- Email: [email protected]
Pharmacokinetic Modeling to Simulate the Concentration-Time Profiles After Dermal Application of Rivastigmine Patch
Journal of Pharmaceutical Sciences xxx (2016) 1-9
Contents lists available at ScienceDirect
Journal of Pharmaceutical Sciences journal homepage: www.jpharmsci.org
Pharmacokinetics, Pharmacodynamics and Drug Transport and Metabolism
Pharmacokinetic Modeling to Simulate the Concentration-Time Profiles After Dermal Application of Rivastigmine Patch vre 2 Sachiko Nozaki 1, *, Masayuki Yamaguchi 1, Gilbert Lefe 1 2
Drug Metabolism & Pharmacokinetics, Translational Medicine, Novartis Pharma K.K., Tokyo, Japan Clinical PK/PD, Translational Medicine, Novartis Institutes for Biomedical Research, Basel, Switzerland
a r t i c l e i n f o
a b s t r a c t
Article history: Received 2 March 2016 Revised 9 April 2016 Accepted 11 April 2016
Rivastigmine is an inhibitor of acetylcholinesterases and butyrylcholinesterases for symptomatic treatment of Alzheimer disease and is available as oral and transdermal patch formulations. A dermal absorption pharmacokinetic (PK) model was developed to simulate the plasma concentration-time profile of rivastigmine to answer questions relative to the efficacy and safety risks after misuse of the patch (e.g., longer application than 24 h, multiple patches applied at the same time, and so forth). The model comprised 2 compartments which was a combination of mechanistic dermal absorption model and a basic 1-compartment model. The initial values for the model were determined based on the physicochemical characteristics of rivastigmine and PK parameters after intravenous administration. The model was fitted to the clinical PK profiles after single application of rivastigmine patch to obtain model parameters. The final model was validated by confirming that the simulated concentration-time curves and PK parameters (Cmax and area under the drug plasma concentrationtime curve) conformed to the observed values and then was used to simulate the PK profiles of rivastigmine. This work demonstrated that the mechanistic dermal PK model fitted the clinical data well and was able to simulate the PK profile after patch misuse. © 2016 American Pharmacists Association®. Published by Elsevier Inc. All rights reserved.
Keywords: physiologically based pharmacokinetic modeling simulations transdermal absorption clinical pharmacokinetics
Introduction Rivastigmine is a slowly reversible (pseudo-irreversible), centrally selective dual inhibitor of acetylcholinesterase and butyrylcholinesterase, which increases the available acetylcholine levels and improves neurotransmission. It has established efficacy in the symptomatic treatment of Alzheimer disease1-3 and Parkinson disease dementia (PDD)4 and was shown to improve activities of daily living, cognition, behavior, and global function.2,5-8 Studies of doseeresponse relationships for cholinesterase inhibitors support greater enzyme inhibition, in turn leading to higher efficacy and long-term benefits with higher drug doses.9 Rivastigmine has been developed for oral twice daily administration as capsule (3, 6, 9, and 12 mg/d) and as solution (2 mg/mL) and
Conflicts of interest: All authors are full-time employees of Novartis, except Masayuki Yamaguchi who was employed with Novartis during the conduct of the research. Currently, Masayuki Yamaguchi is employed by Bristol-Myers Squibb K.K. This article contains supplementary material available from the authors by request or via the Internet at http://dx.doi.org/10.1016/j.xphs.2016.04.011. * Correspondence to: Sachiko Nozaki (Telephone: þ81 3 6899 8545; Fax: þ81 3 6257 3613). E-mail address: [email protected] (S. Nozaki).
for transdermal daily administration as patch (5 cm2 [4.6 mg/24 h], 10 cm2 [9.5 mg/24 h], 15 cm2 [13.3 mg/24 h] and 20 cm2 [17.4 mg/24 h]; of note: the 20-cm2 patch has not been launched to market to date). Transdermal patch is an optimal way to deliver rivastigmine and provides many benefits over conventional oral treatments, allowing patients easier access to optimal therapeutic doses.10 Treatment with rivastigmine patch is initiated with the 5-cm2 (4.6 mg/24 h) patch, which when well tolerated, is uptitrated to the 10-cm2 (9.5 mg/24 h) patch and beyond based on individual responses to obtain the desired therapeutic benefits.11 In Japan, more gradual titration approach starting from 2.5 cm2 and followed with 5 cm2, 7.5 cm2, and 10 cm2 is also available.12 All patch sizes have the same loading dose per area (1.8 mg/cm2), and the amount of rivastigmine delivered from a patch over a 24-h wearing period is approximately 50% of the total loading dose. After launch, there were some case reports of patch misuse, which raised questions related to efficacy and safety risks (e.g., longer application than 24 h, multiple patches applied at the same time, and so forth). Previously, modeling has been performed during the development of rivastigmine to describe PK after applications of rivastigmine patch.10,13 Such noncompartmental or compartmental models, however, were unable to answer these questions, because
http://dx.doi.org/10.1016/j.xphs.2016.04.011 0022-3549/© 2016 American Pharmacists Association®. Published by Elsevier Inc. All rights reserved.
2
S. Nozaki et al. / Journal of Pharmaceutical Sciences xxx (2016) 1-9
Figure 1. Compartment model to describe the PK profile of rivastigmine after rivastigmine patch application. A1, dose point that represents the applied patch in this model case; Cskin, drug concentration in the skin; V2, distribution volume in the skin; Ccentral, drug concentration in the central compartment; V3, distribution volume in the central compartment; A0, elimination compartment.
mostly oral administration models were applied, and the models did not quantitatively reflect the amount of drug delivered per the application time, which will be the basis of quantitative simulation of the systemic exposure. In general terms, pharmacokinetic (PK) models of dermal absorption of chemicals have been created and reported in a series of publications.14-18 Skin is usually represented either as a single compartment or by 2 compartments separately distinguishing the lipophilic and hydrophilic layers of the skin.17 Such modeling attempts were generally used in the toxicology rather than clinical pharmacology fields until recently when Polak et al.18 reported a mechanistic dermal absorption model using Simcyp simulator. Still, there is limited information on dermal absorption models that were put into practical use for simulations of drug concentrations in the clinical setting. In this study, we report a PK model created by integrating a mechanistic dermal absorption model with a systemic PK model to describe the PK after applications of rivastigmine patch for the purpose of simulating misuse scenarios. The model was created on a Phoenix WinNonlin platform that was selected as an optimal tool to fit the model to clinical data for refinement and verification, after which the model was used for the simulations. Methods Compartmental Structure of the Model A 2-compartment model was developed in WinNonlin, version 5.2, using a user-defined American Standard Code for Information Interchange model to describe the PK profile of rivastigmine after single 24-h dermal application of rivastigmine patch (initial model,
Appendix 1). The schematic diagram of the compartmental structure of the model is presented in Figure 1. The model comprised 2 compartments, skin and central compartment, as described by Brown and Hattis15 in which transfers of rivastigmine between compartments are described by rate constants K12 (for dermal absorption), K23 (skin to blood transfer), and K30 (elimination from systemic compartment; Fig. 1). The parameters for volume of skin and central compartment were defined as V2 and V3, respectively. A PK simulation was conducted to draw a PK profile of rivastigmine after 10-cm2 patch application using the initial parameters described in the following sections (bottom-up approach). Initial Values: 1. Dermal Absorption Initial parameter for the dermal absorption rate constant (K12) was estimated assuming a first-order absorption. First, the initial value was calculated using the data of amount of rivastigmine released from the patch after a 24-h application. An average of 9.4 mg (range, 7.3-11.8 mg) of the total drug content of 18 mg was released from 10-cm2 patch during 24-h application (n ¼ 19) in a previous clinical study.19 Calculation using the equation [K12 ¼ ln(released amount/drug content)/24 h] resulted in a mean K12 value of 0.027 h1 (range, 0.018-0.037 h1). The initial value was also calculated from the time course of in vitro human skin permeation data (0.026 h1; data on file). Because both methods gave similar values, 0.026 h1 was used as the initial value for K12. Initial Values: 2. Skin-to-Blood Transfer The rate constant for transfer of the drug from skin to blood (K23) was estimated as follows. The skin and blood are in
Table 1 Initial and Final Parameters Used in the Model Parameters (Unit)
Initial Values
Mean Values of the Estimated Individual Parametersa
Parameters Used in the Final Model
K12 (h1) K23 (h1) K30 (h1)
0.026b 0.3c 0.693d
0.027 0.473 0.521
V2 (mL) V3 (mL)
4e 94,800g
4 141,634
0.027 0.473 K30 ¼ [1/V3] [Vmax/(Km þ Ccentral)], where V3 ¼ 141,634, Km ¼ 24, Vmax ¼ 1,760,000 Note: K30 ¼ 0.497 when Ccentral << Km 4f 141,634
a b c d e f g
Individual parameters are presented in Table 2. Initial value estimated assuming first-order absorption based on (1) released rivastigmine from patch after 24-h application and (2) human skin permeation test data. Parameter for transfer from skin to blood estimated for rivastigmine. Elimination constant determined from T1/2 following intravenous administration of rivastigmine. Volume of skin compartment z patch area (10 cm2) skin thickness (0.4 cm) ¼ 4 mL. V2 was changed corresponding to the size of the patch; 2, 4, 6, and 8 mL for patches 5, 10, 15, and 20 cm2, respectively. See details in Methods section Initial Values. Volume of central compartment (distribution volume) determined from volume of distribution following intravenous administration of rivastigmine.
