Exercise 13

Deconvolution analysis vs. modelling to document the process of drug absorption

Objectives

To implement a WNL user model for a monocompartmental model ith t o processes of absorption either ith an algebraic e!uation or ith a set of differential e!uations. To compare t o concurrent models" monocompartmental# ith a single site of absorption vs. a monocompartmental model ith t o sites of absorption using the $%& criterion To obtain the input function of a drug using deconvolution to reveal information about its process of absorption.

Overview Double pea's in the plasma concentration(time profile follo ing oral administration have been reported for several compounds. )ultiple pea's after oral administration of a drug could have several physiological causes. To describe a complicated drug plasma profile li'e a double pea'# there are t o possible approaches" *i+ curve fitting using a modified customary compartmental model *ii+ numerical deconvolution to identify the input rate of the drug in the central compartment. ,or the present exercise e ill implement a pharmaco'inetic model in WNL that as previously used in exercise - *)onte &arlo simulations# .ioe!uivalence and ithdra al time+ incorporating absorption from t o different sites. The results ill be compared ith those from a deconvolution analysis.

Exercise
The goal of this experiment as to compare t o formulations *$ and .+ of a ne drug product administered by the oral route at a dose of 1// 0g1'g. T elve *12+ animals ere investigated follo ing a cross3over design and the ra data are given in the corresponding Excel sheet. $fter importing the data set in WNL# you ill inspect individual curves# you irregular initial phase ith a shoulder for some animals *e.g. animal 21+. ill see an

%t is clear that fitting this animal ith a conventional model ould most li'ely lead to some misfit especially during the first hours follo ing drug administration and even if a fitting is apparently satisfactory# the estimated parameters could be severely biased. We ill start to fit these different curves ith a classical monocompartmental model ith a single rate constant of absorption .

The next figure sho s the fitting obtained for animal 21 that loo's rather good.

4o ever inspection of residuals is more informative and it immediately enables departure to the model to be detected.

&ompute the different parameters associated ith this monocompartmental model. )onocompartmental model ith a single rate constant of absorption# and mean parameters for the t o formulations"
Variable K10 K10 V_F V_F K01 K01 Formulation A B A B A B N 12 12 12 12 12 12 Mean 0.0418 0.0371 3.1546 4.1748 0.0982 0.1064 SD 0.0038 0.0053 0.3450 1.5003 0.0140 0.0121 Min 0.0377 0.0261 2.4330 2.6644 0.0794 0.0853 Max 0.0490 0.0447 3.5734 7.8159 0.1133 0.1294 CV% 9.0431 14.2316 10.9365 35.9374 14.2178 11.3558 Geometric_Mean 0.0416 0.0368 3.1363 3.9669 0.0973 0.1058

Variable AUC AUC CL_F CL_F K01_HL K01_HL Cmax Cmax K10_HL K10_HL Tmax Tmax

Formulation A B A B A B A B A B A B

N 12 12 12 12 12 12 12 12 12 12 12 12

Mean 766.6596 696.0401 0.1307 0.1480 7.1911 6.5905 16.8647 14.7399 16.7088 19.0709 15.2337 15.2968

SD 36.7152 123.3195 0.0062 0.0272 1.0390 0.7583 1.0422 3.6567 1.4535 3.1261 0.7055 0.7617

Diagnostic table for the monocompartmental model"

Formulation A A A A A A A A A A B B B B B B B B B B Item AIC CORR_(OBS,PR !" CSS !F S SBC SSR #CSS #SSR !"_C#$$_%#&S'($)D* AIC CORR_(OBS,PR !" CSS !F S SBC SSR #CSS #SSR #T_CORR_(OBS,PR !" N 12 12 12 12 12 12 12 12 12 12 12 12 12 12 12 12 12 12 12 12 Mean 14.87 0.9936 478.1272 11.0000 0.6135 16.7937 6.4052 478.1272 6.4052 +.,,1+.2112 0.9925 378.1471 11.0000 0.5048 12.1284 4.2252 378.1471 4.2252 0.9925 SD 27.6988 0.0064 64.4449 0.0000 0.4739 27.6988 6.4875 64.4449 6.4875 +.++ 4 27.2787 0.0074 174.8035 0.0000 0.3756 27.2787 4.4506 174.8035 4.4506 0.0074

