given (mTGR, cvTGR)
parameter set was defined as the fraction of GRI estimates falling
within an arbitrary tolerance (+/- 0.1) of the true GRI values.
Results: While the PTK method was not universally more
accurate than the CFB method over all (mTGR, cvTGR)
parameter values tested, it was consistently more accurate
than CFB over the clinically observed ranges. Specifically,
PTK advantage over CFB was most pronounced in
populations with fast growing tumors and highly
heterogeneous TGRs, while CFB was actually more
accurate than PTK in populations with very slow growing
tumors and relatively homogenous TGRs. Increasing the
time span between the pre-study assessment and the start
of treatment from 4 weeks to 16 weeks further increased
the advantage of PTK over CFB.
Conclusions: While the PTK
method outperforms the CFB
method in all clinically
feasible scenarios tested thus
far, the absolute accuracy
advantage of PTK over CFB
varies from negligible to
significant with increasing
mTGR, cvTGR, and time
between pre-study baseline
scans. This anticipated
accuracy advantage should be
weighed against the minimal
additional cost of reading the
pre-study scan required for
the PTK method.
References
[1] Charles Ferte et al, “Tumor Growth Rate Is an Early Indicator of
Antitumor Drug Activity in Phase I Clinical Trials,” Clin Cancer Res
January 1, 2014 20; 246.
[2] Sylvie Retout et al, “A model-based approach to optimize
detection of treatment effects in early oncology trials,” J Clin Oncol
31, 2013 (suppl; abstr e13508).
[3] Louis Heuser et al, “Growth Rates of Primary Breast Cancers,”
Cancer 43:1888- 1894, 1979.
[4] Katsuo Usuda et al, “Tumor Doubling Time and Prognostic
Assessment of Patients with Primary Lung Cancer,” CANCER
October 15,2994, Volume 74, No. 8.
[5] Geoffrey Oxnard et al, “Variability of Lung Tumor
Measurements on Repeat Computed Tomography Scans Taken
Within 15 Minutes,” J. Clinical Oncology, Volume 29, No. 23,
August 10 2011.