S. Nozaki et al. / Journal of Pharmaceutical Sciences xxx (2016) 1-9 Table 2 Individual Model Parameter Estimates for Pharmacokinetic (PK) Profiles After Single Application of 10-cm2 Patch Subject
K12 /h
K23 /h
K30 /h
V3 mL
CL ¼ K30 V3 mL/h
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 Min Max Mean Standard deviation CV% Variance of intersubject variability (u2)a Initial value Lower limit Upper limit
0.018 0.035 0.040 0.016 0.040 0.037 0.035 0.016 0.019 0.016 0.016 0.016 0.033 0.033 0.026 0.040 0.036 0.026 0.016 0.016 0.040 0.027 0.010
0.377 0.297 0.471 0.365 0.533 0.304 0.249 0.270 1.425 0.287 0.298 0.374 0.353 0.442 0.329 0.289 0.302 1.732 0.280 0.249 1.732 0.473 0.400
0.400 0.303 0.466 0.366 0.527 0.363 1.200 0.321 0.283 0.283 0.295 0.380 0.349 0.521 1.200 0.700 0.815 0.389 0.744 0.283 1.200 0.521 0.286
137,915 193,272 183,716 183,265 131,336 199,218 91,432 199,771 101,735 136,486 190,454 175,175 103,072 186,197 58,895 116,273 105,547 143,605 53,690 53,690 199,771 141,634 47,513
55,183 58,465 85,559 67,109 69,256 72,330 109,681 64,087 28,798 38,633 56,249 66,519 35,978 97,078 70,662 81,366 85,973 55,824 39,951 28,798 109,681 65,195 21,137
37.0 0.137
84.6 0.715
54.9 0.301
33.5 0.113
32.4 0.105
0.026 0.0160 0.0400
0.300 0.030 3.000
0.693 0.100 1.200
94,800 50,000 200,000
d d d
0:98 þ 0:02 Kow 0:993 þ 0:007 Kow
mL 4g 1:5 min100g Skin blood flow ¼ Dermal volume Ksb 4 mL 3
(3)
Initial Values: 3. Systemic Distribution and Elimination The elimination constant K30 was calculated based on the elimination half-life (T1/2) after intravenous administration of rivastigmine (~1 h).
K30 ¼ 0:693=ðT1=2Þ ¼ 0:693=1 ¼ 0:693:21
(4)
Volume of central compartment V3 was determined from distribution volume after intravenous administration of rivastigmine (1.5 L/kg)22 and a typical body weight of human population (63.2 kg).19
V3 ¼ 1:5 L=kg 63:2 kg ¼ 94:8 L:
(5)
Hence, 94.8 L was used as the initial value for V3. Fitting to the Clinically Observed Pharmacokinetic Profile, Model Verification, and Simulations The initial model was fitted to the actual PK profiles after single 24-h application of rivastigmine patch of 10 cm2 (n ¼ 19).19 Uniform weighting was used. Individual model parameters and descriptive statistics such as mean, standard deviation, and coefficient of variation (CV%), were derived (top-down approach). The mean values of the obtained individual parameters values for K12, K23, V2, and V3 (Table 1, middle column) were used for the upcoming final model. The final model was developed in a user-defined Phoenix Modeling Language (PML) using the calculated mean values and the CV% for model parameters K12, K23, K30, V2, and V3 as typical values (fixed effects, q), and random effects (u), respectively (final model, Appendix 2). The modeling and simulations were performed in Phoenix WinNonlin, version 6.1 or higher. To describe the overdose proportional exposure of rivastigmine,10 the elimination constant K30 was modified by assuming MichaeliseMenten kineticsebased saturation. The K30 expressed by the function of rivastigmine (Ccentral) is presented in the following equation.
(1)
where Ksb is a skin/blood partition coefficient, which can be expressed by the following equation:
Ksb ¼
K23 ¼
Hence, 0.3 h1 was used as the initial value for K23.
equilibrium with each other. Brown and Hattis15 proposed the following equation to express the flow rate.
Skin blood flow Dermal volume Ksb
K23 was calculated by applying the values stated above into the Equation 1.
¼ 0:005 min1 ¼ 0:3 hr1
Italic: estimated values that reached the lower or upper limit. The lower and upper limits of each parameter for simulation were determined based on the clinically obtained range when available (K12: range of the absorption rate constant calculated from the remaining amount of rivastigmine in the patch, K30: range of elimination rate constant after intravenous administration, V3: range of volume of distribution after intravenous administration), and 0.1 initial for lower and 10 initial for upper for others (K23 and V2). V2 was not presented because the value remained at 4 mL (initial value) in all individuals. The upper and lower limits for fitting were 0.40 and 40.0, respectively. a Calculated from the equation [(CV%)/100]2.
Skin-to-blood rate ð=minÞ ¼
3
(2)
The parameter Kow represents the distribution coefficient at 37 C in n-octanol/water. Considering that the Kow of rivastigmine is >100 (physical and chemical properties of rivastigmine, data on file), the Ksb can take values from 1.8 (when Kow ¼ 100) to 3 (when Kow ~ ∞), which are within close range. Ksb ¼ 3 was used for the calculation of initial value for K23. Skin blood flow was reported to be 1.5 mL/min/100 g.20 Dermal volume and weight of skin can be estimated to be 4 mL and 4 g, respectively, based on an assumption that the volume of skin compartment can be expressed as the multiplication product of patch area (10 cm2) and skin thickness (0.1-0.4 cm).15 Therefore,
K30 ¼
1 Vmax V3 Km þ Ccentral
(6)
where Ccentral ¼ Acentral V 3 , Acentral ¼ drug amount in the central compartment. The Km value that provided good estimation of the overdose proportional PK was searched manually by testing several values. In the process, the elimination constant K30 was fixed to 0.5 h1 so as to not contradict with the clinical T1/2 (the elimination constant 0.5 h1 corresponds to the T1/2 ¼ 1.4 h). When the Km was 24 ng/mL and the Vmax was 1,760,000 mL, the simulated PK parameters accurately captured the overdose proportional increase in AUCtau and Cmax. These values were used for the final model. The parameter for volume of skin (V2) was changed corresponding to the size of the patch (2, 4, 6, and 8 mL for patches of 5, 10, 15, and 20 cm2, respectively). The parameters used in the final model are presented in Table 1 (right column). The PML code for the final model is presented in Appendix 2.
4
S. Nozaki et al. / Journal of Pharmaceutical Sciences xxx (2016) 1-9
Figure 2. Model fitted to the individual PK profiles after single 24-h application of 10-cm2 patch. The plasma rivastigmine concentrations were measured from time 0 (patch application) to 48 h. Patches were removed at 24 h. Open circle: observed concentrations, solid line: predicted concentrations. The observed concentrations are the PK profiles after single application of rivastigmine patch of 10 cm2 in a previous clinical study.19
S. Nozaki et al. / Journal of Pharmaceutical Sciences xxx (2016) 1-9
5
Simulated Pharmacokinetic Profiles After Normal Applications (Model Verification)
Final Model
dAcentral ¼ Acentral K30 þ Askin K23 dt dA0 ¼ Acentral K30 dt n X dAskin ¼ Askin K23 þ Dosei K12 dt i¼1
dDosei ¼ Dosei K12 dt A Ccentral ¼ central V3 1 Vmax K30 ¼ V3 Km þ Ccentral The parameter Askin represents the drug amount in the skin. The parameter Dosei represents the loaded (time ¼ 0), or remaining (time > 0), amount of the drug in the patch of the i time(s) of the application. Dosei turns into zero at the time of patch removal. n times of application was assumed. The final model was verified by stochastic simulation of steadystate PK profiles following 5, 10, 15, and 20 cm2 patch applications using Phoenix Monte Carlo population PK simulation run mode and by comparing them to clinically observed PK profiles. The clinical data used for the verification10 were from different subjects that were used for the initial fitting.19 Variance of each parameter (CV%) obtained in the initial fitting was used for the simulation. The u2 values were 0.137, 0.715, 0.301, and 0.113 for K12, K23, K30, and V3, respectively, as presented in Table 2. After sampling 1000 replicated data sets, 95% confidence intervals (CIs) for the logs of the simulated data sets were computed and presented graphically with the observed PK profiles. Mean plasma rivastigmine concentration-time profiles after normal and misuse applications were simulated using the final model with Phoenix individual simulations. PK parameters at steady state (Cmax and AUCtau) were calculated for the simulated concentration-time profiles using Phoenix WinNonlin NCA object. Results
The plasma rivastigmine concentration-time profiles after applications of 5-, 10-, 15-, and 20-cm2 patches were simulated and are presented in Figure 3. The comparison between simulated PK profiles/parameters and those of the observed concentrations in a previously conducted clinical study10 is presented in Figure 4 (profiles), Figure 5, and Table 3 (parameters). Overall, the simulated mean concentrations were close to the mean of actual individual concentrations. The actual individual concentrations (n ¼ 13-23) were almost covered by the upper and lower 95% CIs for the logs of the simulated data sets (Fig. 4). Predicted Cmax and AUCtau were close to the observed values, except a tendency of overestimation in AUCtau (Table 3 and Fig. 5).
