Br J Clin Pharmacol 1998; 45: 229–239

Characteristics of indirect pharmacodynamic models and applications to

clinical drug responses
Amarnath Sharma1,2 & William J. Jusko1
1
Department of Pharmaceutics, School of Pharmacy, State University of New York at Buffalo, Buffalo, NY 14260 and 2Eli Lilly and Company,
Indianapolis, IN 46202, USA

This review describes four basic physiologic indirect pharmacodynamic response

(IDR) models which have been proposed to characterize the pharmacodynamics of
drugs that act by indirect mechanisms such as inhibition or stimulation of the
production or dissipation of factors controlling the measured effect. The principles
underlying IDR models and their response patterns are described. The applicability
of these basic IDR models to characterize pharmacodynamic responses of diverse
drugs such as inhibition of gastric acid secretion by nizatidine and stimulation of
MX protein synthesis by interferon a-2a is demonstrated. A list of other uses of
these models is provided. These models can be readily extended to accommodate
additional complexities such as nonstationary or circadian baselines, equilibration
delay, depletion or accumulation of a precursor pool, sigmoidicity, or other
mechanisms. Indirect response models which have a logical mechanistic basis account
for time-delays in many responses and are widely applicable in clinical pharmacology.
Keywords: pharmacodynamics, indirect response models

for their elaboration because of processes such as inhibition

Introduction
or stimulation of the production or dissipation of factors
In order to understand and predict the pharmacologic controlling the measured effects. In these cases, there is
behaviour of drugs, it is important to quantify the time always a mechanistic delay with or without an equilibration
course of pharmacodynamic responses in relation to the delay. The latter is primarily dependent upon the physico-
plasma drug concentrations. The selection of an appropriate chemical properties of the drug such as lipophilicity and the
model should, if possible, be based on the mechanism of actual site of action. Thus, IDR models may be appropriate
action of the drug and other facets of biological reality. In for many drugs that produce responses with a time lag
addition, mechanism-based physiologic models can be useful between biophase drug concentrations and the time course
for estimating inaccessible pharmacodynamic steps and of responses. The mechanism by which drugs cause indirect
parameters. Reversible pharmacodynamic effects of drugs responses was originally described by Ariens in 1964 [3]. He
can be broadly classified as direct and indirect responses. noted that ‘… A number of drugs induce effects not because
Direct responses are effects produced by drugs that act of an interaction with receptors related to the effector, but
immediately on measured variables, and are observed in on the strength of endogenous compounds. That is, via
relation to drug concentrations at the site of action (biophase). their protection or release.’ Warfarin is a classic example of
A static function such as a linear model, an Emax model, or a a drug with an indirect mechanism of action; it inhibits the
sigmoid Emax model can often be applied to characterize the synthesis of prothrombin complex activity to produce an
relationship between plasma concentrations and pharmacodyn- anticoagulant effect [4].
amic profiles of drugs. However, for some drugs, responses This report reviews basic IDR models that may be
take time for their development and are not apparently related suitable for describing the effects of drugs produced by
to plasma concentrations because of an equilibration delay indirect mechanisms. We have also demonstrated the effect
between the plasma compartment and biophase. In such cases, of dose on the response profiles of drugs for each of the
a hypothetical effect-compartment (‘link’) type approach can IDR models, and described experimental designs and
sometimes account for the equilibration delay [1]. d- procedures to estimate the pharmacodynamic parameters of
tubocurarine is a drug that appears to produce its pharmacody- all four models. Furthermore, methods for experimental
namic response by a direct mechanism and exhibits onset and identification of an appropriate IDR model are discussed
offset equilibration delays [2]. and some complexities, which require extension of these
In contrast, for indirect pharmacodynamic responses, there models, are described. Several studies which applied these
is a lag time for development of a response even after the models to clinical response data are reviewed.
drug reaches the site of action. Indirect responses need time
Indirect pharmacodynamic response model
Correspondence: Dr William J. Jusko, 565 Hochstetter Hall, Dept. of Pharmaceutics,
School of Pharmacy, State University of New York at Buffalo, Buffalo, New York A fundamental physiologic model for drugs that produce
14260, USA. pharmacologic effects by indirect mechanisms is shown in

