The Utility of Modeling and Simulation in Understanding Locally Acting Drug Delivery Liang Zhao, Ph.D. Director, Division of Quantitative Methods and Modeling Office of Research and Standards Office of Generic Drugs Center for Drug Evaluation and Research, FDA AAPS, Denver November 13th, 2016
44
Embed
The Utility of Modeling and Simulation in Understanding Locally … · The Utility of Modeling and Simulation in Understanding Locally Acting Drug Delivery Liang Zhao, Ph.D. Director,
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
The Utility of Modeling and Simulation in Understanding Locally Acting Drug Delivery
Liang Zhao, Ph.D.
Director, Division of Quantitative Methods and Modeling
Office of Research and Standards
Office of Generic Drugs
Center for Drug Evaluation and Research, FDA
AAPS, Denver November 13th, 2016
2
The Disclaimer
• This presentation represents the views and perspectives of the speaker and do not necessarily reflect the views of the FDA.
Background• BE demonstration for locally acting products is challenging.
• Recent advancement in science created several innovative pathways for BE of locally acting products in addition to clinical endpoints BE studies.
– Better analytical tools
– Better product characterization
– Quantitative methods and modeling
• M & S Impact: Model based guidance development for complex locally acting products will ensure timely availability of high quality and affordable generics for patients.
5
Quantitative Approaches for Generics
Industry
• Limited inclusion in ANDAs
• Conventionally used in
– Model based product design
– BE study design and assessment
Regulatory
• Complex substance, e.g., Copaxone
• Locally acting products: topical, nasal, inhalation, ocular, and GI
• NTI
• Long acting injectable products
• Drugs with steep exposure-response relationships
• In vitro BE assessment
• Safety surveillance
• Others
6
Core DQMM Tool Set
Oral Drug
PK-PD model
Population model
Our Goal is to support Generic drug research Policy development Regulatory decisions
Release/ Absorption/PBPK Models
Pharmacometrics
Big Data
Non-Oral Drug
Analytics for complex mixturesSystems pharmacologyRisk modelsBusiness process models
7
Role of PBPK and Pharmacometrics for locally acting drugs
• PBPK models– Link in vitro performance with in vivo performance (PK)
– Predict the local tissue concentration in special populations including pediatric, geriatric, and disease populations
– Identify critical quality attributes for BE demonstration
– Define clinically relevant specification
• Pharmacometrics– Perform assessment of new BE criteria
– Evaluate clinical significance of failing BE
– Sensitivity of clinical endpoints
8
Currently Posted BE Recommendation for GI Locally Acting Drug Products
Method Drugs Comment
In vitro only calcium acetate, colesevelam HCl, sevelamer carbonate, sevelamer HCl
• Current BE recommendation– In vivo PK fasting and fed in healthy subjects
– Partial AUCs• AUC0-3 and AUC3-t for ER capsules
• AUC8-48 for DR tablets and DR capsules
– In vitro comparative dissolution• pH 4.5, 6.0, 6.5, 6.8, 7.2, and 7.5
PK or clinical endpoints for different products
PK endpoints + in vitro comparative dissolution
15
Commonly Asked Questions• Why PK?
– PK as a surrogate of in vivo dissolution
• Why pAUC?– Ensure similar distribution in different GI segments
• Why in vitro dissolution?– Capture pH range not covered in healthy populations
– Capture the range of trigger pH of the coating polymer
• Why not in patients?– Comparing formulation differences not performance
differences in different populations
– Extremely high variability in the inflammatory response both inter and intra patient.
Physiologically and pathologically relevant time
Partial AUC Search
Collect PK data after administration for 5 products
Calculate the PK metrics for each subject (dose normalized to 1.2mg)
Design the combinations of PK metrics to be analyzed
Linear Discriminant Analysis (LDA) for each combination of metrics
Randomly split the whole data set into training and testing.
Construct the model using training data and apply the model to the testing set.
AUC0-3,
AUC3-t
16
17
PBPK Absorption Modeling Approaches
Model was developed
based on i.v., suspension, and suppository PK
data.
