1994 IEEE TRANSACTIONS ON BIOMEDICAL ENGINEERING, VOL. 52, NO.
12, DECEMBER 2005
Comprehensive Pharmacokinetic Model of Insulin
Glargine and Other Insulin Formulations
C. Tarín, E. Teufel, J. Picó, Member, IEEE, J. Bondia, and H.-J. Pfleiderer, Fellow, IEEE
Abstract—In this paper, a comprehensive pharmacokinetic
model for different insulin formulations including insulin Glargine
is developed based on the model proposed by Trajanoski et al.
(1993). Current models show limitations for insulin Glargine due
to the appearance of an uncharacteristic peak in the concen-
tration-time evolution of plasma insulin that does not coincide
with real experimental data. This important limitation has been
solved in this paper by introducing a new virtual insulin state
called the bound state, in addition to the dimeric and hexameric
ones. Trying to describe the retarded action of insulin Glargine,
the modeling idea behind this approach is that immediately after
the subcutaneous injection all the insulin resides in the bound
state, and only then small amounts of insulin in the hexameric
form disengage from the bound state. For the model evaluation
different simulation results are compared. Using experimental
data published by Lepore et al. (2000), the developed model turned
out to be capable of at least qualitatively predicting the concen-
tration-time profile of plasma insulin. Both exogenous insulin flow
simulations and spatial diffusion simulations show the plausibility
and correct implementation of the derived model. Considering all
these simulation results, the here presented new pharmacokinetic
model demonstrates to be able to reproduce real patient behavior
simulating even complete insulin regimes including long-acting,
intermediate and short-acting insulin formulations.
Index Terms—Diabetes care, discretization, infusion process, in-
sulin analogues, pharmacokinetic modeling.
I. INTRODUCTION
I NSULIN therapy regimes are the basis of treatment for
type-1 diabetes. A type-1 diabetic patient injects himself
insulin subcutaneously three to four times a day. After the
subcutaneous injection, insulin is absorbed gradually by the
body and transported toward the insulin-sensitive cells, where
it develops its glucose lowering action. There exist many
different insulin formulations. Originally, only animal insulin
has been used to treat type-1 diabetes mellitus. In the 1980s,
synthetically produced insulin with the same structure as
human insulin became available (e.g. Actrapid, Protaphane,
Semilente). Within the last ten years, a new category of insulin
Manuscript received February 18, 2004; revised April 10, 2005. This work
was supported in part by the Deutsche Forschungsgemeinschaft under Grant
DFG—Kennung 248826, in part by the Spanish government under Grant DPI-
2004-07167-C02-01, and in part by the European Union through FEDER funds.
Asterisk indicates corresponding author.
C. Tarín and J. Bondia are with the Department of Systems Engineering
and Control (DISA), Technical University of Valencia (UPV), 46022 Valencia,
Spain.
E. Teufel and H.-J. Pfleiderer are with the Microelectronics Department, Uni-
versity of Ulm, D-89081 Ulm, Germany.
*J. Picó is with the Department of Systems Engineering and Control (DISA),
Technical University of Valencia (UPV), Cami de Vera 14, 46022 Valencia,
Spain (e-mail: jpico@aii.upv.es).
Digital Object Identifier 10.1109/TBME.2005.857681
0018-9294/$20.00 © 2005 IEEE
became popular: the insulin analogues (e.g. Lispro, Humalog,
NovoRapid, Lantus). By means of genetical manipulation, the
human insulin molecule was altered to produce more rapid- and
short-acting as well as slow- and long-acting insulin analogues.
As for modeling the absorption process of subcutaneously
injected insulin, various models have already been developed,
as reported in [3]. Nevertheless, so far no comprehensive model
covering all insulin classes does exist. More specifically, there
is no model describing the new insulin Glargine (Lantus). With
this new insulin analogue having been developed as reported in
[4], [5] and [6], the shortcomings of previous intermediate and
long-acting insulins such as Neutral Protamin Hagedorn (NPH)
have been overcome. In contrast to NPH, insulin Glargine is
able to allow once-daily dosing without showing a pronounced
peak as it is designed for the prolongation of its time action.
This has been achieved by shifting the isoelectric point of the
insulin molecule from pH 5.4 toward more neutral by adding
positively charged amino acids to make the insulin less soluble
at the neutral pH of the injection site as described in [2] and
[7]. Current models show limitations for insulin Glargine due to
the appearance of an uncharacteristic peak in the concentration-
time evolution of plasma insulin that does not coincide with real
experimental data.
In this paper, we extend the model originally proposed in
[8] and later simplified and refined by Trajanoski [1] to cover
Glargine with its peakless time-action profile as well as other
insulin preparations. With this single model, the absorption of
insulin can be described focusing on the subcutaneous route of
administration. Additionally, the spatial diffusion at the injec-
tion site can be simulated.
Commonly, an insulin with a delayed action is needed to
cope with the basal requirement of the diabetic patient and a
short-time acting insulin is used to cope with the bolus meal
episodes. This basal-bolus concept is the cornerstone of any
kind of intensified insulin treatment, which leads to a signifi-
cantly better metabolic control than one or two insulin injections
per day can provide. By means of the developed model the con-
centration-time profile of plasma insulin can be predicted given
an insulin therapy, as well as the exogenous insulin flow.
II. INSULIN MODELLING
In this section, a comprehensive model for all commercially
available insulin classes including the new insulin Glargine
(tradename Lantus) is developed. This model provides a pre-
diction for the concentration-time profile of plasma insulin
depending on the injected insulin amount and preparation. As
far as exogenous insulin flow is concerned, it is suitable for
healthy people and type-1 and type-2 diabetic patients.
TARÍN et al.: COMPREHENSIVE PHARMACOKINETIC MODEL OF INSULIN GLARGINE AND OTHER INSULIN FORMULATIONS 1995