© 1998 Blackwell Science Ltd 229

A. Sharma & W. J. Jusko

Figure 1. In this model, the precursor is converted into or may be an observed response which is directly and
secreted as the response variable or mediator which, in turn, immediately proportional to the concentration of a mediator.
o
is then removed from the system. Drugs with indirect It is assumed that kin and kout fully account for production
actions can produce their effects by acting on one or more and loss of the response. The equations for IDR cannot be
of the indicated steps shown in this model. The agent can fully integrated to explicit functions and thus must be
cause inhibition or stimulation of the synthesis or secretion handled as differential equations and solved by numerical
of the response variable or of its removal, or of processes integration algorithms.
leading to the production of precursor. Models I and II represent processes that inhibit the factors
controlling drug response (Figure 2) where inhibition pro-
cesses operate according to the inhibition function I(t),
Basic indirect response models
which can be the sigmoid function:
Four basic IDR models (Figure 2) were proposed and Imax.Cp
characterized which may be applicable to describe the I(t)=1− (2)
pharmacodynamic responses of drugs with indirect mechan- IC50+Cp
isms of action [5, 6]. The basic premise of these IDR models where the Imax and IC50 value and the plasma drug
is that the measured response (R) to a drug is produced by concentrations (Cp ) control the dynamic function. Imax is
an indirect mechanism. The rate of change of the response the maximum fractional ability of the drug to affect the kin
over time with no drug present can be described as: or kout processes and is always less than or equal to unity,
dR o
i.e., 0<Imax≤1. IC50 is the drug concentration that
=kin−kout.R (1) produces 50% of maximum inhibition achieved at the effect
dt site. The Cp values are determined by the pharmacokinetics
where koin represents the apparent zero-order rate constant of the drug. Any appropriate pharmacokinetic function can
for production of the response, kout defines the first-order be used to obtain the Cp values.
rate constant for loss of the response, and R is assumed to Model I describes drug response resulting from inhibition
o of the factors regulating the production of the response
be stationary with an initial value of Ro (=kin/kout ). The o
response variable, R, can be a directly measured entity or it variable (kin). Inhibition with an Imax and IC50 is considered
o
to act on kin according to:
Production Synthesis Removal dR o
Precursor Response = kin.I(t)−kout.R (3)
Secretion variable dt
In Model I, when plasma drug concentrations are high
(i.e., Cp&IC50 ), IC50 becomes insignificant and the Cp
values cancel out, and if Imax=1, then I(t)=0. At high drug
o
Figure 1 Basic scheme for indirect pharmacodynamic responses. concentrations kin times zero is zero; thus, there is complete
The response variable is affected by drugs which can alter the blockage of the production of the response variable. Later,
formation of precursor, and production or dissipation of response when plasma concentrations decline to low values, Cp
variable. Drugs can inhibit or stimulate any of these processes. would be below IC50 (Cp%IC50 ). When that happens, drug

k 0in k out
Response
(R)

Model: I III II IV

I. Inhibition – k 0in III. Stimulation – k 0in

dR = k 0 .(1– Imax . Cp ) – k .R dR = k 0 .(1+ Smax . Cp ) – k .R

in out in out
dt IC50 + Cp dt SC50 + Cp

0 < Imax ≤ 1 Smax > 0

II. Inhibition – k out IV. Stimulation – k out

dR = k 0 – k (1– Imax . Cp ).R dR = k 0 – k (1+ Smax . Cp ).R

in out in out
dt IC50 + Cp dt SC50 + Cp

0 < Imax ≤ 1 Smax > 0

Figure 2 Four basic indirect response models representing processes that inhibit or stimulate the factors controlling drug response.

230 © 1998 Blackwell Science Ltd Br J Clin Pharmacol, 45, 229–239

 Indirect response models

o
effect would be zero and I(t)=1, meaning kin would be
Assumptions in the proposed models
back to its full value. The system returns toward its baseline
o
as kin will be refilling the pharmacodynamic compartment In these four IDR models, it is assumed that the response
with the response variable. to the drug is independent of the amount of the precursor,
Model II describes drug response resulting from inhibition i.e., the precursor pool is very large and thus is not
o
of the factors governing the dissipation of response variable significantly affected by drug action. Therefore, kin is an
(kout ). Inhibition with an IC50 is considered to act on kout apparent zero-order rate constant. An assumption is also
o
according to: made that the baseline of the system is stationary, and kin
and kout fully account for production and loss of response.
dR o
=kin−kout.I(t).R (4) The drug response (R) begins at a baseline value (Ro ),
dt changes with time following drug administration, and after
o
dissipation of drug effects returns to Ro (i.e., kin=kout.Ro ).
where I(t) is the function shown as Equation 2. Another initial assumption is that drug effects correlate
Models III and IV represent processes that stimulate the directly with plasma drug concentrations as the biophase;
factors controlling drug response (Figure 2) where stimula- thus, functions such as I(t) or S(t) can be used to describe
tion processes operate according to: the intrinsic drug effect.
Smax.Cp
S(t)=1+ (5)
SC50+Cp Effects of dose

The SC50 value represents the drug concentration producing Since dose is the most readily manipulated variable in a
50% of the maximum stimulation achieved at the effect site. pharmacodynamic study, it is of interest to evaluate the
The value of Smax can be any number greater than zero. effect of dose on the properties of the response profiles for
Model III describes drug response resulting from stimula- each of the IDR models (See ref. 6 for details). The
tion of the factors regulating the production of the response simulations presented in Figure 3 utilize a monoexponential
o
variable (kin). Stimulation with SC50 is considered to act function to describe the pharmacokinetics of the drug:
o
on kin according to: D −kel.t
Cp= e (11)
dR o V
=kin.S(t )−kout.R (6)
dt where D=Dose, V =volume of distribution, and kel=
where Equation 5 provides S(t). elimination rate constant.
Model IV represents drug response resulting from The pharmacodynamic profiles of drugs that produce
stimulation of the factors controlling the dissipation of the responses by indirect mechanisms described by Models I to
response variable (kout ). Stimulation with SC50 is considered IV exhibit grossly similar patterns with changes in dose
to act on kout according to: (Figure 3). In all models an increase in dose causes the
maximum response (Rmax ), initial slope (SI ), and the area
dR o between the baseline and effect curve (ABEC) values to
=kin−kout.S(t).R (7) increase, and the time of occurrence of Rmax (tRmax ) to
dt
shift to later times. In addition, the ABEC and tRmax values
The area between the baseline and effect curve (ABEC) continue increasing in proportion to log dose. For Models
can be calculated as a summary parameter to characterize I and IV, Rmax exhibits a lower limiting value with dose,
the overall effect of drug. while for Model III it shows an upper limiting value with
dose. However, Rmax continues increasing nonlinearly with
ABEC=|Ro.tr−AUECo−tr| (8) log dose for Model II when Imax=1. If Imax<1, Rmax has
where Ro is the baseline value and AUEC is the area under an upper limit of Ro/(1−Imax ) with dose for Model II.
or over the response vs. time curve over the time interval In the four IDR models, there is a considerable lag time
of 0 to tr. The value of tr is assumed2. between the occurrence of peak plasma concentrations and
The ABEC summary parameter encompasses all of the the maximum response (Rmax ) as shown in Figure 3. This
determinants of drug action: pharmacokinetic (V, kel for a is due to continuation of drug effects (inhibition or
monoexponential function, Equation 11) and dynamic (IC50 stimulation) for as long as Cp>IC50 (or SC50 ). After the
or SC50, Imax or Smax ). For Models I and III, ABEC is: Rmax has been reached, the return to baseline is governed
o
by both kin and drug elimination (kel ). Therefore, even after