Fit pH dependent dissolution
profiles as model input for in vitro
dissolution.
Adjust pH in the GI lumen against
observed PK profiles for each
subject.
Perform simulation to
answer specific questions.
18
Model Development and Validation
Suspension
ER capsule
Suppository
i.v.
19
Colon and plasma exposure correlation for ER capsules
Partial AUC3-t was identified
20
GDUFA research study supports BE recommendation
• Study Objective: To study the correlation between GI and plasma concentrations
• Treatment 1– Pentasa ER capsules: 500 mg capsule x 2
– Apriso ER capsules : 375 mg capsule x 3
– LialdaDR tablets: 1200 mg tablet x 1
• Treatment 2: – Mesalamine oral solution: 100 mg/125 mL water, followed by 125 mL water
• Each subject will go through treatment 1, then treatment 2
Randomized: Pentasa, Apriso, or
LialdaSolution
GI and plasma concentration was measured
Only plasma concentration was measured
21
Human Subjects
Number of subjects screened 65
Number of subjects participated 34
Number of procedures conducted
38 treatment 1 (MR arm)20 treatment 2 (IR arm)
Number of procedures completed
26 treatment 1 (MR arm)20 treatment 2 (IR arm)
Fasting 10 hours prior to study drug administration
Mean tube depth 170 cm
The study was approved by local IRB and FDA Research Involving Human Subjects Committee
22
0 2 4 6 8 10 120
20
40
60
80
100
Time (hr)
% D
isso
lved
dissolution comparison of Apriso, Pentasa, and Lialda
Apriso, pH6.8
Lialda, pH7.2
P, pH1.5
P, pH4.0
P, pH6.0
P, pH6.5
P, pH6.8
P, pH7.5
Dissolution vs. pH
23
Location Pentasa Apriso Lialda
Plasma
StomachpH ~ 1.5
DuodenumpH ~ 5-6
Proximal Jejunum
pH ~ 6-6.2
Mid JejunumpH ~ 6.4
Distal JejunumpH ~ 6.4
0
200000
400000
600000
0 2 4 6 8
0
200000
400000
600000
0 2 4 6 8
0
200000
400000
600000
0 2 4 6 8
0
200000
400000
600000
0 2 4 6 8
0
200000
400000
600000
0 2 4 6 8
0
1000
2000
0 2 4 6 8
23
0
200000
400000
600000
0 2 4 6 8
0
200000
400000
600000
0 2 4 6 8
0
200000
400000
600000
0 2 4 6 8
0
200000
400000
600000
0 2 4 6 8
0
200000
400000
600000
0 2 4 6 8
0
1000
2000
3000
0 2 4 6 8
0
2000
4000
6000
0 2 4 6 8
0
2000
4000
6000
0 2 4 6 8
0
2000
4000
6000
0 2 4 6 8
0
2000
4000
6000
0 2 4 6 8
0
2000
4000
6000
0 2 4 6 8
0
500
1000
0 2 4 6 8
Time (hr)
Co
nce
ntr
atio
n (
ng
/mL)
24
Conclusions
• Based on modeling and simulation, partial AUC were identifedthat correlates with colon exposure.
• Based on the in vivo GI tube study, GI local concentration measurement results are qualitatively consistent with formulation design and in vitro dissolution.
• Based on the in vivo GI tube study, GI local concentration measurement results are qualitatively correlated with PK metrics (certain partial AUCs).
• In vivo GI tubing study and absorption modeling and simulation confirmed that PK endpoint study is a reasonable surrogate to reflect GI local concentration.
25
Case II: Modeling and simulation evaluation for using residual levonorgestrel as one of BE metrics
26
Background
• Active ingredient: Levonorgestrel (LNG)• Nominal release rate: from 20 μg/day at beginning
to 11 μg/day after 5 years • Indication: Contraception up to 5 years, treatment
for heavy menstrual bleeding• Features: the most effective; long-acting;
reversible; contraception• Modeling and simulation based evaluation of
residual Levonorgestrel can be used for an exploration of BE metric and the associated criteria.
