Presentation to AMIA 2009, San Francisco, CA, in panel S58: Applications of Biomedical Informatics in Clinical Drug Development.

1. Personalized Medicine In Clinical
Drug Development: Opportunities For
Biomedical Informatics
Zhaohui (John) Cai
AMIA 2009
San Francisco, CA
1
Proprietary and Confidential © AstraZeneca 2009
Nov 17, 2009 FOR INTERNAL USE ONLY

2. Presentation Outline
• Background
• Personalized Medicine and Personalized Healthcare
(PHC) in AZ
• Modeling and Simulation (M&S) for PHC: opportunities
for Biomedical Informatics (BioMed Ix)
• Case study 1: modeling for predicting treatment responders
vs. non-responders for better efficacy
• Case study 2: modeling for identifying patients with high
safety risks
• M&S for PHC Integrated into a Clinical Program
2
Proprietary and Confidential © AstraZeneca 2009
Nov 17, 2009 FOR INTERNAL USE ONLY

3. Personalized Medicine or Personalized Healthcare
• Based on the recognition that
unprecedented types of information will be
obtainable from genetic, genomic,
proteomic, imaging, etc, technologies,
which will help us further refine known
diseases into new categories
• Managing a patient's health based on the
individual patient's specific characteristics
vs. “standards of care”
• PHC in AZ to focus on therapies linked to
diagnostics and tools to deliver superior
outcomes to patients
• PHC in AZ to deliver:
• Disease segmentation
• Patient selection
• Improved dosing
3
Proprietary and Confidential © AstraZeneca 2009
Nov 17, 2009 FOR INTERNAL USE ONLY

4. M&S for PHC: Opportunities for Biomedical Informatics
• Wide application of the new technologies to clinical trials
has not come to reality in the pharmaceutical industry, for
all kinds of reasons, such as
• Limitations in trial designs
• Extra cost and time
• Uncertainly in regulatory and commercial consequences
• A cost-effective approach is M&S using available data and
technologies
• The industry and FDA have now a broader use and
acceptance of M&S
• Cheaper, faster, and easier to integrate into clinical programs
(arguable)
• Many M&S application types: biological (from cell to system to
disease), pharmacological (PK/PD, drug-disease/drug-
patients/efficacy and safety*), clinical trial modeling and
simulation, HEOR modeling, etc.
4 * Opportunities for BioMed Ix
Proprietary and Confidential © AstraZeneca 2009
Nov 17, 2009 FOR INTERNAL USE ONLY

5. Case Study 1: Identify Treatment Responders
Treatment effect in overall patient population
Placebo Treatment
Treatment effect in patient subpopulations defined by
baseline biomarker levels
Placebo
Treatment Blue: survivors
Red: non-survivors
Marker A ≤ xxx Marker A > xxx
5
Proprietary and Confidential © AstraZeneca 2009
Nov 17, 2009 FOR INTERNAL USE ONLY

6. Models to Predict Survival In Treatment Group
24 hrs
72 hrs
120 hrs
Baseline
Random Forest models using
common 46 variables across 4
time points
6
Proprietary and Confidential © AstraZeneca 2009
Nov 17, 2009 FOR INTERNAL USE ONLY

7. Models to Predict Survival In Placebo Group
120 hrs
72 hrs
24 hrs
Baseline
Random Forest using common
46 variables across 4 time points
7
Proprietary and Confidential © AstraZeneca 2009
Nov 17, 2009 FOR INTERNAL USE ONLY

8. Variable Importance Plots
Top predictors in placebo Top predictors in in treatment
group (prognosis markers) group (efficacy markers)
Variables
Variables
8
Nov 17, 2009 Importance Score
Proprietary and Confidential © AstraZeneca 2009
FOR INTERNAL USE ONLY

9. Predictive Biomarkers: Most Important Variables For Survival On
Treatment But Least Important On Placebo
9
Proprietary and Confidential © AstraZeneca 2009
Nov 17, 2009 FOR INTERNAL USE ONLY