A B
Imax D/V drug concentrations have declined below IC50 (or SC50 ),
ABEC=Ro ln 1+ Model I (9)
kel IC50 the response remains for some time owing to the time
o
needed for the system to regain equilibrium (when kin=

A B
Smax D/V kout.Ro ).
ABEC=Ro ln 1+ Model III (10)
kel SC50
Model identification
while it has a more complex relationship for Models II and
IV [6]. These equations predict an initial threshold then a To assign an appropriate model to pharmacodynamic data,
linear ABEC vs log dose profile for all drug responses. knowledge of the mechanism of drug action is essential.

© 1998 Blackwell Science Ltd Br J Clin Pharmacol, 45, 229–239 231

A. Sharma & W. J. Jusko

Model I Model II 1000

30 1000
100
100
10
20 10 80 10 10
00 00

100 10 10 1
0 60 100 0
10 10 10
Pharmacological effect

1000 40

Drug concentration
10
0

60
Model III Model IV 1000
1000 30
100
50
100 10 25 10 10
00 00 10

40 10 10 1
0 20 0
100
10 10
10 1000
15
30
0 10 20 30 0 10 20 30
Time
Figure 3 Simulations of the pharmacodynamic response variables (solid lines) with respect to time after single i.v. bolus doses. Simulated
pharmacokinetic profiles at the corresponding doses using Equation 11 are shown by dashed lines. The indicated values of doses (10 to
−1
1000) were used to invoke pharmacodynamic responses for the four indirect response models. Values of V =90 l, kel=0.3 h , Imax=
−1 −1 o −1 −1
1.0, Smax=1.0, IC50=100 ng ml , SC50=100 ng ml , kin=9 unit h , kout=0.3 h , and Ro=30 were used for simulations.

However, in cases where the mechanism is not known, Rmax=Ro/(1−Imax ) (Model II)
experimental designs that can discriminate between models
Rmax=Ro(1+Smax ) (Model III)
would be very useful. Furthermore, these study designs can
also be used to validate the mechanism of action for drugs Rmax=Ro/(1+Smax ) (Model IV)
with known actions which, in turn, reinforces the biologic
Therefore, for drugs with incomplete information on the
plausibility of the proposed model.
mechanism of action, an appropriate IDR model can still
Among these four IDR models, response profiles show
be chosen by comparison of experimental Rmax values
downward trends for Models I and IV, and upward behaviour
obtained at larger doses with estimated Rmax values for
for Models II and III (Figure 3). Models I and IV may be
models with similar downward or upward responses.
applicable if the drug causes a decrease in the response
However, with Models II and III, it is difficult to know
variable from its baseline value. Similarly, Models II and III
when the dose is sufficiently high to produce maximal
may appear to be suitable for describing data if the response
responses (Figure 3). With Models I and IV, the range of
variable increases from its baseline value. Thus, it is useful
responses has a clearer lower limit (e.g. zero).
to have method(s) that can be applied for experimental
In an infusion study, the length of infusion should be
identification of an appropriate indirect response model.
sufficiently long not only to produce steady-state pharmaco-
Two such methods have been proposed to assign an
kinetics but also steady-state conditions in the pharmacodyn-
appropriate IDR model for drug response (6): (i) single i.v.
amic system. More than one administration rate is needed
dose study design; and (ii) a steady-state i.v. infusion
and the time of infusion should based on the kout value. For
study design.
instance, if kout is small, a longer infusion time is required,
In a single i.v. dose study, administration at more than
and vice-versa.
one dose level is required with one dose level high enough
In Models I and III, the tRmax would remain constant
to produce either full inhibition or stimulation of the system.
with the change in the infusion rate because the drug affects
Depending on the nature of drug action (stimulation or o
kin for these two IDR models (inhibits for Model I and
inhibition), the experimental data obtained can be charac- o
stimulates for Model III), and kin has no influence on the
terized using two of four IDR models to calculate
time required by the pharmacodynamic system to reach
pharmacodynamic parameters such as Imax or Smax and IC50
steady-state under continuous drug infusion. However, in
or SC50. These parameters, in turn, can be used to estimate
Models II and IV, the tRmax would change with the infusion
the maximum responses (Rmax ) at large doses (Dose2)
rate because the drug affects kout for these models (inhibits
according to:
for Model II and stimulates for Model IV). The kout value
Rmax=Ro(1−Imax ) (Model I) influences the time required for the response to reach