Corrected sum of squared observations (CSS), weighted corrected sum of squared observations (WCSS), sum of squared residuals (SSR), weighted sum of squared residuals (WSSR), estimate of residual standard deviation (S) and degrees of freedom (DF), the correlation between observed Y and redicted Y, the weighted correlation, and two measures of goodness of fit! the "#ai#e $nformation Criterion ("$C) and Schwar% &a'esian Criterion (S&C)(

)odel ith t o sites of absorption

We ant to compare results obtained ith this classical model ith those of a user model able to describe the same data but ith a model incorporating t o sites of absorption. This is the model that e previously used to simulate our data set ith &rystal .all *exercise -+. %n this monocompartmental model# a first fraction *,1+ of the administered dose is available from site1 ith a rate constant of absorption 5a16 and the second fraction *,2+ of the administered dose is available from site 2 ith a rate constant 5a2. We assumed that the availability is total and that ,2713,16 8c is the volume of distribution of the central compartment6 This model can describe a double pea' in plasma concentration3time curves"

,1 x Dose ,raction 1 *:+ 5a1

;lasma 8c
,2 x Dose ,raction 2 *:+ 5a2

51/

The system can be very easily described by a set of differential e!uations such as" D<*1+ 7 35a1=<*1+ D<*2+ 7 35a2=<*2+ D<*3+ 7 5a1=<*1+>5a2=<*2+351/=<*3+ Where"

D<*1+ 7 35a1=<*1+ is the differential e!uation describing the disappearing of the drug from the first site of absorption *<*1++ throughout 5a1# a first order rate constant of absorption# D<*2+ 7 35a2=<*2+ is the differential e!uation describing the disappearing of the drug from the second site of absorption *<*2++ throughout 5a2# a first order rate constant of absorption. ,or both e!uations# the minus sign indicates that the drug disappears from their respective sites of administration.

D<*3+ 7 5a1=<*1+>5a2=<*2+351/=<*3+ is the differential e!uation describing the arrival of drug in the central compartment *<3+ i(e( *5a1=<*1+>5a2=<*2++ minus its disappearance from the central compartment i(e( *351/=<*3++ throughout 51/# the first order rate constant of elimination. $ny ;5 model can be described li'e that ith a rather natural description of the drug disposition. The next step to describe the model is to !ualify the so3call initial condition of the system to tell the numerical solver ho to proceed at time /6 here e 'no that a fraction of the dose *noted ,?+ is in the first site of absorption and the rest of the dose *noted 13,?+ in the other site thus" <*1+ 7 ,?=Dose <*2+ 7 *13,?+=Dose <*3+ 7 / Where"

<*1+ 7 ,?=Dose indicates that a ,? fraction of the dose *from / to 1+ is located in site 1# i.e. *<1+ at time / <*2+ 7 *13,?+=Dose# indicates that the rest of the dose should be located at time / in the second site of administration# i.e. < *2+ <*3+ 7 / indicates that there is nothing in the central compartment <*3+ at time /

The number of parameters to estimate is 9 namely @,?@# @5a1@# @5a2@# and @51/@. No e have to rite a set of statements to actually implement this model in WNL. ritten ith

The follo ing set of statements corresponds to the 2 site model differential e!uations to be actually run in WNL.