Simulated Pharmacokinetic Profile After Misuse Applications The simulated concentration-time profiles based on a series of misuse scenarios are presented in Figure 6. Simulations were conducted under steady state conditions for scenarios A, B, and C, more specifically on third day of repeated administrations or after.10 (a) A new 10-cm2 patch was applied while the previous one was kept for an additional 0.5, 1, 2, 6, 10, 18, or 24 h: When the previous patch was kept for an additional 2 h, there was a limited change in the concentration-time profiles. When the previous patch was kept for additional 6 h, the peak concentrations were significantly elevated, close to the concentration levels reached with a 15-cm2 patch. (b) Simulated plasma rivastigmine concentration-time profiles of 10 days application with new patches applied every day without removing previous ones: From time 0 to 48 h (days 2, 1, and 1), 1 patch applied per day (normal applications). From time 48 to 288 h (days 2 to 10), each daily patch was not removed once applied, meaning that 1, 2, 3, …, 10 patches were applied at the same time on days 1, 2, 3, …, 10, respectively, mimicking off-label use. All the patches were removed at time 288 h (day 10). The simulations showed that such usage of 10-cm2 patch resulted in elevated peak concentrations at steady state comparable to the levels
“Bottom-Up” ApproachdSimulated Plasma Rivastigmine Concentration Using Initial Parameters PK profile of rivastigmine after 10-cm2 patch application was simulated using the initial parameters described in the Methods section Initial Values: 1-3. The simulated plasma rivastigmine concentration-time curve was close to the actual concentrations, indicating that the initial parameters were good enough to start the fitting from (data not presented). “Top-Down” ApproachdFitting the Model to Clinical Data The model was then fitted to the actual PK profiles after single application of rivastigmine patch of 10 cm2 in a previous clinical study.19 Individual parameters are presented in Table 2, and the individual PK profiles are shown in Figure 2. The model fitted well, not only to the individual concentration-time profiles during patch application (0-24 h), but also to those after patch removal (24-48 h). Mean values of the individual parameters for K12, K23, V2, and V3 were similar to the respective initial values, reconfirming that the prediction by “bottom-up” approach was good.
Figure 3. Simulated PK profiles after multiple applications. Black lines: simulated concentration-time profiles, lower to upper: 5, 10, 15, and 20 cm2. For the simulations, mean values of each model parameters were used without considering the variability.
6
S. Nozaki et al. / Journal of Pharmaceutical Sciences xxx (2016) 1-9
Figure 4. Simulated PK profiles overlaid with the observed individual PK profiles at steady state for patches of 20 cm2 (a), 10 cm2 (b), and 5 cm2 (c). Red line: simulated mean concentration-time profiles. Dotted red line: lower/upper limits of 95% CIs for the logs of the generated 1000 sets of concentration-time profiles by Monte Carlo simulation. Black line: mean of the observed individual concentration-time profiles. Gray lines: observed individual concentration-time profiles. (a) n ¼ 13, (b) n ¼ 22, and (c) n ¼ 23. Left: linear scale and right: log scale.
S. Nozaki et al. / Journal of Pharmaceutical Sciences xxx (2016) 1-9
7
(d) Patches of 15 cm2 changed every 1 h instead of every 24 h: The simulated concentration reached steady state around 20 h. The concentrations were similar to those reached with 20-cm2 patch under normal application conditions. Discussion
Figure 5. Simulated and observed PK parameters (a) Cmax: Cmax at steady state, (b) AUC: AUCtau at steady state. Closed square: observed parameters. Open diamond: simulated parameters.
reached with the 20-cm2 patch. Exposure was ~2.6 times higher relative to normal use in the case of 10 cm2 and ~3.8 times higher in the case of 15 cm2, exceeding the levels reached with 20 cm2. (c) Simulated plasma rivastigmine concentration-time profiles after 1, 2, 3, 4, or 5 patches of 10 cm2 applied at the same time and remained on the skin: The peak concentrations in these misuse case scenarios were 1-, 2.4-, 4.7-, 8.0-, and 12.7-fold those obtained under normal (label) application conditions when 1, 2, 3, 4, or 5 patches were applied at the same time, respectively. The total exposure (AUC calculated from 48 h to infinity) increased 1-, 2.3-, 4.3-, 7.5-, and 12.7-fold, respectively.
In this study, a model was developed to describe the PK profiles of rivastigmine patch and to simulate its PK under misuse situations (e.g., multiple patches applied at the same time, patches replaced more frequently than as per label, and so forth). This model is the first to describe the PK of rivastigmine using dermal absorption model. In addition, to our knowledge, this is the first case example that leveraged a dermal absorption model into practical use for simulations of systemic drug concentrations. Previous models to describe the PK profiles of rivastigmine patch10,13,19 were noncompartment or 1-compartment models with first-order administration and elimination process. Such models were unable to simulate the PK after patch misuse, on account of inaccurate mass balance in drug amount delivered from the patch. The amount of drug available in the body as a function of time was modeled as A(t) ¼ Dose(t0) exp(ka t). The population fixed estimate for the ka parameter obtained by fitting the 1-compartment model to the clinical data was 0.0667 h1 (data on file). Therefore, the amount of dose remaining into the patch after 24 h according to the 1-compartment model is Dose(t0) exp(0.0667 24) ¼ Dose(t0) 0.201. This quantity represents approximately 20% of the initial amount of drug available in the patch. This figure differs significantly from the measured value of approximately 50% of the drug load remaining into the patch after the 24-h wearing period as reported in the original article.10 Although the 1-compartment model could adequately serve its initial purpose of describing the individual PK profiles and could be used for subsequent PK modeling, that model was not flexible enough to answer other questions. In addition, the model could not be used to support (for instance, by running simulations) further statements on the impact of noncompliance to treatment on individual PK profiles. The newly created model in this work was based on a mechanistic dermal absorption model assuming a barrier from patch to the skin. Mechanistic dermal absorption models have been reported mostly in a toxicology area.17 Generally speaking, skin compartment has been expressed as 1- or 2-compartments depending on the model's purpose.17 Dividing the skin part into 2 compartments is useful when considering the skin as a storage compartment, especially when the purpose of the model is to estimate the toxicity of the compound in the skin.16 However, 1 compartment was used in this study for the skin part because it was considered adequate to serve the purpose of this modeling exercise, which was to describe the systemic concentration of the drug. In addition, limiting the number of compartments is recommended to avoid overparameterization when fitting to clinical PK. A rate constant expression by Brown and Hattis15 was used to describe the transfer from skin to blood (Eq. 1). This equation is one of the several similar models based on a common assumption that
Table 3 Simulated and Observed Pharmacokinetic (PK) Parameters Dose (cm2)
5 10 15 20
Cmax at Steady State (ng/mL)
AUCtau at Steady State (ng h/mL)
Simulated
Observed
Ratio (Simulated/Observed)
Simulated
Observed
Ratio (Simulated/Observed)
3.17 7.18 12.3 19.1
2.71 7.88 14.1 19.5
1.17 0.91 0.87 0.98
65.3 147 254 397
46.3 127 233 345
1.41 1.16 1.09 1.15
8
S. Nozaki et al. / Journal of Pharmaceutical Sciences xxx (2016) 1-9
Figure 6. Simulated concentration-time profiles following various scenarios of rivastigmine patch applications. (a) A new 10-cm2 patch (fourth patch) was applied while the previous one (third patch) was kept for additional 0.5, 1, 2, 6, 10, 18, or 24 hr (black line) vs normal applications (red-dotted lines, 10 cm2, 15 cm2, 20 cm2 from lower to upper). (b) Simulated plasma rivastigmine concentration-time profiles of 10-day application with new patches applied every day without removing previous ones, starting from fourth application (black solid line: 10 cm2, dotted line: 15 cm2) vs normal applications (red-dotted lines, 10 cm2, 15 cm2, 20 cm2 from lower to upper). (c) Simulated plasma rivastigmine concentration-time profiles after 1, 2, 3, 4, or 5 patches of 10 cm2 applied at the same time and remained on the skin (black lines) vs normal applications (red-dotted lines, 10 cm2, 15 cm2, 20 cm2 from lower to upper). The misuse was assumed to happen on the third day. (d) Patches of 15-cm2 changed every 1 h instead of every 24 h (black solid line), Normal applications after 15- and 20-cm2 patch applications from lower to upper (red-dotted lines, for reference). For the simulations, mean values of each model parameters were used without considering the variability.