232 © 1998 Blackwell Science Ltd Br J Clin Pharmacol, 45, 229–239

 Indirect response models

steady-state during continuous drug infusion. For instance, H2-receptor antagonist: Inhibition of gastric secretion Nizatidine
in Model II where a drug produces its response by inhibition is an H2-receptor antagonist that blocks the gastric acid
of kout, the tRmax would shift to later times because the secretion stimulated by histamine and other H2-receptor
decrease in kout due to the drug would result in an increased agonists. Callaghan et al. [7] evaluated effects of nizatidine
time required by the system to reach steady-state during at four dose levels on gastric acid secretion in which one
continuous infusion of the drug. Similarly, in Model IV, dose (250 mg ) was high enough to produce full inhibition
the tRmax would shift to earlier times because an increased of the system (gastric acid secretion). In this study, healthy
kout value produces the opposite behavior. Thus, one can subjects received 5 min i.v. infusions of nizatidine at 25, 50,
determine which IDR model is suitable for a drug by 100 or 250 mg dose levels. Gastric acid secretion was
−1 −1
comparing experimental tRmax values obtained at two steady- induced by i.v. infusion of pentagastrin at 2 mg kg h
state infusion rates. for 4.25 h starting 45 min before the nizatidine dose. Their
data were reanalyzed using IDR Model I and are shown in
Figure 4. In our proposed model the parietal cells in the
stomach are assumed to produce gastric acid at a constant
Estimation of the pharmacodynamic parameters o
rate (kin) under stimulation of H2-receptor agonists such as
The study design is critical for estimating the pharmacodyn- histamine and the acid is removed from the stomach by
amic parameters of an IDR model. The following design is gastric emptying via a first-order rate process (kout ).
needed to elucidate fully the pharmacodynamic parameters Nizatidine reduces the amount of gastric acid secretion by
of IDR models: (i) administration of two or more bolus i.v. inhibiting the interaction of the H2-receptor agonist with
doses of drug; (ii) one of the dose levels should be H2-receptors. As the drug is eliminated from the body,
sufficiently high to produce either full inhibition or gastric acid secretion returns gradually to baseline. The
stimulation of the system; (iii) determination of the baseline pharmacodynamic data were fitted to Model I (Figure 4).
value (Ro ) and behaviour of the system. The fitted line shows that the proposed model adequately
The experimental values of the baseline (Ro ), the initial described the data. The pharmacodynamic parameters
o −2 −1
slope (SI ), and maximum drug response (Rmax ) obtained at estimated were: kin=184 mEq h , kout=5.6 h , Imax=
−1
large doses can be first used to estimate maximum inhibitory 1.0, and IC50=164 ng ml . The coefficients of variation
(Imax ) or stimulatory (Smax ) factors as listed in the first row (CV) of these parameters were reasonable, ranging from 2
of Table 1. The estimated Imax (or Smax ) and the initial slope to 6%.
(SI ) of the response versus time curve can then be used to Modeling IDR data such as this was and should be a
o
calculate kin values as indicated by the second row in two-stage process. The plasma drug concentration (or other
o
Table 1. Then kout is obtained from kin/Ro for all models biophase) data are first fitted to produce a suitable descriptive
as indicated by the third row in Table 1. equation. This serves as a forcing function in Equation 2 or
The experimental value of the time to reach maximum 5 to produce the inhibition or stimulation values. Then the
response (tRmax ) and Rmax obtained at a given dose can be dynamic data are fitted by nonlinear regression analysis
o o
used to interpolate the plasma drug concentration at tRmax where either kin or kout (NB kin=kout.Ro ), IC50 (or SC50 ),
(CRmax ) from the pharmacokinetic data. Very roughly, the and Imax (or Smax ) are sought as regression parameters.
IC50 or SC50 is the drug concentration at the time of Rmax Ideally, data from the baseline and several dose levels should
for a drug causing a moderate change in response. More be fitted simultaneously in order to generate one set of
exact IC50 or SC50 values can be estimated as shown in the parameters to describe the effects of the drug. When one
fourth row of Table 1. dose level is fitted alone, there may be inadequate
These equations provide a rationale for examining IDR discrimination of the model parameters. A Hill coefficient
profiles such as shown in Figure 3 and using graphical means (c) may be added once it is determined that lack of
of recovering the major parameters governing the overall coefficient (c=1) is insufficient to characterize the data.
system. However, because of the non-linearity and time-
dependence of IDR, final estimation of parameters should Induction of MX protein synthesis Interferon a-2a induces
be based on nonlinear least-squares regression analysis using the synthesis of MX protein, a factor that interferes with
computer programs (WinNonlin, Adapt II, P-Pharm, viral replication. Nieforth et al. [8] used IDR Model III to
NonMem, and others) capable of performing numerical characterize the pharmacodynamics of interferon a-2a after
integration of differential equations. subcutaneous administration. It was assumed that MX
o
protein is produced as a zero-order process (kin), a first-
order process (kout ) accounts for its degradation, and
o
interferon a-2a stimulates kin (Figure 5). There occurs an
Clinical examples
initial dose-proportional increase in MX protein in plasma
The applicability of IDR models has been demonstrated for and, as the drug is eliminated from the body, MX protein
diverse clinical pharmacodynamic responses such as those concentrations return to the baseline. Response data for four
listed in Table 2. More specific responses that can be doses (3, 6, 9, and 18 MIU) of interferon a-2a [8] were
characterized with IDR models are demonstrated below. simultaneously fitted to Model III (Figure 5). The pharmaco-
−1 −1
For the examples given below, we have used published data dynamic parameters estimated were: kin=0.65 ng ml h ,
−1 −1
for the pharmacokinetics and pharmacodynamics of drugs kout=0.013 h , Smax=28.5, and SC50=29.9 units ml .
with indirect responses, and data were reanalyzed with the The coefficients of variation (CV) of these parameters ranged
most plausible of the IDR models shown in Figure 2. from 7 to 34%.