$ command file is made up of bloc's of text# including a model bloc'# and other bloc's specifying all or most of the information re!uired to run the model. Each user model must begin ith the 'ey ord MODEL. %t can be follo ed by a number to identify the model# such as )ADEL 1. The model ends ith the command EOM hich indicates the Bend of the model definitionC. 1. Command block The WinNonlin commands are a group of commands *in red+ to define values such as NP ! ME"E!# *the number of parameters+. 4ere e have 9 parameters to estimate. PN ME# $%!$& $'a1$& $'a($& $'1)$ indicates the name of the 9 parameters to be estimated *do not forget !uotation mar's to declare your parameters+. The NCON#" N"# command specifies the number of constants to be used in the model. ,or our model there ill be only 1 constant that is the dose6 this constant must be initialiDed via the WinNonlin interface as for a classical model. The NDE!*+ "*+E# command tells WinNonlin the number of differential e!uations in the model *here n73+. The N%,NC"*ON# command specifies the number of different e!uations that have observations associated ith them6 here e declared 3 because if % ish to simulate my 3 compartments *2 sites of absorption plus the central compartment+# WNL should solve 3 functions. END bloc' allo s WinNonlin commands to be associated ith a model. ,ollo ing the &A))$NDE statement# any WinNonlin command such as N;$?$)ETE?E# N&ANET$NTE# ;N$)E# etc. may be given. The &A))$ND section concludes ith an END statement. (. "em-orar. variables block 8ariables defined in the temporary bloc' are general variables6 that is# they may be used in any bloc'. 4ere % have only one temporary variable * Dose/CON0111 meaning t2at * can re-lace CON011 in m. model b. t2e dose 0for e3am-le in t2e initial condition block1. 4. #tarting values block When a model is defined by a system of one or more differential e!uations *i.e. NDE!*+ "*+E# 4+# the starting values corresponding to each differential e!uation in the system# must be specified. 5. %unction block This group of statements defines the model to be fitted to a set of data. The letter N *here 1# 2 or 3+ denotes the function number. $ function bloc' must appear for each of the N%,NC"*ON# functions of the command bloc'. The variable , should be assigned to the function value. The function may be"

 an algebraic function *e.g. &7Dose18=*'a1=,1=exp*3'a1=T+1*'3'a1+ > 'a2=,2=exp*3'a2=T+1*'3'a2+ > *'a1=,1=*'a23'+>'a2=,2=*'a13'++=exp*3'=T+1**'a13 '+=*'a23'+++ for our algebraic model or a function of a differential e!uation or differential e!uations defined in one or more differential bloc's *e.g." ,7<*1+# or ,7 <*1+ etc+ 6. Differential e7uation block This set of statements is used to define a system of differential e!uations defines a model hich

M#D).
$%ma$&'''(((((((((((((((((((((((((((((((((((((((((((((((((((((( $%ma$&'''!%)%*+,%$- PL'T+./a01 $%ma$&'''2+3%*'!a/%1241242009 $%ma$&'''2+3%*'V%$50+11.0 $%ma$&'''(((((((((((((((((((((((((((((((((((((((((((((((((((((( $%ma$& $%ma$&'4'3%601%'m+3%*45,%70607'7+mma135'

C#MMANDS/ NC#NS"AN"/1 NF0NC"I#NS/ND)$IVA"IV)S/N(A$AM)")$S/4 (NAM)S//1F$1'/12a11'/12a21'/121+1 )ND

$%ma$&'4'3%601%'/%m,+$a$8')a$0a9*%5

")M(#$A$3/ Do4e5C#N%1* )ND'

$%ma$&'4'3%601%'3066%$%1/0a*'%:.a/0+15'5/a$/01;')a*.%5 START' <(1"'='FR(!+5% <(2"'='(14FR"(!+5% <(3"'='0 >! $%ma$&'4'3%601%'3066%$%1/0a*'%:.a/0+15

DIFF)$)N"IA./ D6%1*/5/72a186%1* D6%2*/5/72a286%2* D6%-*/5/2a186%1*92a286%2*721+86%-* )ND

