Transcript Practical implementation of CRM in real clinical settings for oncology

Practical Implementation
of CRM in Real Clinical Settings for
Oncology Dose-Finding Trials
Xiaobu Ye
Sidney Kimmel Cancer Center,
Biostatistics and Clinical Trials
Johns Hopkins University School
of Medicine
Talk Outline

How are we doing?

What might be the reasons?

What could we do to help?

Are there more challenges
ahead ?
Goal of Dose-Finding Trial in
Oncology
Dose-finding trials in oncology are a
broad class of clinical experiments to
determine an optimal dose (MTD or OBD)
of drug for cancer related treatment or
prevention.
Two Types of Drugs of Interest


Cytotoxic agents (toxicity)
A higher therapeutic index for most cytotoxic drugs is
obtained using a higher dose which yields higher
side-effects
Molecular target agents (mechanism of action)
– Toxicity
– Biological activities which are assumed to be
associated with the clinical outcome of interests
Type of Measurements Used in
Dose-Finding Trials





Toxicity
Pharmacokinetics
Pharmacodynamics
Biomarker
Imaging
Model-based Approach for Dose-Finding Trial
Definition of Dose-response relationship
The relationships among dose, drug concentration in blood, and
clinical response (effectiveness and undesirable effects). ICH-E4
Model-based approaches are generally under some
assumptions

The true dose-response relationship has a biological form;

A mathematical model could mimic observation if empirical data
were collected; and

A model could capture and represent biological knowledge.
CRM is one of the model-based approaches of dose-finding
methods in oncology drug development, and was first proposed
by O’Quigley et al (Biometrics, 1990)
Popularity = Reality
From 1991-2006, among 1,235 phase I oncology trials in US, only
20 (1.6%) were identified using model based approach (A. Rogatko
et al 2007)
There are three parties involved that created the reality:
Statisticians develop sophisticated model-based approaches
and desire for accuracy and precision in estimates;

Clinicians are satisfied with having sufficient assurance that
the selected dose is reasonably safe and desire for simplicity of
trial execution;

Regulatory agency has the primary concern for the safety of
using human subjects for testing without pre-specified dose.

Special Characteristics of Model-based
Approach Compared to Simple 3+3 Design








complex
no assumption of actual dose used
no assumption of response
no assumption of cohort size
not intuitive
use clinical inference throughout trial
need statistical expert
have to use computer program
How Statisticians Deal with the
Challenge
To identify the necessary steps that ease the adaption of CRM
into clinical practice (focus on “simplicity” for clinicians and safety for
regulatory agency)
Planning stage
Working with investigators
Working with regulatory agency (CTEP)
Execution stage
Toxicity grading and modeling
working with investigators
Conclusion stage
detailed written documentation of model-based
dose selection process.
Example
New Approaches to Brain Tumor Therapy
NABTT-Consortium has been
funded by the NCI since 1994 for
therapeutic studies of central
nervous system malignancies
Member Institutions
Cleveland Clinic
Emory University
Henry Ford Hospital
Johns Hopkins University
Massachusetts General Hospital
Primary goal of the consortium
is to improve the therapeutic
outcome for adults with primary
brain tumors.
Moffitt Cancer Center
NCI Neuro-Oncology Program
University of Alabama at Birmingham
University of Pennsylvania
Wake Forest University
Example NABTT
The main task is early anti-cancer drug screen
including dose-finding and safety / efficacy
clinical trials
All NABTT trials

Approved by CTEP and local IRBs

Involve multiple institutions

phase I trial designs were either rule-based or
model-based (modified CRM)

single agent or combination agents
Trials used mCRM method






9-AMINO-20(s)-CAMPTOTHECIN
Pyraxoloacridin
Irinotecan (CPT-11)
Karenitecin
BMS-247550
TMZ+BSI-201
Modified CRM
by Dr.Steven Piantadois
The main points in modification of CRM
used in the NABTT :
1.
A simple probability model, assuming a true
dose-toxicity response is a logistic curve, to
guide data interpolation:
Pr[ toxicity ] 
1
1 e
  ( d  d 50 )
Assumed Underline Dose-Toxicity
Function
Modified CRM
The log-likelihood function for binomial
outcomes and logistic dose response:
  ni 
L (  , d 50 )   log   ni log 1  e   ( di  d50 )
i 1 
 ri 
 (ni  ri ) log 1  e  ( di d50 )