© 1998 Blackwell Science Ltd Br J Clin Pharmacol, 45, 229–239 233

234

A. Sharma & W. J. Jusko

Table 1 Use of indirect response profiles to estimate model parameters. Measured characteristics: Ro, SI, Rmax, CRmax at large dosesa.

Parameter Model I Model II Model III Model IV

Imax or Smax (Ro−Rmax )/Ro (Rmax−Ro )/Rmax (Rmax−Ro )/Ro (Ro−Rmax )/Rmax
koin −SI/Imax SI/Imax SI/Smax −SI/Smax
kout koin/Ro koin/Ro koin/Ro koin/Ro
IC50 or SC50 CRmax.(Rmax−(1−lmax ).Ro) CRmax.(Ro−(1−lmax ).Rmax ) CRmax.(Ro.(1+Smax)−Rmax ) CRmax.(Rmax(1+Smax )−Ro )
(Ro−Rmax ) (Rmax−Ro ) (Rmax−Ro ) (Ro−Rmax )

a
Symbols are defined in Glossary.

Table 2 Drugs which produce indirect pharmacodynamic responses.

Drug Mechanism of action Measured PD response Indirect model Reference

Warfarin Vitamin K epoxide reductase inhibitor Decreased prothrombin complex activity I 4

Nizatidine H2-receptor antagonist Decreased secretion of gastric acid I 7
Tolrestat Aldose reductase inhibitor Decreased RBC sorbitol I 17
Methylprednisolone Inhibits corticotropin and CRF Decreased cortisol secretion I 9, 30
Fluticasone propionate Inhibits corticotropin and CRF Decreased cortisol secretion I 11
© 1998 Blackwell Science Ltd Br J Clin Pharmacol, 45, 229–239

Prednisolone Inhibits tissue efflux Decreased basophil trafficking I 30

Prednisolone Inhibits tissue efflux Decreased T-cell trafficking I 27, 30
Prednisolone Altered tissue efflux Increased NK cells in blood I 31
Prednisolone Suppression of bone metabolism Decreased serum osteocalcin concentration I 32
Ibuprofen Inhibition of prostaglandin E2 Reduced fever I 28
DCN 203–922 Dopaminomimetic Decreased plasma prolactin concentration I 12
Pyridostigmine Cholinesterase inhibitor Increased muscle response II 30
Frusemide Inhibitor of Na+-K+-2Cl− symport Increased urinary excretion of Na+, Cl− II 30
Cimetidine H2-receptor antagonist Increased prolactin release III 30
Interferon a-2a Induces MX protein synthesis Increased blood MX protein concentration III 8
Terbutaline b2-adrenergic receptor agonist Increased pulmonary air flow III 30
Terbutaline b2-adrenergic receptor agonist Hypokalaemia IV 30
 Indirect response models

k 0in k out
Gastric juice
(H+)
IC50

10000 40
Plasma nizatidine concentration (ng ml–1)

Gastric acid secretion (mEq h–1)

30
1000

20

100
10

10 0
0 1 2 3 4 0 1 2 3 4
Time (h)
Figure 4 Composite graph showing the pharmacodynamic model, pharmacokinetics of nizatidine, and gastric acid secretion profiles after
single i.v. administration of 25 ($), 50 (&), 100 (+) and 250 (,) mg nizatidine doses. Symbols are experimental data from Callaghan
et al. [7], and lines are simultaneous least-squares regression fitting of all data to Model I.

These studies also included a comparison of the effects of et al. [10] used a ramp function consisting of a linear
interferon a-2a and pegylated interferon seeking to determine decrease of cortisol secretion during the day and a shorter
whether the latter, because of its longer t1/2 (5.3 vs 11.9 h) linear increase at night as an alternative to the cosine
would be suitable for once-weekly dosing. Unfortunately, function to handle the circadian behaviour of cortisol. It
pegylation also produced a reduction in Smax (67 vs 39) appears to handle better the rapid rise in cortisol which
−1
while SC50 remained the same (60 units ml ) as for occurs in the early morning hours. Rohatagi et al. have
interferon a-2a. Simulations demonstrated that an extension modeled the cortisol suppressant effects of fluticasone
of the dosing interval was inadequate to maintain desired propionate using an IDR model with the ramp function
plasma concentrations of MX protein, thus leading to a described above [11]. The mean IC50 values obtained in
−1
company decision not to further develop this pegylated healthy volunteers were 0.171, 0.105 and 0.126 ng ml for
form of the drug. This was an excellent demonstration of 0.5, 1, and 2 mg doses. A good correlation between in vivo
the value of PK/PD modeling at a critical stage in drug pharmacodynamic response (decrease in cortisol levels) and
development which saved considerable time and expense of in vitro receptor binding affinity of fluticasone propionate
further clinical trials. was observed.
Francheteau et al. [12] characterized the prolactin sup-
pressant effects of a dopaminomimetic drug DCN 203–922
Complexities and applications
using an IDR model which required both a link compart-
The four models reviewed in this article are the most basic ment and a ‘fluctuation model’ for the irregular rhythmic
of IDR models and can be extended to incorporate baseline of prolactin. The latter was handled using a spline
additional complexities in pharmacodynamic systems. Several fitting of the placebo response. They showed greater
interesting applications of more complex IDR models will reductions in plasma concentrations of prolactin when giving
be described. the drug as 2 mg at 6 hour intervals rather than a single
If the baseline is nonstationary such as for a circadian 6 mg dose. Therefore, diverse functions can be used easily
o o
rhythm, a cosine function can be used to describe kin [9]. to describe nonstationary baselines as part of kin or kout in
This approach has been used for assessing the effect of Equation 1 in application of IDR models for drug responses.
exogenous corticosteroids (methylprednisolone and predni- For some drugs, the production of the drug response may
solone) on plasma concentrations of cortisol [9]. Rohatagi be dependent on the amount of precursor, which may