Figure 7. Impact of varying K23 values on the simulated concentration-time profiles. Plasma rivastigmine concentrations after single 24-h application of 10-cm2 patch were simulated using the final model. K23 values were varied while all the other parameters remained constant. Lines represent the results for K23 ¼ 0.249 (blue), 0.473 (black), and 1.732 (red), respectively.
clearance from the skin is controlled by the solubility of the compound in the blood relative to that in the skin and the skin blood flow rate and area, as reviewed by McCarley and Bunge.17 This equation was selected because it was based solely on physicochemical or physiological parameters and did not require parameters that need to be determined experimentally. In addition, models using this type of equation have successfully predicted clinical systemic PK profiles of several compounds.18 The simulated PK profile by “bottom-up” approach (simulation based on initial values, Table 1 left column) was close to the clinically observed PK profiles, suggesting that the model structure was valid and conforming to the clinical situations. When the model was fitted to the clinically observed data, the model was well applicable and allowed to describe most of the individual PK profiles. In some subjects, however, predicted curve did not fully capture the steep peak concentrations (e.g., subjects 1, 6, and 8). Reason behind those sharp peaks is unknown, but one possibility could be that there was a transient increase in skin absorption rate during application (e.g., due to temporal heating), resulting in temporal rise in K12. Such profile cannot be described by this model where K12 is constant over time. Although this might be a
S. Nozaki et al. / Journal of Pharmaceutical Sciences xxx (2016) 1-9
limitation of this model, the overall fit was good and considered adequate to serve its initial purpose of describing the typical PK profiles. Estimated mean parameters as a result of the fitting were similar to the respective initial values, reconfirming that the prediction by “bottom-up” approach was correct. It was notable that the parameter for volume of skin (V2) did not vary at all (i.e., remained as the initial value; 4 mL). This means that the volume of skin was not a sensitive parameter to determine systemic exposure. Some of the estimated values for K12 and K30 reached the lower or upper limit (8 and 2 subjects of 19 for each). The lower and upper limits of these parameters were determined based on ranges of clinical observations, which acted as a gatekeeper preventing the estimated parameters from being clinically irrelevant. Other parameters (K23 and V3) varied within the range without reaching the limits. There was a large variability associated with the K23 estimate (range, 0.249-1.732/h, CV ¼ 84.6%). To assess the impact of K23 to the results, we have conducted a sensitivity analysis by varying K23 values (0.249, 0.473, and 1.732) while all the other parameters remained constant (Fig. 7). The result showed that the K23 values have a clear effect on the shape of the PK profile before and after removal of the patch. This suggested that K23 was mathematically useful to describe the varying shape of individual PK profile. From physiological point of view, variation in K23 values may be attributable to individual differences in fat content and/or blood flow rate in the skin, although this could not be verified. Overall, these facts indicated that the model fit was adequate. Because rivastigmine shows overdose proportional exposure,10 the model was modified accordingly to adequately describe the PK across the patch doses of 5, 10, 15, and 20 cm2. Considering that rivastigmine is not metabolized in the skin (2% metabolized after 72 h, data on file) and the major route of elimination is hydrolysis by esterase, a saturation factor was incorporated into the elimination constant K30. Predicted Cmax and AUC conformed well to the observed values, although AUC showed a tendency of ~15% overestimation (Fig. 5). However, a slight overestimation in total exposure was considered acceptable (and preferable to an underestimation) from a safety risk assessment perspective. In addition, the simulated time-concentration profile was within the range of variation when compared with observed individual values (Fig. 4). Furthermore, the model was verified by simulating 1000 replications by Monte Carlo simulations that resulted in the 95% CIs containing most subjects' concentration-time profiles (n ~ 20). In other words, the model and the variance parameter (u) were adequate to simulate PK profiles for various scenarios with good estimation of 95% CIs. It is of note that simulation of mean profiles was sufficient to answer most questions relative to PK differences between normal application and misuse conditions. This work demonstrated the utility of the PK modeling in simulating PK under various situations. Prospective predictions of PK after off-label misuse are useful to answer frequently asked questions in the case of patch or tape formulations as unintended removal or application could happen. The simulations demonstrated that the model was flexible enough to address these questions. We could compare simulated PK profiles relative to the typical normal profiles of marketed patch as a tool to assess the safety and efficacy implications of the off-label use (Fig. 6). Although the clinical implication of the simulation results is out of scope of this article, most of the scenarios tested (Fig. 6) resulted in a quick increase in exposures reaching PK levels comparable to those measured at the next upper dose level. This indicates that these off-label/misuses of Exelon Patch must be avoided in clinical practice, especially when the simulated concentrations exceed the typical exposures observed with the 20-cm2 patch, which is the highest ever-tested patch dose in clinical, and would require a safety alert.
9
It should be emphasized that the users of the model need to be fully aware of the model assumptions and limitations to understand the differences in outcomes from other models. Limitations of the model include potential overestimation of the exposures at higher levels than those ever observed in clinical setting (i.e., Cmax ~ 100 ng/mL). In an actual situation, another elimination route may contribute more if the original route or enzyme is saturated and possibly compensate for the decreased clearance. Hence, extrapolation to higher concentrations should be treated with caution. Limitations of the model also include that the model may not be appropriate to simulate the PK when system parameters in the elimination process could be altered, for example, special populations, because the model does not include mechanistic parameters in the elimination process and cannot accommodate for situations when these processes are altered. References 1. Farlow M, Anand R, Messina Jr J, Hartman R, Veach J. A 52-week study of the efficacy of rivastigmine in patients with mild to moderately severe Alzheimer's disease. Eur Neurol. 2000;44(4):236-241. 2. Rosler M, Anand R, Cicin-Sain A, et al. Efficacy and safety of rivastigmine in patients with Alzheimer's disease: international randomised controlled trial. BMJ. 1999;318(7184):633-638. 3. Karaman Y, Erdogan F, Koseoglu E, Turan T, Ersoy AO. A 12-month study of the efficacy of rivastigmine in patients with advanced moderate Alzheimer's disease. Dement Geriatr Cogn Disord. 2005;19(1):51-56. 4. Emre M, Aarsland D, Albanese A, et al. Rivastigmine for dementia associated with Parkinson's disease. N Engl J Med. 2004;351(24):2509-2518. 5. Onor ML, Trevisiol M, Aguglia E. Rivastigmine in the treatment of Alzheimer's disease: an update. Clin Interv Aging. 2007;2(1):17-32. 6. Finkel SI. Effects of rivastigmine on behavioral and psychological symptoms of dementia in Alzheimer's disease. Clin Ther. 2004;26(7):980-990. 7. Farlow MR, Cummings JL, Olin JT, Meng X. Effects of oral rivastigmine on cognitive domains in mild-to-moderate Alzheimer's disease. Am J Alzheimers Dis Other Demen. 2010;25(4):347-352. 8. Burns A, Spiegel R, Quarg P. Efficacy of rivastigmine in subjects with moderately severe Alzheimer's disease. Int J Geriatr Psychiatry. 2004;19(3):243-249. 9. Cutler NR, Polinsky RJ, Sramek JJ, et al. Dose-dependent CSF acetylcholinesterase inhibition by SDZ ENA 713 in Alzheimer's disease. Acta Neurol Scand. 1998;97(4):244-250. vre G, Sedek G, Jhee SS, et al. Pharmacokinetics and pharmacodynamics of the 10. Lefe novel daily rivastigmine transdermal patch compared with twice-daily capsules in Alzheimer's disease patients. Clin Pharmacol Ther. 2008;83(1):106-114. 11. Exelon Patch® US Prescribing Information 2015. Available at: http://www. pharma.us.novartis.com/product/pi/pdf/exelonpatch.pdf. Accessed May 11, 2016. 12. Exelon Patch product information (Package Insert). Available at: http://www. info.pmda.go.jp/downfiles/ph/PDF/300242_1190700S1029_1_07.pdf Accessed April 7, 2016. vre G, Huang HL, Schmidli H, Amzal B, Appel-Dingemanse S. 13. Mercier F, Lefe Rivastigmine exposure provided by a transdermal patch versus capsules. Curr Med Res Opin. 2007;23(12):3199-3204. 14. Reddy MB, McCarley KD, Bunge AL. Physiologically relevant one-compartment pharmacokinetic models for skin. 2. Comparison of models when combined with a systemic pharmacokinetic model. J Pharm Sci. 1998;87(4):482-490. 15. Brown HS, Hattis D. The role of skin absorption as a route of exposure to volatile organic-compounds in household tap waterda simulated kinetic approach. J Am Coll Toxicol. 1989;8(5):839-851. 16. Shatkin JA, Brown HS. Pharmacokinetics of the dermal route of exposure to volatile organic chemicals in water: a computer simulation model. Environ Res. 1991;56(1):90-108. 17. McCarley KD, Bunge AL. Pharmacokinetic models of dermal absorption. J Pharm Sci. 2001;90(11):1699-1719. 18. Polak S, Ghobadi C, Mishra H, et al. Prediction of concentration-time profile and its inter-individual variability following the dermal drug absorption. J Pharm Sci. 2012;101(7):2584-2595. vre G, Buche M, Sedek G, et al. Similar rivastigmine pharmacokinetics and 19. Lefe pharmacodynamics in Japanese and white healthy participants following the application of novel rivastigmine patch. J Clin Pharmacol. 2009;49(4):430-443. 20. Kai K, Inoue S, Higaki Y, Tomokuni K. Effects of a cool environment on the health of female office workers and students. J Physiol Anthropol. 2008;27(3): 153-159. 21. Hossain M, Jhee SS, Shiovitz T, et al. Estimation of the absolute bioavailability of rivastigmine in patients with mild to moderate dementia of the Alzheimer's type. Clin Pharm. 2002;41(3):225-234. 22. Exelon Patch product information (Interview Form) Available at: http://www. info.pmda.go.jp/go/interview/1/300242_1190700S1029_1_EXP_1F Accessed April 7, 2016.