k



The best estimated dose is obtained by using prespecified target toxicity rate and empirical data to fit
the logistic function through maximum likelihood
estimates of Beta and d50.
Modified CRM
2. Use three patients at each dose to stabilize
estimates
3. Use investigator clinical knowledge in the form of
data to make the process easier to understand
4. A flexible computer program to facilitate calculation
with an intuitive user interface to guide through the
dose-finding process
Reference: Piantadosi et al Practical implementation of a modified
continual reassessment method for dose-finding trials, Cancer
Chemother Pharmacol (1998)
The Computer Program User Interface
sp
Initiating the CRM Requires
Information from:




Observations of patients
Quantitative specification of a model
Assumed probability distribution for the
model parameters
Clinical knowledge formalized as “data”
Software Website
Current website:
http://www.cancerbiostats.onc.jhmi.edu/softw
are.cfm
Potential future website:
http://www.csmc.edu/15108.html
Planning Stage
1 working with investigators
The goal is to simplify and ease investigator’s
participation








Prior knowledge in study drug including biological mechanism, side-effect, PD,
PK, and drug half-life etc. from preclinical , or health volunteers or other type of
cancer that had been studied
Type of toxic (side-effects) and its severity by dose level
Formulate a mathematical model that capture the dose and response
relationship
Model specification with a range of a prior initial lower dose to the lethal dose
Modeling the dose-finding trial with several different scenarios
Conceptualizing the definition of dose-limiting toxicity ( this definition may vary
according to different types of cancer)
preparing protocol documentation with dose escalation or de-escalation rule,
procedure and the stopping rule for declare a MTD
Scheduling a 30-60 minute meeting with PI when all information is ready
Issues Requiring PI’s Confirmation

Using CRM method (Giving a demo to investigator for future dose-finding trial
with several different scenarios)









Number of patients per dose cohort
Initial prior dose and toxicity used in the model
Choice of initial testing dose
Definition of dose limiting toxicity
Duration of the treatment
Toxicity evaluation period
Dose escalation or de-escalation rule, procedure and
the stopping rule for declare MTD
Documentation of the first meeting with both
investigator and statistician signatures
Protocol preparation after the initial meeting
Planning Stage
2. working with regulatory agency (CTEP)
The goal is to get approval of an algorithm rather than a set
of pre-specified doses and demonstrate it is safe to perform
a dose-find trial in human subjects using the algorithm






To provide documentation of theoretical elements of using the modelbased approach and include it in the clinical protocol
To provide clinical references (rationale) for initial data (prior) used in
the model
To limit the uncertainty about unspecified testing doses by providing
several steps of potential dose escalation and de-escalation scenarios
using the model predicted results in the protocol
Clearly defined stopping rule (stop when estimated targeting dose
become sufficient accurate)
To define an upper boundary of does increment to an adjacent cohort
If it is possible, to do a real-time demo with CTEP biostatisticians
Example Table Provided in a Protocol
CRM cycle 1
CRM cycle 2
Toxicities
Next Dose
Toxicities
Next Dose
0/3
7.5
0/3
8.7
1/3
7.5
2/3
6.3
3/3
5.7
0/3
6.2
1/3
5.0
2/3
3.2
3/3
Re-evaluate the starting dose
1/3
2/3
5.1
2.3
This dose is below the d10 and will not be
considered as a testing dose
Currently, a reported safe dose from an on-going phase I trial in solid
tumors is XX.
Execution Stage
Statisticians could help:

Study toxicity report

Working with investigator using patient data
to fit the model and estimate next dose for
testing

Prepare an operational report for each
dose cohort including type, severity, and
frequency of the toxicity used to fit the
model for dose estimation
Required Information to Run the Model






Quantified clinical intuition about drug
behavior at higher and lower doses
Target toxicity rate ( assuming a highest
therapeutic index within tolerable side-effect)
Dose
number of patients
r (number of responses (toxicity))
Weight
Initial Dose
Second Dose
Final Dose
Cautions during the Execution Period