© 1998 Blackwell Science Ltd Br J Clin Pharmacol, 45, 229–239 235

A. Sharma & W. J. Jusko

k 0in k out
MX protein

SC50

80
400
Serum INFa-2a concentration (units ml–1)

Doses, MIU
18

MX protein concentration (ng ml–1)

60
300

40
200
9

20 6
100
3

0
0
0 20 40 60 80 0 100 200 300
Time (h)

Figure 5 Composite graph showing the pharmacodynamic model, pharmacokinetics of interferon a-2a, and blood MX protein
stimulation profiles after a single subcutaneous administration of 3 (#), 6 (%), 9 ($) and 18 (&) MIU interferon a-2a doses. Symbols
are experimental data from Nieforth et al. [8], and lines are simultaneous least-squares regression fitting of all data to Model III. Note
difference in PK and PD time scales.

change significantly due to the drug. Ekblad & Licko [13] of a lag-time for tolerance and a rate constant for tolerance
suggested a general model similar to that shown in Figure 1 development.
to characterize transient responses for physiological sub- IDR modeling approaches were extended to handle
stances. Basic IDR models can be extended to accommodate multiple-dose kinetics and dynamics of drugs by Van
dependence of response on the amount of precursor by Griensven and co-workers [17]. Model I was used to
making kin a first-order rate constant [13–15]. These characterize the effects of tolrestat on RBC sorbitol levels
extended IDR models can describe tolerance and rebound after administration of single and multiple-doses. For
phenomena, and are appropriate for drugs that affect modeling of multiple-dose dynamics, a continuous PK/PD
precursor pools and/or the amount of endogenous com- relationship was generated by using the method of super-
pounds that are required for the production of drug response. positioning to extrapolate the effects from the initial to the
This approach was used to model the effects of the dopamine final tolrestat dose [17]. Like for many pharmacologic effects
antagonist remoxipride on stimulation of prolactin release it was found that responses or IC50 values related better to
into circulation upon repeated administration [14,15]. free than total drug in plasma. There was close agreement
−1
When a second dose of drug is given before the precursor of the in vivo IC50 for free tolrestat (12–16 ng ml ) with
−1
pool has refilled (refractory period), the response will values obtained in vitro (13 ng ml ) measuring sorbitol
be blunted and tolerance will appear to occur. A linear production in human erythrocytes.
rather than sigmoid stimulation model was suitable for The classic example and the first quantitative modeling of
relating remoxipride plasma concentrations to release of an IDR was the measurement of warfarin effects on
prolactin. prothrombin complex activity (PCA) [4]. Pitsiu et al. [18]
IDR models were also extended by addition of a applied both population and two-stage analyses to examine
‘modifier’ compartment where kout was altered in relation PCA and factor VII activities after dosing warfarin in a
to the measured response to cause development of furosem- group of 48 healthy volunteers. The IDR required both the
ide tolerance [16]. This extended model well characterized c shape factor and a lag time to allow for a delay in the
the diuretic and natriuretic response patterns after adminis- onset of response after warfarin administration. The c values
tration of serial doses of furosemide where responses were 1.03 for PCA and 2.63 for factor VII. The lag time
diminished with successive doses and allowed the estimation was about 8 h for both responses. The population model