$%ma$&'4'3%601%'a*;%9$a07'6.17/0+15

F0NC"I#N/1 F5/6%1* )ND F0NC"I#N/2 F5/6%2* )ND F0NC"I#N// F5/6%-* )ND

$%ma$&'4'3%601%'a18'5%7+13a$8',a$am%/%$5 $%ma$&'4'%13'+6'm+3%*

)#M
G

%f you ant to plot the 3 functions * e(g( ith a simulation+ you have to edit your data set indicating the time etc. $lternatively# you can declare only a number of functions e!ual to 1 i(e( only ,7<*3+. This model using differential e!uations is easy to rite but it is preferable# henever possible# to use the corresponding algebraic e!uations. %n this case# using the Laplace transform# the e!uation giving the plasma concentration is"
C (t ) = Dose ( A + B + C) Vc

With"
A=

Ka1 F1 Exp( Ka1 t ) ( K10 Ka1)

B=

Ka 2 (1 F1) Exp( Ka 2 t ) ( K10 Ka 2)

Ka1 F1 ( Ka 2 K10) + Ka 2 F 2 ( Ka1 K10) Exp( K10t ) ( Ka1 K10) ( Ka 2 K10)

C=

This model does not exist in the WNL library and e ill implement it. The follo ing set of statements corresponds to the 2 site model ritten ith an algebraic e!uation. The structure of the model is the same as for the preceding one6 % Hust add t)* as a tem orar' variable to tell WNL that the independent variable *al ays I+ can be ritten ith a t in my e!uation"
2O! L
$%ma$&'''(((((((((((((((((((((((((((((((((((((((((((((((((((((( $%ma$&'''!%)%*+,%$- P*'/+./a01 $%ma$&'''2+3%*'!a/%0342242011 $%ma$&'''2+3%*'V%$50+11.0 $%ma$&'''(((((((((((((((((((((((((((((((((((((((((((((((((((((( $%ma$& $%ma$&'4'3%601%'m+3%*45,%70607'7+mma135' CO22A>!S' >FU>CT?O>S'1 >CO>'1 >PARA2 T RS'5 P>A2 S''@&a1@,'@&a2@,'@&10@,'@V@,'@F1@ >! $%ma$&'4'3%601%'/%m,+$a$8')a$0a9*%5 T 2PORARA' !+5%=CO>(1" $%ma$&-'/'05'/B%'013%,%13%1/')a$0a9*% /=C >!' $%ma$&'4'3%601%'a*;%9$a07'6.17/0+15 FU>CT?O>'1' F='(!+5%DV"((((&a1(F1(%x,(4&a1(/"D(&104&a1""'E'(&a2((14F1"(%x,(4

&a2(/"D(&104&a2""'E'(((&a1(F1((&a24&10""E(&a2((14F1"((&a14 &10"""(%x,(4&10(/"D((&a14&10"((&a24&10"""""
>! $%ma$&'4'3%601%'a18'5%7+13a$8',a$am%/%$5 $%ma$&'4'%13'+6'm+3%*

O2

&reating a ne $E&%% model in WNL

Eelect +ser ,odel in the )odel Types dialog and clic' -e.t. The $E&%% )odel Eelection dialog appears. To create this ne Kser )odel# there are t o options"

The first option consists of riting your o n model ith the assistance of the WNL dialog box that offers a template to edit. T o user model templates are provided ith WinNonlin" one for models that include only algebraic e!uations and one for models that include differential e!uations.

The second option consists of selecting one of the t o options *differential or algebraic+ but# instead of riting the hole model yourself# you can copy the present ord document and paste it directly in WNL.

?unning the algebraic model corresponding to 2 sites of absorption

Ksing the algebraic model# estimate yourself the corresponding parameters for the 29 animals. Estimate l each individual curve6 loo' at the different graphs including residuals and then ith the statistical tool of WNL# ma'e the follo ing tables.