• PK/PD modeling approach to assess the feasibility to use plasma concentration as a BE metrics
27
Questions to be Addressed and Challenges
• Conventional plasma pharmacokinetic (PK) endpoints to establish bioequivalence challenging:
• M & S was used to identify and assess potential BE metrics and statistical criteria to promote development of generic LNG IUS:
– Time
– # of Subjects
28
Source Data for Analysis• Residual Levonorgestrel (LNG) data from an array of IND and NDAs with
study durations of 1, 3, and 5 years
• Product Specification: No more than 1 of 10 results lies outside of 47 – 57 mg/system and no result lies outside of 45 – 60 mg/system.
• Created virtual population (n = 1000) incorporating the variability, under following assumptions:
1. SD of 1.56 (CV of 3%) for A (mean 52 mg) provided a range of initial LNG in 47 – 57 mg/system.
2. For release rate constant different CVs (i.e. 5 %, 10% and 15%) for k were tested and a CV of 15 % (SD of 5.835 × 10-5) with a mean release rate constant of 3.89 × 10-4 per day provided a reasonable explanation of observed residual LNG. Additive error model for release rate constant was used.
NDA 21-225 S-030 CMC Review
29
Study Design
• BE analysis was performed on residual LNG from virtual subjects ( n = 16) for RLD and hypothetical generics with different mean release rate constants (k) varying from 5% to 100% at 6 months, 1, 2, 3, 4 and 5 years using 80% - 125% BE limit.
• BE analysis was also performed between formulation D vs C and D vs Another Drug X to check if they would have passed current BE criteria at the end of 5 years.
• Phoenix/R/Berkeley Madonna software were used for the analysis.
30
Residual LNG from Virtual Population (n = 1000) and Observed data
A (CV = 0.03), K (CV = 0.15)
31
Study Simulations
• BE analysis was performed on residual LNG from virtual subjects ( n = 16) for RLD and hypothetical generics with different mean release rate constants (k) varying from 5% to 100% at 6 months, 1, 2, 3, 4 and 5 years using 80% - 125% BE limit.
• Phoenix/R/Berkeley Madonna software were used for the analysis.
32
Hypothetical Generics with Different Release Rate Constants
Rate Constant Change % Value
K RLD 3.89
K + 5 % 5% 4.0845
K + 10 % 10% 4.279
K + 20 % 20% 4.668
K + 40 % 40% 5.446
K + 60 % 60% 6.224
K + 80 % 80% 7.002
K + 100% 100% 7.78
K - 5 % 5% 3.6955
K - 10 % 10% 3.501
K - 20 % 20% 3.112
K - 40 % 40% 2.334
K - 60 % 60% 1.556
K - 80 % 80% 0.778
K - 100% 100% 0
33
Parallel BE Study Results at Different Time Points for Hypothetical Generics with Increase in Release
Observed Residual LNG from Different Formulations and Model Simulated Residual LNG in Virtual Population (n = 16) with
Hypothetical Generic Formulations
35
Parallel BE Study Results at Different Time Points for Hypothetical Generics with Decrease in
Release Rate ConstantChange Minus 0 %
TIME 6 Months 1 Year 2 Year 3 Year 4 Year 5 YearGMR 100 100 100 100 100 100
Lower 98.04 97.69 96.81 95.81 94.78 93.73
Upper 102 102.36 103.3 104.37 105.51 106.69Change Minus 5 %
GMR 100.4 100.7 101.5 102.2 102.9 103.7
Lower 98.4 98.42 98.27 97.99 97.67 97.32
Upper 102.36 103.09 104.75 106.58 108.49 110.47Change Minus 10 %
GMR 100.7 101.5 102.9 104.4 106 107.5
Lower 98.77 99.16 99.75 100.22 100.65 101.05
Upper 102.73 103.82 106.23 108.84 111.56 114.39Change Minus 20 %
GMR 101.5 102.9 106 109.1 112.3 115.6