236 © 1998 Blackwell Science Ltd Br J Clin Pharmacol, 45, 229–239

 Indirect response models

identified sources of both pharmacokinetic and pharmaco- require more complex indirect models and represent a major
dynamic variances with inter-individual differences in kout frontier in the future of PK/PD modeling.
and IC50 contributing most to variability in responses. A
time analysis also showed when each parameter was most Comprehensive pharmacokinetic/pharmacodynamic model
influential in the observed responses.
The equations used to describe the inhibitory (Equation A comprehensive scheme that accounts for the various steps
2) and stimulatory (Equation 5) effects of drugs need not be involved in a complex PK/PD system is depicted in Figure 6
limited to the sigmoid or Hill equations. In fact, it is [24]. This diagram shows four intermediary components
important to identify the biosensor process and to use an between drug in blood and the measured response, with
equation that best reflects the mechanism of action. A simple drug action divided into pharmacokinetic and pharmacodyn-
linear function served well for remoxipride stimulation of amic steps. Drug administration, distribution and elimination
prolactin release [14]. When two agonists are present (C1 are considered under pharmacokinetics. In most cases, the
and C2 ), it is possible to describe their additive effects as: drug response may be directly related to the plasma
C=C1+e.C2 where e is a potency ratio. This was done by concentration or free drug in plasma. In others, the biophase
Milad et al. [19] to describe the joint effects of cortisol and may be relevant. For example, in the case of frusemide and
methylprednisolone on lymphocyte trafficking in studies in most other diuretics, the biophase is urine [25, 26]. If drug
normal volunteers. concentrations in the biophase cannot be directly measured,
The IDR models can accommodate an irreversible then use of a link model may be appropriate in considering
component as well. The antiplatelet effect of aspirin was drug equilibration with a site separate from plasma.
modeled by Yamamoto et al. [20] with the relationship For pharmacodynamic modeling, it is essential to have
knowledge of the mechanism of action of the drug. In
dR Figure 6, the mechanism of drug action is represented as
=kin−kout.R−k.C.R
dt biosignal flux. The biosignal may be a mediator, which
undergoes altered synthesis or degradation due to drug
where the last term reflects a bimolecular drug-platelet action. The H indicates that the Hill equation might be
interaction. The modeling yielded a realistic responses and o
o
applied as an inhibitory or stimulatory function to the kin
regeneration rate constant (kin) for platelet production. A or kout processes as used in Equations 2 and 5. The
similar concept is embodied in some simple cancer and pharmacodynamic model based on biosignal flux may also
antibiotic chemotherapy models [21, 22] where cell repli- be applicable to receptor-mediated action. For instance, one
cation represent the kin process. can consider kin to be binding of drug to the receptor, kout
The greatest complexities come with transduction pro- the rate constant for drug dissociating from the receptor,
cesses. In the basic IDR models, the response is assumed to and bound receptor concentrations to be the biosignal.
be directly proportional, in a linear and immediate manner, These IDR models can adequately describe the receptor-
to a biosignal. The response and the biosignal become the mediated processes.
same. However, in some pharmacodynamic responses, there
may be signal transduction mechanisms involved which can
Conclusions
cause nonlinearities and time-delays between the biosignal
generated by the drug (bound receptor) and the response Modeling pharmacodynamic effects requires a broad appreci-
being measured. This occurs with corticosteroids [23]. ation of pharmacology, pharmacokinetics, and dynamics.
Following receptor binding, processes involving nuclear The mechanism of action of the drug should be known to
binding, synthesis of mRNA, and synthesis of effector define the appropriate pharmacodynamic model. It is
proteins are major steps in propagation of a response. The important to understand the biophase responsible for the
receptor-gene mediated actions of various steroids and the drug action, be it a real one or a hypothetical compartment.
numerous signal-transduction processes of various drugs will It is advisable to start with the simple models (observe

Drug
k in

Figure 6 A comprehensive scheme CP k eo Biosignal R

Ce
proposed to include the major
pharmacokinetic and pharmacodynamic
processes governing drug action including
distribution (rate constant, keo ) to a k out
biophase, inhibition or stimulation (H) of
the production (kin ) or dissipation (kout )
of a mediator or biosignal, and Disposition Biophase Biosensor Biosignal Trans- Response
transduction of the response (R). kinetics distribution process flux duction
(Reprinted with permission from Jusko
et al. J Pharmacokinet Biopharm 1995; 23:
Pharmacokinetics Pharmacodynamics
5–8).