The first table gives you mean parameters for the 29 animals and the second table is the diagnostic table provided by WNL and that ill be used to compare the merits of the present model against those of the classical monocompartmental model. )odel ith 2 sites of absorption# mean parameters for the t o formulations"
Formulation A B A B A B A B A B N 12 12 12 12 12 12 12 12 12 12 Mean 0.0543 0.0542 0.0017 0.0018 0.6605 0.5808 0.0815 0.0814 1.1852 1.1944 SD 0.0097 0.0097 0.0004 0.0004 0.0694 0.1806 0.0193 0.0194 0.2066 0.2210 S) 0.0028 0.0028 0.0001 0.0001 0.0200 0.0521 0.0056 0.0056 0.0596 0.0638 CV% 17.8621 17.8241 25.2762 24.9447 10.5044 31.1003 23.6972 23.8438 17.4331 18.4994 Geometric_Mean 0.0535 0.0534 0.0016 0.0017 0.6570 0.5549 0.0794 0.0793 1.1690 1.1756

Variable &a1 &a1 &a2 &a2 F1 F1 &10 &10 V V

Diagnostic table for the model ith 2 sites of absorption"

Formu lation A A A A A A A A A A B B B B B B B B B B Item A?C CORR_(OBS,PR !" CSS !F S SBC SSR #CSS #SSR #T_CORR_(OBS,PR !" A?C CORR_(OBS,PR !" CSS !F S SBC SSR #CSS #SSR #T_CORR_(OBS,PR !" N 12 12 12 12 12 12 12 12 12 12 12 12 12 12 12 12 12 12 12 12 Mean 442.3582 0.9939 478.1272 9.0000 0.5761 439.1629 5.9788 478.1272 5.9788 0.9939 445.4551 0.9937 378.1471 9.0000 0.4388 442.2598 3.5768 378.1471 3.5768 0.9937 SD 90.9333 0.0064 64.4449 0.0000 0.6022 90.9333 6.3681 64.4449 6.3681 0.0064 85.6979 0.0066 174.8035 0.0000 0.4728 85.6979 4.2168 174.8035 4.2168 0.0066 Min 4137.2456 0.9877 414.1760 9.0000 0.0017 4134.0503 0.0000 414.1760 0.0000 0.9877 4132.5252 0.9867 115.9450 9.0000 0.0021 4129.3299 0.0000 115.9450 0.0000 0.9867 Max 47.9836 1.0000 627.5470 9.0000 1.2943 51.1788 15.0761 627.5470 15.0761 1.0000 44.3703 1.0000 670.3750 9.0000 1.1376 47.5656 11.6466 670.3750 11.6466 1.0000

Corrected sum of squared observations (CSS), weighted corrected sum of squared observations (WCSS), sum of squared residuals (SSR), weighted sum of squared 1/

residuals (WSSR), estimate of residual standard deviation (S) and degrees of freedom (DF), the correlation between observed Y and redicted Y, the weighted correlation, and two measures of goodness of fit! the "#ai#e $nformation Criterion ("$C) and Schwar% &a'esian Criterion (S&C)( Com-arison of t2e two concurrent models8

&omparison of the 2 modelling approaches *classical monocompartmental model against our user model including t o phases of absorption+ can be done based on different arguments *goodness of fitting# plot of residuals# and the consideration of the $'ai'e %nformation &riterion *$%&++. $%& is a measure of goodness of fit based on maximum li'elihood. When comparing several models for a given data set# the model associated ith the smallest *C is regarded as giving the best fit. $%& is appropriate only for comparing models that use the same eighting scheme *no eighing here+. *C / 9N log 0:!##1 ; (P for modelling in WinNonlin. N is the number of observations ith positive eight. W?EE is the eighted residual sum of s!uares. ; is the number of parameters that has been estimated.

Muestions"

What are your conclusionsN Why is the terminal half3life estimated to be J.-1h */.FL31'1/+ hile the classical monocompartmental model gives an estimate of 1F31L hN %s it a flip3flop phenomenonN