Lower 99.5 100.64 102.77 104.83 106.86 108.91
Upper 103.46 105.29 109.26 113.51 117.98 122.67Change Minus 40 %
GMR 102.9 106 112.3 119 126.1 133.6
Lower 100.97 103.66 109.07 114.63 120.41 126.44
Upper 104.94 108.32 115.59 123.5 132.03 141.18Change Minus 60 %
GMR 104.4 109.1 119 129.8 141.6 154.4
Lower 102.47 106.77 115.73 125.29 135.56 146.64
Upper 106.45 111.44 122.34 134.45 147.85 162.64Change Minus 80 %
GMR 106 112.3 126.1 141.6 159 178.5
Lower 103.99 109.95 122.74 136.85 152.49 169.86
Upper 107.98 114.67 129.52 146.46 165.72 187.57
Change Minus 100 %
GMR 107.5 115.6 133.6 154.4 178.5 206.3
Lower 105.52 113.22 130.13 149.38 171.38 196.54
Upper 109.54 118 137.17 159.64 185.92 216.59
36
Observed Residual LNG from Tested Formulations and Model Simulated Residual LNG in Virtual Population (n =
16) with Hypothetical Generic Formulations
37
Proposed BE Criteria for Ex-Vivo Study
80
85
90
95
100
105
110
115
120
125
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5 5.5
Co
nfi
de
nce
Inte
rval
Time (Yrs)
Lower
Upper
TIME 6 Months 1 Year 2 Year 3 Year 4 Year 5 Year
Lower 96 95 90 87 84 80
Upper 104 105 111 115 119 125
• Tighter BE Limit Criteria for Residual LNG can be proposed for different time duration (up to 5 Year)
38
Kt = Kr+0.20*Kr
Sample Size BE Limits at 1 year(%) # Passing BE Limits at 5 years(%) # Passing
(20,20) 93 – 100/0.93 998 80-125 687
94 – 100/0.94 954 80-125 687
95 – 100/0.95 651 80-125 687
96 – 100/0.96 164 80-125 687
Kt = Kr-0.20*Kr
Sample Size BE Limits at 1 year(%) # Passing BE Limits at 5 years(%) # Passing
(20,20) 93 – 100/0.93 1000 80-125 859
94 – 100/0.94 992 80-125 859
95 – 100/0.95 804 80-125 859
96 – 100/0.96 219 80-125 859
Potential Selection of BE Limit at One Year
Initial LNG content (A, mean = 52 mg, CV=0.03), Release Rate Constant (k, mean = 3.89 × 10-4 per day, CV = 0.15)
• Purpose of this approach was to find BE limit at one year which will ensure similar passing rate at 5 years, assuming conventional 80% - 125% BE criteria applied to residual LNG at 5 years.
• Then 90% confidence interval of geometric mean ratio of the RLD and TEST at 1 year and 5 years were computed and the procedure was repeated 1000 times simulating 1000 studies.
39
Evaluation of the Proposed (95% – 100/0.95%) BE Criteria at 1 Year
• Two approaches: 1. Modeling and simulation as described above was utilized to create
virtual subjects and parallel BE comparison was conducted at 1 and 5 years. Release rate constants retrieved from relevant submitted applications were utilized.
2. In the second approach observed residual LNG data at one year was retrieved and parallel BE comparison was conducted. Residual LNG data in between 330 to 390 days were considered for one year analysis.
• For both approaches two cases were evaluated and BE analysis showed that criteria were met:– Formulation C vs formulation D
– Formulation D vs Another Drug
40
Summary of Case II
A lower limit of 95.00% and upper limit of 100/0.95% (105.26) seem to be appropriate BE limits at 1 year, which can be used to ensure equivalence of 80% - 125% at 5 years for residual LNG parameter.
41
Overall Summary
• PBPK or PKPD models are critical techniques to inform decision making regarding locally acting drug delivery
• Relevant BE metrics and assessment criteria can be established accordingly
• Modeling based recommendations can be confirmed by observed outcomes