Contents lists available at ScienceDirect
Journal of Pharmaceutical Sciences journal homepage: www.jpharmsci.org
Pharmacokinetics, Pharmacodynamics and Drug Transport and Metabolism
Pharmacokinetic Modeling to Simulate the Concentration-Time Profiles After Dermal Application of Rivastigmine Patch vre 2 Sachiko Nozaki 1, *, Masayuki Yamaguchi 1, Gilbert Lefe 1 2
Drug Metabolism & Pharmacokinetics, Translational Medicine, Novartis Pharma K.K., Tokyo, Japan Clinical PK/PD, Translational Medicine, Novartis Institutes for Biomedical Research, Basel, Switzerland
a r t i c l e i n f o
a b s t r a c t
Article history: Received 2 March 2016 Revised 9 April 2016 Accepted 11 April 2016
Rivastigmine is an inhibitor of acetylcholinesterases and butyrylcholinesterases for symptomatic treatment of Alzheimer disease and is available as oral and transdermal patch formulations. A dermal absorption pharmacokinetic (PK) model was developed to simulate the plasma concentration-time profile of rivastigmine to answer questions relative to the efficacy and safety risks after misuse of the patch (e.g., longer application than 24 h, multiple patches applied at the same time, and so forth). The model comprised 2 compartments which was a combination of mechanistic dermal absorption model and a basic 1-compartment model. The initial values for the model were determined based on the physicochemical characteristics of rivastigmine and PK parameters after intravenous administration. The model was fitted to the clinical PK profiles after single application of rivastigmine patch to obtain model parameters. The final model was validated by confirming that the simulated concentration-time curves and PK parameters (Cmax and area under the drug plasma concentrationtime curve) conformed to the observed values and then was used to simulate the PK profiles of rivastigmine. This work demonstrated that the mechanistic dermal PK model fitted the clinical data well and was able to simulate the PK profile after patch misuse. © 2016 American Pharmacists Association®. Published by Elsevier Inc. All rights reserved.
Keywords: physiologically based pharmacokinetic modeling simulations transdermal absorption clinical pharmacokinetics
Introduction Rivastigmine is a slowly reversible (pseudo-irreversible), centrally selective dual inhibitor of acetylcholinesterase and butyrylcholinesterase, which increases the available acetylcholine levels and improves neurotransmission. It has established efficacy in the symptomatic treatment of Alzheimer disease1-3 and Parkinson disease dementia (PDD)4 and was shown to improve activities of daily living, cognition, behavior, and global function.2,5-8 Studies of doseeresponse relationships for cholinesterase inhibitors support greater enzyme inhibition, in turn leading to higher efficacy and long-term benefits with higher drug doses.9 Rivastigmine has been developed for oral twice daily administration as capsule (3, 6, 9, and 12 mg/d) and as solution (2 mg/mL) and
Conflicts of interest: All authors are full-time employees of Novartis, except Masayuki Yamaguchi who was employed with Novartis during the conduct of the research. Currently, Masayuki Yamaguchi is employed by Bristol-Myers Squibb K.K. This article contains supplementary material available from the authors by request or via the Internet at http://dx.doi.org/10.1016/j.xphs.2016.04.011. * Correspondence to: Sachiko Nozaki (Telephone: þ81 3 6899 8545; Fax: þ81 3 6257 3613). E-mail address: [email protected] (S. Nozaki).
for transdermal daily administration as patch (5 cm2 [4.6 mg/24 h], 10 cm2 [9.5 mg/24 h], 15 cm2 [13.3 mg/24 h] and 20 cm2 [17.4 mg/24 h]; of note: the 20-cm2 patch has not been launched to market to date). Transdermal patch is an optimal way to deliver rivastigmine and provides many benefits over conventional oral treatments, allowing patients easier access to optimal therapeutic doses.10 Treatment with rivastigmine patch is initiated with the 5-cm2 (4.6 mg/24 h) patch, which when well tolerated, is uptitrated to the 10-cm2 (9.5 mg/24 h) patch and beyond based on individual responses to obtain the desired therapeutic benefits.11 In Japan, more gradual titration approach starting from 2.5 cm2 and followed with 5 cm2, 7.5 cm2, and 10 cm2 is also available.12 All patch sizes have the same loading dose per area (1.8 mg/cm2), and the amount of rivastigmine delivered from a patch over a 24-h wearing period is approximately 50% of the total loading dose. After launch, there were some case reports of patch misuse, which raised questions related to efficacy and safety risks (e.g., longer application than 24 h, multiple patches applied at the same time, and so forth). Previously, modeling has been performed during the development of rivastigmine to describe PK after applications of rivastigmine patch.10,13 Such noncompartmental or compartmental models, however, were unable to answer these questions, because
http://dx.doi.org/10.1016/j.xphs.2016.04.011 0022-3549/© 2016 American Pharmacists Association®. Published by Elsevier Inc. All rights reserved.
2
S. Nozaki et al. / Journal of Pharmaceutical Sciences xxx (2016) 1-9
Figure 1. Compartment model to describe the PK profile of rivastigmine after rivastigmine patch application. A1, dose point that represents the applied patch in this model case; Cskin, drug concentration in the skin; V2, distribution volume in the skin; Ccentral, drug concentration in the central compartment; V3, distribution volume in the central compartment; A0, elimination compartment.
mostly oral administration models were applied, and the models did not quantitatively reflect the amount of drug delivered per the application time, which will be the basis of quantitative simulation of the systemic exposure. In general terms, pharmacokinetic (PK) models of dermal absorption of chemicals have been created and reported in a series of publications.14-18 Skin is usually represented either as a single compartment or by 2 compartments separately distinguishing the lipophilic and hydrophilic layers of the skin.17 Such modeling attempts were generally used in the toxicology rather than clinical pharmacology fields until recently when Polak et al.18 reported a mechanistic dermal absorption model using Simcyp simulator. Still, there is limited information on dermal absorption models that were put into practical use for simulations of drug concentrations in the clinical setting. In this study, we report a PK model created by integrating a mechanistic dermal absorption model with a systemic PK model to describe the PK after applications of rivastigmine patch for the purpose of simulating misuse scenarios. The model was created on a Phoenix WinNonlin platform that was selected as an optimal tool to fit the model to clinical data for refinement and verification, after which the model was used for the simulations. Methods Compartmental Structure of the Model A 2-compartment model was developed in WinNonlin, version 5.2, using a user-defined American Standard Code for Information Interchange model to describe the PK profile of rivastigmine after single 24-h dermal application of rivastigmine patch (initial model,
Appendix 1). The schematic diagram of the compartmental structure of the model is presented in Figure 1. The model comprised 2 compartments, skin and central compartment, as described by Brown and Hattis15 in which transfers of rivastigmine between compartments are described by rate constants K12 (for dermal absorption), K23 (skin to blood transfer), and K30 (elimination from systemic compartment; Fig. 1). The parameters for volume of skin and central compartment were defined as V2 and V3, respectively. A PK simulation was conducted to draw a PK profile of rivastigmine after 10-cm2 patch application using the initial parameters described in the following sections (bottom-up approach). Initial Values: 1. Dermal Absorption Initial parameter for the dermal absorption rate constant (K12) was estimated assuming a first-order absorption. First, the initial value was calculated using the data of amount of rivastigmine released from the patch after a 24-h application. An average of 9.4 mg (range, 7.3-11.8 mg) of the total drug content of 18 mg was released from 10-cm2 patch during 24-h application (n ¼ 19) in a previous clinical study.19 Calculation using the equation [K12 ¼ ln(released amount/drug content)/24 h] resulted in a mean K12 value of 0.027 h1 (range, 0.018-0.037 h1). The initial value was also calculated from the time course of in vitro human skin permeation data (0.026 h1; data on file). Because both methods gave similar values, 0.026 h1 was used as the initial value for K12. Initial Values: 2. Skin-to-Blood Transfer The rate constant for transfer of the drug from skin to blood (K23) was estimated as follows. The skin and blood are in
Table 1 Initial and Final Parameters Used in the Model Parameters (Unit)
Initial Values
Mean Values of the Estimated Individual Parametersa
Parameters Used in the Final Model
K12 (h1) K23 (h1) K30 (h1)
0.026b 0.3c 0.693d
0.027 0.473 0.521
V2 (mL) V3 (mL)
4e 94,800g
4 141,634
0.027 0.473 K30 ¼ [1/V3] [Vmax/(Km þ Ccentral)], where V3 ¼ 141,634, Km ¼ 24, Vmax ¼ 1,760,000 Note: K30 ¼ 0.497 when Ccentral << Km 4f 141,634
a b c d e f g
Individual parameters are presented in Table 2. Initial value estimated assuming first-order absorption based on (1) released rivastigmine from patch after 24-h application and (2) human skin permeation test data. Parameter for transfer from skin to blood estimated for rivastigmine. Elimination constant determined from T1/2 following intravenous administration of rivastigmine. Volume of skin compartment z patch area (10 cm2) skin thickness (0.4 cm) ¼ 4 mL. V2 was changed corresponding to the size of the patch; 2, 4, 6, and 8 mL for patches 5, 10, 15, and 20 cm2, respectively. See details in Methods section Initial Values. Volume of central compartment (distribution volume) determined from volume of distribution following intravenous administration of rivastigmine.