11

Deconvolution analysis
Overview Deconvolution is primarily used in ;5 to obtain the input function# i(e( the rate *0g1h or 0g.'g311h+ of drug entry into the central compartment. Deconvolution is useful to reveal in vivo drug release from a given pharmaceutical form or in our case the delivery from the 2 sites of absorption based on data for a 'no n drug input *generally an %8 administration+. "2e integral of t2e in-ut function gives t2e bioavailabilit. . Deconvolution based bioavailability estimation methods are a more general approach to estimate bioavailability than conventional methods since the former provide estimates of both the rate of systemic upta'e and the total systemic dose. Estimated rates of bioavailability may potentially be used to" gain insights into the mechanisms of upta'e enable better predictions of bioavailability under varying conditions Technically# deconvolution is a techni!ue that can be used to estimate an in-ut function& given the corresponding in-ut9res-onse function 0our data to anal.<e1 and the im-ulse9res-onse function *concentration profile follo ing an %8 bolus dose+ for the system. Deconvolution is t2e inverse of convolution .

Literally# if e 'no the disposition curve from a unit %8 dose *bolus+ and the rate of input of the drug into the plasma *input rate function+ then e can calculate the plasma disposition curve from that input by convolution.

12

Deconvolution is the inverse problem# i(e( to determine the input function from the plasma disposition curve given the unit disposition function# i.e. to determine the input function f*t+# given the unit impulse response &O*t+ and the input response &*t+. 5ey assumptions for deconvolution are" 1+ linearity" f*D1>D2+ 7 f*D1+ > f*D2+ 2+ time invariance" f*D+ has the same shape no matter hen D is given. We ill illustrate these concepts ith our example" Apen WNL ith our data sheet. Eelect Deconvolution from the Tools menu"

The Deconvolution dialog appears!

13

Then drag the appropriate variables to the /ime and Concentration fields# drag formulation and animals to the Sort 0ariables field. $ separate analysis is done for every uni!ue combination of sort variable values. Enter the dosing units in the Dose units field *here 0g+. The dosing units are used only hen the input data set has units associated ith the Time and &oncentration data *ng1mL+. The administered dose as 1//0g1'g.

%n the Settings fields# select the number of exponential terms *N+ in the unit impulse response *here it is 1 because it is a mono3exponential model+ Enter the values for $ and alpha in the unit impulse response function *see belo + as described by

The %8 bolus *unit impulse response+ after a 1// 0g1'g bolus dose the follo ing e!uation"
C (t ) = 800 Exp (0.1 time)

With &*t+ in ng1mL thus $7J// ng1mL and '1/7/.1 per h ,or this deconvolution# '1/7/.1 e have to enter a scaled $# i(e( $11//0g7/.J and alpha is

Eelect the setting for Smoothing to be used. "utomatic should be selected if you ant the program to find the optimal value for the dispersion parameter delta. This is the default. Eelect the $nitial Change in Rate is 1ero chec' box to constrain the derivative of the estimated input rate to be Dero at the initial time *lag time+.

19

&lic' Calculate. Deconvolution generates a ne chart and Deconvolution generates three charts for each of the sorted 'eys.

or'boo'.

The Fitted Curve output plot depicts concentration data from the input data set plotted against time for animal P21# the curve sho s a shoulder in the ascending phase. This is the deconvoluted curve.

The next figure gives the $n ut Rate *0g1h+ plot depicting the rate of drug input against time for each profile. 8isual inspection of this figure sho s clearly the presence of t o pea's in the input rate describing a rapid and a slo absorption process.

1-

The next figure is the Cumulative $n ut plot that displays cumulative drug input *in 0g1h+ against time for animal 21. 8isual inspection of the curve clearly sho s that there is a biphasic absorption explaining the initial shoulder seen on the observed curve.

The deconvolution or'boo' contains one or'boo' ith different or'sheets. The values or'sheet includes values for Time# %nput ?ate# &umulative %nput and ,raction %nput. %nspection of the results for animal 1 sho s that the total amount of absorbed drug as 1/G 0g1'g and that -/: of the drug has been absorbed after a delay of 1-.J9h.

1F

%t should be stressed that all the animals do not sho animal 21.

such a nice profile as for

Conclusion
Deconvolution is a po erful tool to document a process of absorption. &onvolution is also extensively used in industry for in vitro2in vivo *%8%8+ extrapolations.

1G