© 1998 Blackwell Science Ltd Br J Clin Pharmacol, 45, 229–239 237

A. Sharma & W. J. Jusko

parsimony) and add complexities in an attempt to account 5 Dayneka NL, Garg V, Jusko WJ. Comparison of four basic
for all sources of variability in drug action. The IDR models models of indirect pharmacodynamic responses.
reviewed in this report are relevant to numerous drugs. J Pharmacokinet Biopharm 1993; 21: 457–478.
They require numerical integration and nonlinear regression 6 Sharma A, Jusko WJ. Characterization of four basic models of
programs for data analysis and simulation and should be indirect pharmacodynamic responses. J Pharmacokinet Biopharm
1996; 24: 611–634.
considered as a possible choice in modeling the time course
7 Callaghan JT, Bergstrom RF, Obermeyer BD, et al.
of drug action.
Intravenous nizatidine kinetics and acid suppression. Clin
Pharmacol Ther 1985; 37: 162–165.
8 Nieforth KA, Nadeau R, Patel IH, et al. Use of an indirect
Glossary
pharmacodynamic stimulation model of MX protein
induction to compare in vivo activity of interferon alfa-2a and
ABEC Area between the baseline and the response curve a polyethylene glycol-modified derivative in healthy subjects.
(0 to tr ) Clin Pharmacol Ther 1996; 59: 636–646.
Cp Plasma concentration of drug at any time 9 Lew KH, Ludwig EA, Milad MA, et al. Gender-based effects
CRmax Plasma concentration of drug at the time of on methylprednisolone pharmacokinetics and
pharmacodynamics. Clin Pharmacol Ther 1993; 54: 402–414.
maximal response
10 Rohatagi S, Bye A, Mackie AE, et al. Mathematical modeling
D Dose of drug
of cortisol circadian rhythm and cortisol suppression. Eur
I(t) Inhibitory function J Pharm Sci 1996; 4: 341–350.
IC50 Drug concentration producing 50% of maximum 11 Rohatagi S, Bye A, Falcoz C, et al. Dynamic modeling of
inhibition cortisol reduction after inhaled administration of fluticasone
Imax Maximum inhibitory factor attributed to drug propionate. J Clin Pharmacol 1996; 36: 938–941.
(0<Imax≤1) 12 Francheteau P, Steimer, J-L, Dubray C, et al. Mathematical
IDR Indirect pharmacodynamic response model for in vivo pharmacodynamics integrating fluctuation of
kel First-order rate constant for drug elimination the response: Application to the prolactin suppressant effect of
o
kin Apparent zero-order rate constant for production the dopaminomimetic drug DCN 203–922. J Pharmacokinet
of drug response Biopharm 1991; 19: 287–309.
kout First-order rate constant for loss of drug response 13 Ekblad EBM, Licko V. A model eliciting transient responses.
R Response variable Am J Physiol 1984; 246: R114–R121.
Rmax Maximal response 14 Movin-Osswald G, Hammarlund-Udenaes M. Prolactin
release after remoxipride by an integrated pharmacokinetic-
Ro Baseline response prior to drug administration
pharmacodynamic model with intra- and interindividual
SC50 Drug concentration producing 50% of maximum
aspects. J Pharmacol Exp Ther 1995; 274: 921–927.
stimulation
15 Sharma A, Jusko WJ. Indirect pharmacodynamic models for
Smax Maximum stimulatory factor attributed to drug tolerance and rebound phenomena. IBC Library Series 1997;
(Smax>0) Southborough, MA.
SI Initial slope of the response versus time curve 16 Wakelkamp M, Alvan G, Gabrielsson J, et al.
S(t) Stimulatory function Pharmacodynamic modeling of furosemide tolerance after
t Time after drug administration multiple intravenous administration. Clin Pharmacol Ther 1996;
tr Time when response returns to baseline 60: 75–88.
tRmax Time to reach maximum response following drug 17 Van Griensven JMT, Jusko WJ, Lemkes HHPJ, et al. Tolrestat
administration pharmacokinetic and pharmacodynamic effects on red blood
cell sorbitol levels in normal volunteers and patients with
insulin-dependent diabetes. Clin Pharmacol Ther 1995; 58:
This work is supported in part by Grant No. 24211 from 631–640.
the National Institute of General Medical Sciences, National 18 Pitsiu M, Parker EM, Aarons L, et al. Population
Institutes of Health. pharmacokinetics and pharmacodynamics of warfarin in
healthy young adults. Eur J Pharm Sci 1993; 1: 151–157.
19 Milad MA, Ludwig EA, Anne S, et al. Pharmacodynamic
References model for joint exogenous and endogenous corticosteroid
1 Holford NHG, Sheiner LB. Understanding the dose-effect suppression of lymphocyte trafficking. J Pharmacokinet
relationship: Clinical application of pharmacokinetic- Biopharm 1994; 22: 469–480.
pharmacodynamic models. Clin Pharmacokin 1981; 6: 20 Yamamoto K, Abe M, Katashima M, et al. Pharmacodynamic
429–453. analysis of antiplatelet effect of aspirin in the literature-
2 Sheiner LB, Stanski DR, Vozeh S, et al. Simultaneous modeling based on inhibition of cyclooxygenase in the
modeling of pharmacokinetics and pharmacodynamics: platelet and the vessel wall endothelium. Jpn J Hosp Pharm
Application to d-tubocurarine. Clin Pharmacol Ther 1979; 25: 1996; 22: 133–141.
358–371. 21 Jusko WJ. Pharmacodynamics of chemotherapeutic effects:
3 Ariens EJ. In Molecular Pharmacology: The mode of action of Dose-time-response relationships for phase-nonspecific agents.
biologically active compounds. Vol. 1, ed Ariens EJ, New York: J Pharm Sci 1971; 60: 892–895.
Academic Press, 1964. 22 Zhi J, Nightingale CH, Quintilliani R. Microbial
4 Nagashima R, O’Reilly RA, Levy G. Kinetics of pharmacodynamics of piperacillin in neutropenic mice of
pharmacologic effects in man: the anticoagulant action of systemic infection due to pseudomonas aeruginosa.
warfarin. Clin Pharmacol Ther 1969; 10: 22–35. J Pharmacokinet Biopharm 1988; 16: 355–375.

238 © 1998 Blackwell Science Ltd Br J Clin Pharmacol, 45, 229–239

 Indirect response models

23 Xu Z-X, Sun Y-N, DuBois DC, et al. Third-generation 28 Garg V, Jusko WJ. Pharmacodynamic modeling of
model for corticosteroid pharmacodynamics: Roles of nonsteroidal anti-inflammatory agents: antipyretic effect of
glucocorticoid receptor mRNA and tyrosine aminotransferase ibuprofen. Clin Pharmacol Ther 1994; 55: 87–88.
mRNA in rat liver. J Pharmacokinet Biopharm 1995; 23: 29 Ko HC, Jusko WJ. Pharmacodynamic modeling of finasteride,
163–181. a 5a-reductase inhibitor. Pharmacotherapy 1995; 15: 509–511.
24 Jusko WJ, Ko HC, Ebling WF. Convergence of direct and 30 Jusko WJ, Ko HC. Physiologic indirect response models
indirect pharmacodynamic response models. J Pharmacokinet characterize diverse types of pharmacodynamic effects. Clin
Pharmacol Ther 1994; 56: 406–419.
Biopharm 1995; 23: 5–8.
31 Xu Z-X, Lee M-J, Blum RA, Jusko WJ. Pharmacodynamic
25 Kaojarern S, Day B, Brater DC. The time course of delivery
modeling of prednisolone effects on natural killer cell
of furosemide into urine: an independent determinant of trafficking. Pharm Res 1994; 11: 848–854.
overall response. Kidney Int 1982; 22: 69–74. 32 Wald JA, Jusko WJ. Corticosteroid pharmacodynamic
26 Brater DC. Pharmacodynamic consideration in use of modeling: Osteocalcin suppression by prednisolone. Pharm Res
diuretics. Ann Rev Pharmacol Toxicol 1983; 23: 45–62. 1992; 9: 1096–1098.
27 Fisher LE, Ludwig EA, Jusko WJ. Pharmacoimmunodynamics
of methylprednisolone: trafficking of helper T lymphocytes. (Received 10 July 1997,
J Pharmacokinet Biopharm 1992; 20: 319–331. accepted 22 October 1997)

© 1998 Blackwell Science Ltd Br J Clin Pharmacol, 45, 229–239 239