S. Nozaki et al. / Journal of Pharmaceutical Sciences xxx (2016) 1-9 Table 2 Individual Model Parameter Estimates for Pharmacokinetic (PK) Profiles After Single Application of 10-cm2 Patch Subject
K12 /h
K23 /h
K30 /h
V3 mL
CL ¼ K30 V3 mL/h
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 Min Max Mean Standard deviation CV% Variance of intersubject variability (u2)a Initial value Lower limit Upper limit
0.018 0.035 0.040 0.016 0.040 0.037 0.035 0.016 0.019 0.016 0.016 0.016 0.033 0.033 0.026 0.040 0.036 0.026 0.016 0.016 0.040 0.027 0.010
0.377 0.297 0.471 0.365 0.533 0.304 0.249 0.270 1.425 0.287 0.298 0.374 0.353 0.442 0.329 0.289 0.302 1.732 0.280 0.249 1.732 0.473 0.400
0.400 0.303 0.466 0.366 0.527 0.363 1.200 0.321 0.283 0.283 0.295 0.380 0.349 0.521 1.200 0.700 0.815 0.389 0.744 0.283 1.200 0.521 0.286
137,915 193,272 183,716 183,265 131,336 199,218 91,432 199,771 101,735 136,486 190,454 175,175 103,072 186,197 58,895 116,273 105,547 143,605 53,690 53,690 199,771 141,634 47,513
55,183 58,465 85,559 67,109 69,256 72,330 109,681 64,087 28,798 38,633 56,249 66,519 35,978 97,078 70,662 81,366 85,973 55,824 39,951 28,798 109,681 65,195 21,137
37.0 0.137
84.6 0.715
54.9 0.301
33.5 0.113
32.4 0.105
0.026 0.0160 0.0400
0.300 0.030 3.000
0.693 0.100 1.200
94,800 50,000 200,000
d d d
0:98 þ 0:02 Kow 0:993 þ 0:007 Kow
mL 4g 1:5 min100g Skin blood flow ¼ Dermal volume Ksb 4 mL 3
(3)
Initial Values: 3. Systemic Distribution and Elimination The elimination constant K30 was calculated based on the elimination half-life (T1/2) after intravenous administration of rivastigmine (~1 h).
K30 ¼ 0:693=ðT1=2Þ ¼ 0:693=1 ¼ 0:693:21
(4)
Volume of central compartment V3 was determined from distribution volume after intravenous administration of rivastigmine (1.5 L/kg)22 and a typical body weight of human population (63.2 kg).19
V3 ¼ 1:5 L=kg 63:2 kg ¼ 94:8 L:
(5)
Hence, 94.8 L was used as the initial value for V3. Fitting to the Clinically Observed Pharmacokinetic Profile, Model Verification, and Simulations The initial model was fitted to the actual PK profiles after single 24-h application of rivastigmine patch of 10 cm2 (n ¼ 19).19 Uniform weighting was used. Individual model parameters and descriptive statistics such as mean, standard deviation, and coefficient of variation (CV%), were derived (top-down approach). The mean values of the obtained individual parameters values for K12, K23, V2, and V3 (Table 1, middle column) were used for the upcoming final model. The final model was developed in a user-defined Phoenix Modeling Language (PML) using the calculated mean values and the CV% for model parameters K12, K23, K30, V2, and V3 as typical values (fixed effects, q), and random effects (u), respectively (final model, Appendix 2). The modeling and simulations were performed in Phoenix WinNonlin, version 6.1 or higher. To describe the overdose proportional exposure of rivastigmine,10 the elimination constant K30 was modified by assuming MichaeliseMenten kineticsebased saturation. The K30 expressed by the function of rivastigmine (Ccentral) is presented in the following equation.
(1)
where Ksb is a skin/blood partition coefficient, which can be expressed by the following equation:
Ksb ¼
K23 ¼
Hence, 0.3 h1 was used as the initial value for K23.
equilibrium with each other. Brown and Hattis15 proposed the following equation to express the flow rate.
Skin blood flow Dermal volume Ksb
K23 was calculated by applying the values stated above into the Equation 1.
¼ 0:005 min1 ¼ 0:3 hr1
Italic: estimated values that reached the lower or upper limit. The lower and upper limits of each parameter for simulation were determined based on the clinically obtained range when available (K12: range of the absorption rate constant calculated from the remaining amount of rivastigmine in the patch, K30: range of elimination rate constant after intravenous administration, V3: range of volume of distribution after intravenous administration), and 0.1 initial for lower and 10 initial for upper for others (K23 and V2). V2 was not presented because the value remained at 4 mL (initial value) in all individuals. The upper and lower limits for fitting were 0.40 and 40.0, respectively. a Calculated from the equation [(CV%)/100]2.
Skin-to-blood rate ð=minÞ ¼
3
(2)
The parameter Kow represents the distribution coefficient at 37 C in n-octanol/water. Considering that the Kow of rivastigmine is >100 (physical and chemical properties of rivastigmine, data on file), the Ksb can take values from 1.8 (when Kow ¼ 100) to 3 (when Kow ~ ∞), which are within close range. Ksb ¼ 3 was used for the calculation of initial value for K23. Skin blood flow was reported to be 1.5 mL/min/100 g.20 Dermal volume and weight of skin can be estimated to be 4 mL and 4 g, respectively, based on an assumption that the volume of skin compartment can be expressed as the multiplication product of patch area (10 cm2) and skin thickness (0.1-0.4 cm).15 Therefore,
K30 ¼
1 Vmax V3 Km þ Ccentral
(6)
where Ccentral ¼ Acentral V 3 , Acentral ¼ drug amount in the central compartment. The Km value that provided good estimation of the overdose proportional PK was searched manually by testing several values. In the process, the elimination constant K30 was fixed to 0.5 h1 so as to not contradict with the clinical T1/2 (the elimination constant 0.5 h1 corresponds to the T1/2 ¼ 1.4 h). When the Km was 24 ng/mL and the Vmax was 1,760,000 mL, the simulated PK parameters accurately captured the overdose proportional increase in AUCtau and Cmax. These values were used for the final model. The parameter for volume of skin (V2) was changed corresponding to the size of the patch (2, 4, 6, and 8 mL for patches of 5, 10, 15, and 20 cm2, respectively). The parameters used in the final model are presented in Table 1 (right column). The PML code for the final model is presented in Appendix 2.
4
S. Nozaki et al. / Journal of Pharmaceutical Sciences xxx (2016) 1-9
Figure 2. Model fitted to the individual PK profiles after single 24-h application of 10-cm2 patch. The plasma rivastigmine concentrations were measured from time 0 (patch application) to 48 h. Patches were removed at 24 h. Open circle: observed concentrations, solid line: predicted concentrations. The observed concentrations are the PK profiles after single application of rivastigmine patch of 10 cm2 in a previous clinical study.19
S. Nozaki et al. / Journal of Pharmaceutical Sciences xxx (2016) 1-9
5
Simulated Pharmacokinetic Profiles After Normal Applications (Model Verification)
Final Model
dAcentral ¼ Acentral K30 þ Askin K23 dt dA0 ¼ Acentral K30 dt n X dAskin ¼ Askin K23 þ Dosei K12 dt i¼1
dDosei ¼ Dosei K12 dt A Ccentral ¼ central V3 1 Vmax K30 ¼ V3 Km þ Ccentral The parameter Askin represents the drug amount in the skin. The parameter Dosei represents the loaded (time ¼ 0), or remaining (time > 0), amount of the drug in the patch of the i time(s) of the application. Dosei turns into zero at the time of patch removal. n times of application was assumed. The final model was verified by stochastic simulation of steadystate PK profiles following 5, 10, 15, and 20 cm2 patch applications using Phoenix Monte Carlo population PK simulation run mode and by comparing them to clinically observed PK profiles. The clinical data used for the verification10 were from different subjects that were used for the initial fitting.19 Variance of each parameter (CV%) obtained in the initial fitting was used for the simulation. The u2 values were 0.137, 0.715, 0.301, and 0.113 for K12, K23, K30, and V3, respectively, as presented in Table 2. After sampling 1000 replicated data sets, 95% confidence intervals (CIs) for the logs of the simulated data sets were computed and presented graphically with the observed PK profiles. Mean plasma rivastigmine concentration-time profiles after normal and misuse applications were simulated using the final model with Phoenix individual simulations. PK parameters at steady state (Cmax and AUCtau) were calculated for the simulated concentration-time profiles using Phoenix WinNonlin NCA object. Results
The plasma rivastigmine concentration-time profiles after applications of 5-, 10-, 15-, and 20-cm2 patches were simulated and are presented in Figure 3. The comparison between simulated PK profiles/parameters and those of the observed concentrations in a previously conducted clinical study10 is presented in Figure 4 (profiles), Figure 5, and Table 3 (parameters). Overall, the simulated mean concentrations were close to the mean of actual individual concentrations. The actual individual concentrations (n ¼ 13-23) were almost covered by the upper and lower 95% CIs for the logs of the simulated data sets (Fig. 4). Predicted Cmax and AUCtau were close to the observed values, except a tendency of overestimation in AUCtau (Table 3 and Fig. 5).