Subjectivity in toxicity grading and attribution
External drug information becomes available during
the trial
Clinical judgment versus model prediction
Decision rule to declare an MTD ( avoid split-hair
issues )
Predetermined number of iterations ( revisit model
specification if estimates do not converge after a
predetermined number of steps)
A Partial Operational Report
X number of patients were treated on dose level 1. Two patients
had grade 4 thrombosis during first cycle of the treatment. One
thrombosis was attributed to drug A with possible relationship given
by the treating physician and it was deemed as a DLT based on
pre-specified criteria. The other case of thrombosis was attributed
as unlikely to either drug A or drug B given by a different treating
physician. Due to this attribution, this case of thrombosis will be
weighted as zero with respect to treatment related toxicity in
estimation of next testing dose by CRM method.
The toxicity profile is attached to this report. Dr. X and
statistician Y run the CRM model on <date> to obtain the next
testing dose, dose level 2, for the group2. The new dose was
reviewed by the central office on <date>.
Reporting Stage
Information should be provided in the statistical report:








The type of mathematical model used to guide data interpolation
Rationale for the target toxicity rate ( clinically and biologically)
Dosing steps
Number of patients per dose cohort ( enrolled and actually used for
fitting dose-response model)
Major deviation in toxicity attribution which had effect on estimating the
best dose
Overall model fitting with cumulative data from all cohorts being tested
Clear description on decision of the best dose based on estimation
convergence with sufficient accuracy
Percent of patients treated by dose above the best dose
Limitations


Current mathematical model used to describe
the dose and toxicity relationship is based on
cytotoxic agents. It does not necessarily fit
new paradigm of target agents.
The fraction of increment of the dose works
only best for agent with a continuous dose
such as given through IV, not for discrete
dose ranges, such as by tablets.
Popularity = Implementation
The three parties in the challenging reality:



Clinical investigators: to understand CRM and use the
method in their dose-find trial
NCI statisticians: to confidently accept that the model-based
approach is more efficient than and as safe as conventional
3+3 design
Statisticians: to implement the method in general use with
simple execution procedure and safety boundary for over
dosing control (development and implement)
More Challenges ahead
in Oncology Dose-Finding Trials
What are we looking for in a dose-finding
trial?
A dose that has higher therapeutic effect
for a medical condition and with tolerable
side-effects
Challenges: 1. Model Selection


Cytotoxic anticancer drugs: the optimal dose has
usually been defined as the maximum tolerated dose
(MTD). This toxicity-based dosing approach is under
the assumption that the mechanisms of action of the
toxic and therapeutic effects are the same.
Molecular target based drugs: the dose-effect
relationship is likely to be a biological rather than a
toxicity. Without induction of acute cellular damage,
they are likely to be cytostatic. Most molecularly
targeted drugs are expected to be more selective
and less toxic than conventional cytotoxic drugs (E.
Fox 2002).
Challenges: 1. Model Selection
Mathematical models commonly used to fit
dose-toxicity relationship for cytotoxic drugs are
not necessarily suitable for describing the
relationship of dose-biological activities unless
the dose-biological function is similar to the
relationship of dose-toxicity
Challenges: 1. Model Selection
Probability of Response
1.0
A
C
B
0.5
D
0.0
0
Dose
Challenges: 2. Endpoint Selection






Toxicity
PK guided dose escalation is based on extracellular drug
delivery (plasma concentrations). it dose not have direct
indication of drug uptake at a specific tumor site. It also
requires real-time PK.
PD using biomarker as a therapeutic endpoint requires
sequential tumor biopsies.
Biomarkers require well defined appropriate measure of
achieved target effect and reliable assay given a small cohort
size
Imaging (functional imaging) quantifies the level of target
function in vivo.
Multiple endpoints (toxicity and biological activity) (P.Hung2009)
Challenges: 2. Endpoint Selection
The optimal biological dose based on a therapeutic
end point :
The assays used to measure the biological effect
need to be stabilized (sensitivity and variability
assessment) and validated prior to the initiation of the
phase I trial (E. Fox 2002).
These surrogate measures must be validated and
correlated with the effect of the drug on the target in the
tumor prior to using them as primary end points in
clinical trials (KA. Gelmon, 1999)
Challenges: 3. Joint Effect from
Combined Regimes



Combination of two cytotoxic
agents
Combination of one cytotoxic agent
and another a target agent
Combination of two target agents
Can we capture the complex information we need to
define a best dose and deliver it through a simple
platform for general usage?
Is this A
Challenge ?