10. Potential Application to Phase 3: Marker-based Vs. Traditional
Design (with and without stratified analysis)
Traditional design: Placebo
Register Randomize
Treatment Traditional analysis
Stratified analysis
Marker-based design: Placebo
Marker A > cutoff Randomize
Treatment
Register Test marker
Placebo
Marker A <= cutoff Randomize
Treatment
Interim analysis Final analysis
• Additional risk: a test with a quick turn around time for Marker A
• Benefit:
Better chance to demonstrate mortality improvement and allow a personalized medicine
approach with this product
Smaller sample size and shorter trial duration if interim analysis shows significance for
Marker A <= cutoff arms.
10
More ethical if the treatment is not beneficial to patients with Marker A >cutoff
Proprietary and Confidential © AstraZeneca 2009
Nov 17, 2009 FOR INTERNAL USE ONLY

11. Case Study 2: Identify Patients at High Safety Risk
.
Using biomarkers to predict individual patient risk of developing liver
signals in response to a drug
11
Proprietary and Confidential © AstraZeneca 2009
Nov 17, 2009 FOR INTERNAL USE ONLY

12. Question: Who will develop liver signals during the trial?
Purpose: patient selection / risk stratification (tolerators vs. adaptors and susceptibles)
Data: Baseline Labs+ Demographics + Concomitant Medications + Medical History
Classification models: Abnormals (FDA-guideline cut-offs) vs. Normals
Abnormals
Max(LFT)
Liver Chemistry test
1xULN
Normals
1 2 i-2 i-1 i 5 Visits
Result (based on 5 projects, 24 studies)
Baseline values Marker A Markers B+C Marker D
Model Important for predicting
Normals Abnormals AT>3 ALP>1.5 Bilirubin>1.5
12
Proprietary and Confidential © AstraZeneca 2009
Nov 17, 2009 FOR INTERNAL USE ONLY

13. Predictive Models Using Baseline Information
13
Proprietary and Confidential © AstraZeneca 2009
Nov 17, 2009 FOR INTERNAL USE ONLY

14. Predictive Baseline Variables for Biochemical Hy’s Law Cases
During The Trials
Importance
Baseline variables
14
Proprietary and Confidential © AstraZeneca 2009
Nov 17, 2009 FOR INTERNAL USE ONLY

15. Biomarker-based Risk Stratification to Improve Patient Safety
• Identify patients with high risk of developing liver signals
• Better patient risk management
• Cost-effective biomarker research
• Being applied to a live project in transition to phase III
• Potential applications of the predictive biomarkers
• Trial protocol for close monitoring of the high-risk subpopulation
(e.g. those with marker A > xxx)
• New exclusion criterion for trials as appropriate (e.g. excluding
those with marker A > xxx)
• Warnings in product label: marker A should be obtained before
starting therapy. If marker A > xxx, do not start therapy or apply
close monitoring
15
Proprietary and Confidential © AstraZeneca 2009
Nov 17, 2009 FOR INTERNAL USE ONLY

16. M&S for PHC Integrated into a Clinical Program
Which patients will benefit most from the therapy (i.e. w/
most effectiveness and least safety risk)?
Data Model/ Historical Preclinical/
mining Hypothesis Literature trials
Initial Phases 1 & 2a
question
Literature Biological
mining interpretation Hypothesis &
Candidate Biomarker(s) initial modeling
/model Learn*
Phase 2b
Model validation Validated Biomarker(s) Design and analysis
/model
Phase 3
Model application Patient stratification
Design and analysis
Confirm
Outcome
* Opportunities for BioMed Ix (a PHC product)
16
Proprietary and Confidential © AstraZeneca 2009
Nov 17, 2009 FOR INTERNAL USE ONLY

17. Acknowledgements
• AZ Biomedical Informatics Network
• AZ Hepatotoxicity Safety Knowledge Group
• AZ Clinical Project Team for AZDxxxx
• AZ Clinical Information Science Leadership
• AZ Discovery Information
• AMIA CRI-WG
17
Proprietary and Confidential © AstraZeneca 2009
Nov 17, 2009 FOR INTERNAL USE ONLY