Simulated Pharmacokinetic Profile After Misuse Applications The simulated concentration-time profiles based on a series of misuse scenarios are presented in Figure 6. Simulations were conducted under steady state conditions for scenarios A, B, and C, more specifically on third day of repeated administrations or after.10 (a) A new 10-cm2 patch was applied while the previous one was kept for an additional 0.5, 1, 2, 6, 10, 18, or 24 h: When the previous patch was kept for an additional 2 h, there was a limited change in the concentration-time profiles. When the previous patch was kept for additional 6 h, the peak concentrations were significantly elevated, close to the concentration levels reached with a 15-cm2 patch. (b) Simulated plasma rivastigmine concentration-time profiles of 10 days application with new patches applied every day without removing previous ones: From time 0 to 48 h (days 2, 1, and 1), 1 patch applied per day (normal applications). From time 48 to 288 h (days 2 to 10), each daily patch was not removed once applied, meaning that 1, 2, 3, …, 10 patches were applied at the same time on days 1, 2, 3, …, 10, respectively, mimicking off-label use. All the patches were removed at time 288 h (day 10). The simulations showed that such usage of 10-cm2 patch resulted in elevated peak concentrations at steady state comparable to the levels
“Bottom-Up” ApproachdSimulated Plasma Rivastigmine Concentration Using Initial Parameters PK profile of rivastigmine after 10-cm2 patch application was simulated using the initial parameters described in the Methods section Initial Values: 1-3. The simulated plasma rivastigmine concentration-time curve was close to the actual concentrations, indicating that the initial parameters were good enough to start the fitting from (data not presented). “Top-Down” ApproachdFitting the Model to Clinical Data The model was then fitted to the actual PK profiles after single application of rivastigmine patch of 10 cm2 in a previous clinical study.19 Individual parameters are presented in Table 2, and the individual PK profiles are shown in Figure 2. The model fitted well, not only to the individual concentration-time profiles during patch application (0-24 h), but also to those after patch removal (24-48 h). Mean values of the individual parameters for K12, K23, V2, and V3 were similar to the respective initial values, reconfirming that the prediction by “bottom-up” approach was good.
Figure 3. Simulated PK profiles after multiple applications. Black lines: simulated concentration-time profiles, lower to upper: 5, 10, 15, and 20 cm2. For the simulations, mean values of each model parameters were used without considering the variability.
6
S. Nozaki et al. / Journal of Pharmaceutical Sciences xxx (2016) 1-9
Figure 4. Simulated PK profiles overlaid with the observed individual PK profiles at steady state for patches of 20 cm2 (a), 10 cm2 (b), and 5 cm2 (c). Red line: simulated mean concentration-time profiles. Dotted red line: lower/upper limits of 95% CIs for the logs of the generated 1000 sets of concentration-time profiles by Monte Carlo simulation. Black line: mean of the observed individual concentration-time profiles. Gray lines: observed individual concentration-time profiles. (a) n ¼ 13, (b) n ¼ 22, and (c) n ¼ 23. Left: linear scale and right: log scale.
S. Nozaki et al. / Journal of Pharmaceutical Sciences xxx (2016) 1-9
7
(d) Patches of 15 cm2 changed every 1 h instead of every 24 h: The simulated concentration reached steady state around 20 h. The concentrations were similar to those reached with 20-cm2 patch under normal application conditions. Discussion
Figure 5. Simulated and observed PK parameters (a) Cmax: Cmax at steady state, (b) AUC: AUCtau at steady state. Closed square: observed parameters. Open diamond: simulated parameters.
reached with the 20-cm2 patch. Exposure was ~2.6 times higher relative to normal use in the case of 10 cm2 and ~3.8 times higher in the case of 15 cm2, exceeding the levels reached with 20 cm2. (c) Simulated plasma rivastigmine concentration-time profiles after 1, 2, 3, 4, or 5 patches of 10 cm2 applied at the same time and remained on the skin: The peak concentrations in these misuse case scenarios were 1-, 2.4-, 4.7-, 8.0-, and 12.7-fold those obtained under normal (label) application conditions when 1, 2, 3, 4, or 5 patches were applied at the same time, respectively. The total exposure (AUC calculated from 48 h to infinity) increased 1-, 2.3-, 4.3-, 7.5-, and 12.7-fold, respectively.
In this study, a model was developed to describe the PK profiles of rivastigmine patch and to simulate its PK under misuse situations (e.g., multiple patches applied at the same time, patches replaced more frequently than as per label, and so forth). This model is the first to describe the PK of rivastigmine using dermal absorption model. In addition, to our knowledge, this is the first case example that leveraged a dermal absorption model into practical use for simulations of systemic drug concentrations. Previous models to describe the PK profiles of rivastigmine patch10,13,19 were noncompartment or 1-compartment models with first-order administration and elimination process. Such models were unable to simulate the PK after patch misuse, on account of inaccurate mass balance in drug amount delivered from the patch. The amount of drug available in the body as a function of time was modeled as A(t) ¼ Dose(t0) exp(ka t). The population fixed estimate for the ka parameter obtained by fitting the 1-compartment model to the clinical data was 0.0667 h1 (data on file). Therefore, the amount of dose remaining into the patch after 24 h according to the 1-compartment model is Dose(t0) exp(0.0667 24) ¼ Dose(t0) 0.201. This quantity represents approximately 20% of the initial amount of drug available in the patch. This figure differs significantly from the measured value of approximately 50% of the drug load remaining into the patch after the 24-h wearing period as reported in the original article.10 Although the 1-compartment model could adequately serve its initial purpose of describing the individual PK profiles and could be used for subsequent PK modeling, that model was not flexible enough to answer other questions. In addition, the model could not be used to support (for instance, by running simulations) further statements on the impact of noncompliance to treatment on individual PK profiles. The newly created model in this work was based on a mechanistic dermal absorption model assuming a barrier from patch to the skin. Mechanistic dermal absorption models have been reported mostly in a toxicology area.17 Generally speaking, skin compartment has been expressed as 1- or 2-compartments depending on the model's purpose.17 Dividing the skin part into 2 compartments is useful when considering the skin as a storage compartment, especially when the purpose of the model is to estimate the toxicity of the compound in the skin.16 However, 1 compartment was used in this study for the skin part because it was considered adequate to serve the purpose of this modeling exercise, which was to describe the systemic concentration of the drug. In addition, limiting the number of compartments is recommended to avoid overparameterization when fitting to clinical PK. A rate constant expression by Brown and Hattis15 was used to describe the transfer from skin to blood (Eq. 1). This equation is one of the several similar models based on a common assumption that
Table 3 Simulated and Observed Pharmacokinetic (PK) Parameters Dose (cm2)
5 10 15 20
Cmax at Steady State (ng/mL)
AUCtau at Steady State (ng h/mL)
Simulated
Observed
Ratio (Simulated/Observed)
Simulated
Observed
Ratio (Simulated/Observed)
3.17 7.18 12.3 19.1
2.71 7.88 14.1 19.5
1.17 0.91 0.87 0.98
65.3 147 254 397
46.3 127 233 345
1.41 1.16 1.09 1.15
8
S. Nozaki et al. / Journal of Pharmaceutical Sciences xxx (2016) 1-9
Figure 6. Simulated concentration-time profiles following various scenarios of rivastigmine patch applications. (a) A new 10-cm2 patch (fourth patch) was applied while the previous one (third patch) was kept for additional 0.5, 1, 2, 6, 10, 18, or 24 hr (black line) vs normal applications (red-dotted lines, 10 cm2, 15 cm2, 20 cm2 from lower to upper). (b) Simulated plasma rivastigmine concentration-time profiles of 10-day application with new patches applied every day without removing previous ones, starting from fourth application (black solid line: 10 cm2, dotted line: 15 cm2) vs normal applications (red-dotted lines, 10 cm2, 15 cm2, 20 cm2 from lower to upper). (c) Simulated plasma rivastigmine concentration-time profiles after 1, 2, 3, 4, or 5 patches of 10 cm2 applied at the same time and remained on the skin (black lines) vs normal applications (red-dotted lines, 10 cm2, 15 cm2, 20 cm2 from lower to upper). The misuse was assumed to happen on the third day. (d) Patches of 15-cm2 changed every 1 h instead of every 24 h (black solid line), Normal applications after 15- and 20-cm2 patch applications from lower to upper (red-dotted lines, for reference). For the simulations, mean values of each model parameters were used without considering the variability.
