Title: Impact of Prior Knowledge on Drug Development Decisions: Case studies across companies
1Impact of Prior Knowledge on Drug Development
Decisions Case studies across companies
- Jaap W Mandema, PhD
- Quantitative Solutions Inc.
- 845 Oak Grove Ave, Suite 100
- Menlo Park, CA 94025
- Ph 650-743-9790
- Email jmandema_at_wequantify.com
- ACPS 10-19-2006
October 19, 2006
2Prior Information is always used for decision
making
- Topic of today
- The use of mathematical models to formally
(quantitatively) use prior information to enhance
decision making
3What do models provide?
- Enhanced Data analysis
- More effective use of the available data,
resulting in increased knowledge and better (more
precise) decision making - Enhanced Trial design
- Better understanding of the data we need and how
best to obtain it to inform future decisions.
4Models improve decision making by combining
multiple pieces of information
- Include information across time points
- Understanding of the time course of response
- Include information across doses
- Understanding of the shape of the dose response
relationship (e.g. Emax model) - Include information across trials
- Accounting for differences in patient populations
(e.g. disease severity) - Include information across drugs
- Understanding similarities in dose response (e.g.
similar Emax for analogues) - Include information across endpoints
- Understanding of link between preclinical,
biomarker and clinical endpoints (e.g. similar
relative potency/ efficacy)
5Trade-off between improved decisions and validity
of assumptions
- Advantage
- Better decisions
- Disadvantage
- Validity of assumptions
6Scope of data integration
- Several to 500 clinical trials
- Several to 15 endpoints
- Preclinical, biomarker, clinical efficacy and
tolerability - Summary level data /- individual patient level
data - Better understanding of impact of patient level
covariates such as disease severity
7Scope of application
- Investment of several large pharma companies
- All therapeutic areas
- From late pre-clinical through post approval
- Models are continuously updated as new
information is obtained - Close collaboration between clinical
pharmacology, statistics and medical specialties
8Example Importance of accounting for differences
between patient populations
9One of the conclusions of the meta-analysis
- The net change in LDL-C is
- Bezafibrate 8 (p0.04)
- Fenofibrate 11 (p0.01)
- Ciprofibrate 8 (p0.005)
- Clofibrate 3 (p0.53)
- Gemfibrozil 1 (p0.68)
- However, the LDL-C response is dependent on the
baseline Lipid profile, which is quite different
from trial-to-trial - Very different relative effects are calculated
when the differences in baseline lipids are
accounted for
10Dependency of LDL effect of Fibrates on baseline
triglycerides
- mean LDL effect in trial normalized for dose and
fibrate - (size sample size)
11Example value of pharmacological assumption
- Meta-analysis of Statins, Ezetimibe, Fibrates,
and Niacin to compare effectiveness/ tolerability
profile as function of dose - Focus on combination products
12With respect to LDL the only difference between
Statins is dose
After adjusting for differences in potency (ED50)
all statins share a common dose response
relationship for LDL
13Interaction between statins and ezetimibe is
characterized by simple interaction model
14A simple interaction model for ezetimibe and
statins
- The interaction for lipid effects could be
described by a simple interaction model - Only 1 additional parameter, ? required to
characterize the magnitude of interaction - ? gt 0 means that the combined effect is greater
than the sum of the effects of the drugs when
given alone. - ? of 0 means that the combined effect is the sum
of the effects of the drugs when given alone. - ? of -1 indicates a pharmacologically independent
interaction. - ? lt -1 indicates a reduced benefit
15Interaction model also characterized statin
gemcabene combination
16Interaction between Atorvastatin and gemcabene
(600 mg) and ezetimibe (10 mg)
17Value of model for development of novel lipid
altering agent
- Validated methodology of response-surface
analysis - Significantly increased power of phase II design
- Enabled assessment of the competitive clinical
profile of a new lipid altering agent when given
alone or in combination with a statin. - Precise quantitative assessment of benefit of
gemcabene/ atorvastatin vs. ezetimibe/
atorvastatin combination
18Example accounting for random differences in
patient populations
- Meta analysis of 19 trials that evaluate
Eletriptan and/ or Sumatriptan
19Pain relief at 2 hoursObserved response (mean,
95 CI)
20Pain relief at 2 hoursResponse adjusted for
differences in placebo effect
21Trial specific Random effects logistic regression
model
- P(Pain Relief)i represents the probability of a
patient achieving pain relief in the jth
treatment arm of the ith trial. - E0 represents the Placebo response Emax is the
maximum response ED50 is the dose required to
get 50 of maximum response. - ?i is a trial specific random effect with mean 0
and variance ?2 to account for the heterogeneity
among the trials. - No additional heterogeneity was found for Emax
22Key question Encapsulation does not impact the
time course of response to Sumatriptan
o Commercial Sumatriptan ? Encapsulated
Sumatriptan