Figure 7. Impact of varying K23 values on the simulated concentration-time profiles. Plasma rivastigmine concentrations after single 24-h application of 10-cm2 patch were simulated using the final model. K23 values were varied while all the other parameters remained constant. Lines represent the results for K23 ¼ 0.249 (blue), 0.473 (black), and 1.732 (red), respectively.
clearance from the skin is controlled by the solubility of the compound in the blood relative to that in the skin and the skin blood flow rate and area, as reviewed by McCarley and Bunge.17 This equation was selected because it was based solely on physicochemical or physiological parameters and did not require parameters that need to be determined experimentally. In addition, models using this type of equation have successfully predicted clinical systemic PK profiles of several compounds.18 The simulated PK profile by “bottom-up” approach (simulation based on initial values, Table 1 left column) was close to the clinically observed PK profiles, suggesting that the model structure was valid and conforming to the clinical situations. When the model was fitted to the clinically observed data, the model was well applicable and allowed to describe most of the individual PK profiles. In some subjects, however, predicted curve did not fully capture the steep peak concentrations (e.g., subjects 1, 6, and 8). Reason behind those sharp peaks is unknown, but one possibility could be that there was a transient increase in skin absorption rate during application (e.g., due to temporal heating), resulting in temporal rise in K12. Such profile cannot be described by this model where K12 is constant over time. Although this might be a
S. Nozaki et al. / Journal of Pharmaceutical Sciences xxx (2016) 1-9
limitation of this model, the overall fit was good and considered adequate to serve its initial purpose of describing the typical PK profiles. Estimated mean parameters as a result of the fitting were similar to the respective initial values, reconfirming that the prediction by “bottom-up” approach was correct. It was notable that the parameter for volume of skin (V2) did not vary at all (i.e., remained as the initial value; 4 mL). This means that the volume of skin was not a sensitive parameter to determine systemic exposure. Some of the estimated values for K12 and K30 reached the lower or upper limit (8 and 2 subjects of 19 for each). The lower and upper limits of these parameters were determined based on ranges of clinical observations, which acted as a gatekeeper preventing the estimated parameters from being clinically irrelevant. Other parameters (K23 and V3) varied within the range without reaching the limits. There was a large variability associated with the K23 estimate (range, 0.249-1.732/h, CV ¼ 84.6%). To assess the impact of K23 to the results, we have conducted a sensitivity analysis by varying K23 values (0.249, 0.473, and 1.732) while all the other parameters remained constant (Fig. 7). The result showed that the K23 values have a clear effect on the shape of the PK profile before and after removal of the patch. This suggested that K23 was mathematically useful to describe the varying shape of individual PK profile. From physiological point of view, variation in K23 values may be attributable to individual differences in fat content and/or blood flow rate in the skin, although this could not be verified. Overall, these facts indicated that the model fit was adequate. Because rivastigmine shows overdose proportional exposure,10 the model was modified accordingly to adequately describe the PK across the patch doses of 5, 10, 15, and 20 cm2. Considering that rivastigmine is not metabolized in the skin (2% metabolized after 72 h, data on file) and the major route of elimination is hydrolysis by esterase, a saturation factor was incorporated into the elimination constant K30. Predicted Cmax and AUC conformed well to the observed values, although AUC showed a tendency of ~15% overestimation (Fig. 5). However, a slight overestimation in total exposure was considered acceptable (and preferable to an underestimation) from a safety risk assessment perspective. In addition, the simulated time-concentration profile was within the range of variation when compared with observed individual values (Fig. 4). Furthermore, the model was verified by simulating 1000 replications by Monte Carlo simulations that resulted in the 95% CIs containing most subjects' concentration-time profiles (n ~ 20). In other words, the model and the variance parameter (u) were adequate to simulate PK profiles for various scenarios with good estimation of 95% CIs. It is of note that simulation of mean profiles was sufficient to answer most questions relative to PK differences between normal application and misuse conditions. This work demonstrated the utility of the PK modeling in simulating PK under various situations. Prospective predictions of PK after off-label misuse are useful to answer frequently asked questions in the case of patch or tape formulations as unintended removal or application could happen. The simulations demonstrated that the model was flexible enough to address these questions. We could compare simulated PK profiles relative to the typical normal profiles of marketed patch as a tool to assess the safety and efficacy implications of the off-label use (Fig. 6). Although the clinical implication of the simulation results is out of scope of this article, most of the scenarios tested (Fig. 6) resulted in a quick increase in exposures reaching PK levels comparable to those measured at the next upper dose level. This indicates that these off-label/misuses of Exelon Patch must be avoided in clinical practice, especially when the simulated concentrations exceed the typical exposures observed with the 20-cm2 patch, which is the highest ever-tested patch dose in clinical, and would require a safety alert.
9
It should be emphasized that the users of the model need to be fully aware of the model assumptions and limitations to understand the differences in outcomes from other models. Limitations of the model include potential overestimation of the exposures at higher levels than those ever observed in clinical setting (i.e., Cmax ~ 100 ng/mL). In an actual situation, another elimination route may contribute more if the original route or enzyme is saturated and possibly compensate for the decreased clearance. Hence, extrapolation to higher concentrations should be treated with caution. Limitations of the model also include that the model may not be appropriate to simulate the PK when system parameters in the elimination process could be altered, for example, special populations, because the model does not include mechanistic parameters in the elimination process and cannot accommodate for situations when these processes are altered. References 1. Farlow M, Anand R, Messina Jr J, Hartman R, Veach J. A 52-week study of the efficacy of rivastigmine in patients with mild to moderately severe Alzheimer's disease. Eur Neurol. 2000;44(4):236-241. 2. Rosler M, Anand R, Cicin-Sain A, et al. Efficacy and safety of rivastigmine in patients with Alzheimer's disease: international randomised controlled trial. BMJ. 1999;318(7184):633-638. 3. Karaman Y, Erdogan F, Koseoglu E, Turan T, Ersoy AO. A 12-month study of the efficacy of rivastigmine in patients with advanced moderate Alzheimer's disease. Dement Geriatr Cogn Disord. 2005;19(1):51-56. 4. Emre M, Aarsland D, Albanese A, et al. Rivastigmine for dementia associated with Parkinson's disease. N Engl J Med. 2004;351(24):2509-2518. 5. Onor ML, Trevisiol M, Aguglia E. Rivastigmine in the treatment of Alzheimer's disease: an update. Clin Interv Aging. 2007;2(1):17-32. 6. Finkel SI. Effects of rivastigmine on behavioral and psychological symptoms of dementia in Alzheimer's disease. Clin Ther. 2004;26(7):980-990. 7. Farlow MR, Cummings JL, Olin JT, Meng X. Effects of oral rivastigmine on cognitive domains in mild-to-moderate Alzheimer's disease. Am J Alzheimers Dis Other Demen. 2010;25(4):347-352. 8. Burns A, Spiegel R, Quarg P. Efficacy of rivastigmine in subjects with moderately severe Alzheimer's disease. Int J Geriatr Psychiatry. 2004;19(3):243-249. 9. Cutler NR, Polinsky RJ, Sramek JJ, et al. Dose-dependent CSF acetylcholinesterase inhibition by SDZ ENA 713 in Alzheimer's disease. Acta Neurol Scand. 1998;97(4):244-250. vre G, Sedek G, Jhee SS, et al. Pharmacokinetics and pharmacodynamics of the 10. Lefe novel daily rivastigmine transdermal patch compared with twice-daily capsules in Alzheimer's disease patients. Clin Pharmacol Ther. 2008;83(1):106-114. 11. Exelon Patch® US Prescribing Information 2015. Available at: http://www. pharma.us.novartis.com/product/pi/pdf/exelonpatch.pdf. Accessed May 11, 2016. 12. Exelon Patch product information (Package Insert). Available at: http://www. info.pmda.go.jp/downfiles/ph/PDF/300242_1190700S1029_1_07.pdf Accessed April 7, 2016. vre G, Huang HL, Schmidli H, Amzal B, Appel-Dingemanse S. 13. Mercier F, Lefe Rivastigmine exposure provided by a transdermal patch versus capsules. Curr Med Res Opin. 2007;23(12):3199-3204. 14. Reddy MB, McCarley KD, Bunge AL. Physiologically relevant one-compartment pharmacokinetic models for skin. 2. Comparison of models when combined with a systemic pharmacokinetic model. J Pharm Sci. 1998;87(4):482-490. 15. Brown HS, Hattis D. The role of skin absorption as a route of exposure to volatile organic-compounds in household tap waterda simulated kinetic approach. J Am Coll Toxicol. 1989;8(5):839-851. 16. Shatkin JA, Brown HS. Pharmacokinetics of the dermal route of exposure to volatile organic chemicals in water: a computer simulation model. Environ Res. 1991;56(1):90-108. 17. McCarley KD, Bunge AL. Pharmacokinetic models of dermal absorption. J Pharm Sci. 2001;90(11):1699-1719. 18. Polak S, Ghobadi C, Mishra H, et al. Prediction of concentration-time profile and its inter-individual variability following the dermal drug absorption. J Pharm Sci. 2012;101(7):2584-2595. vre G, Buche M, Sedek G, et al. Similar rivastigmine pharmacokinetics and 19. Lefe pharmacodynamics in Japanese and white healthy participants following the application of novel rivastigmine patch. J Clin Pharmacol. 2009;49(4):430-443. 20. Kai K, Inoue S, Higaki Y, Tomokuni K. Effects of a cool environment on the health of female office workers and students. J Physiol Anthropol. 2008;27(3): 153-159. 21. Hossain M, Jhee SS, Shiovitz T, et al. Estimation of the absolute bioavailability of rivastigmine in patients with mild to moderate dementia of the Alzheimer's type. Clin Pharm. 2002;41(3):225-234. 22. Exelon Patch product information (Interview Form) Available at: http://www. info.pmda.go.jp/go/interview/1/300242_1190700S1029_1_EXP_1F Accessed April 7, 2016.