23But so much more was learned about the
differences in speed of onset and magnitude of
response between Eletriptan and Sumatriptan
24Benefit of Eletriptan 40 mg over Sumatriptan 100
mg
25Example value of understanding comparative
clinical profile of anti epileptic drugs (AEDs)
- Comparative trials are limited because of large
sample sizes required - Meta-analysis of 8 newer AEDs to compare
effectiveness/ tolerability profile as function
of dose - Literature data
- FDA/ EMEA websites
- Summary level data on almost 7000 patients with
refractory partial seizures - Efficacy endpoints
- reduction in seizure frequency
- proportion of patients with 50 or greater
reduction in seizure frequency (responders) - Tolerability endpoint
- proportion of patients withdrawing from trial due
to AEs
26Dose response relationship for seizure frequency
27Dose response analysis major findings
- Significant random trial effect (heterogeneity)
on mean response but not on treatment effect,
validating placebo as an internal reference - Significant dose response relationship for each
compound and each endpoint - High correlation between potency estimates for
seizure frequency and responder endpoints - Significant differences between the AEDs in
potency and selectivity for each endpoint, i.e. - Therapeutic window is significantly different
between compounds
28Comparison of Efficacy and Tolerability of AEDs
29Comparison of Efficacy and Tolerability of AEDs
30Value of model for novel AED development
- Provided understanding of competitive landscape
and product opportunities - Aided in quick assessment of potential of new AED
- It is possible to get a good understanding of the
competitive profile of the NCE with limited phase
II data, i.e. small number of doses and limited
sample size
31Example value of biomarker-endpoint models
- Novel anti-coagulant for VTE prophylaxis
- Analyzed dose response data for VTE and bleeding
risk for Heparin, LMWH, Thrombin inhibitors, and
FXa inhibitors after hip and knee surgery - Set targets and identify opportunity
- Scale to NCE on basis of bio-marker data
- Generated biomarker data internally because of
inconsistency of methods - Used to optimize Phase II design for prophylaxis
- Established link between efficacy and safety for
prophylaxis of VTE and treatment of VTE - Acute and chronic treatment period
- Used to select dose for VTE treatment
32Example value of biomarker-endpoint models
- Novel PDE-5 inhibitor intended for the treatment
of male erectile dysfunction - Scale clinical profile of PDE5 inhibitors to NCE
on basis of relative potency (and efficacy)
estimates from preclinical studies and Biomarker
studies (efficacy) and first in man dose
escalation studies (tolerability) - Model identified dose range to study
- Wider instead of narrow range because of
differences among bio-markers - Model allowed for scaling to moderate/mild
patient population to set appropriate targets and
expectations in that patient population. - Model enhanced power of phase II design
- Analysis of prior data jointly with NCE data
reduced sample size from 350 to 200 for equal
decision making power - Model put trial in decision context of ability to
identify dose and competitive positioning for
phase III and not solely showing statistical
benefit vs. placebo. - Better tolerability predicted by biomarker was
confirmed in clinical trial
33Example value of biomarker-endpoint models
- Preclinical and biomarker data show increased
selectivity for beneficial effect vs. AEs for NCE - Biomarker-endpoint model put potency and
selectivity from the biomarker study in a
clinical context - How much more effect can we expect at similar AEs
- Short and directed phase II study can quickly
answer key development uncertainties - Does biomarker selectivity translate into
clinical selectivity? - Is Emax for clinical efficacy large enough to
allow for a meaningful benefit
34Opportunities at FDA
- Important to engage with Industry
- Wealth of Information to mine that can be used
for patient benefit - Understanding of trial-to-trial variability in
response - Explanatory covariates (disease severity)
- Magnitude of random (non-explained) variability
- Safety modeling
- Therapeutic index across drugs is reduced safety
a drug effect or dose effect. - Biomarker linking
- Predictive power of biomarkers (QTc)
35Summary of Value
- Better understanding of competitive landscape and
targets - Better understanding of NCE earlier in
development - Learn from other compounds, endpoints, and
species - Enabling major improvements in clinical trial
design - Better understanding of impact of patient and
disease characteristics - Disease severity
- Special populations
- Objective quantitative assessment of information
- as long as we state our assumptions
36The current trend towardsModel-Based Drug
Development
- There is a tremendous opportunity to integrate
the wealth of public and proprietary data
spanning discovery and clinical into a
probabilistic model of the compounds product
profile in relation to the compounds
competitors. - Utilize the smooth relationship across time, dose
patient characteristics, and endpoints from our
understanding of the underlying pharmacology and
(patho)-physiology. - Models become knowledge repository and provide a
quantitative basis for certain drug development
and regulatory decisions