with rotational symmetry with respect to the origin (that corre-

sponds to the injection site) [1], [8] and, therefore, (1) reduce in
spherical coordinates to

Fig. 1. Block diagram of the developed model.

The problem of shaping the concentration-time profile for dif-

ferent insulin preparations is addressed by an adequate parame-
terization. The developed model is based on Trajanoski’s model
described in [1], which is a simplification of the model proposed
in [8]. According to Nucci [3], it is definitely one of the most
complete models that are available. In contrast to Trajanoski’s
model, where only two different association states are consid-
ered (namely the hexameric and the dimeric state), here a new
virtual insulin association state is introduced: the bound state. (2)
Therefore, in the here developed model, insulin is modeled to
appear in three different states: the hexameric, the dimeric and
the bound state. As can be observed from these equations, the insulin concentra-
The model will be derived in two steps. First, the exogenous tions depend on time and on the distance from the injection
insulin flow into the blood stream is derived from a subcuta- site. The concentrations , , and will
neous injection. This model includes the diffusion at the injec- locally be defined for through a model of infinitesimal thin
tion site, molecular dissociation and absorption. Second, from spheric concentric shells with homogeneous concentrations.
the exogenous insulin flow, a model describing the concentra- The origin of the diffusion process and, therefore, the centre of
tion-time evolution of the exogenous plasma insulin will be de- the shells, is the injection site. This is of course an abstraction
rived. The block diagram of the complete model is represented and corresponds to an approximation of the actual diffusion
in Fig. 1. process.
In an effort to integrate physiological knowledge in the 2) Hexameric-Dimeric Dissociation: For the interaction be-
model, a mechanistic approach is followed. Albeit mechanistic tween the dimeric and the hexameric insulin an equilibrium re-
models exhibit the disadvantage of being harder to adapt to action of third order is modeled as proposed in [8] yielding
experimental data, they tend to be more reliable than mere
phenomenological models like for example artificial neural (3)
networks [9], that are heavily data-oriented.
where and are parameters depending on the insulin class.
A. Insulin Exogenous Flow Applying the law of mass action to (3), the following system of
In the following, the system of differential equations that de- differential equations is obtained:
scribes the variation of the dimeric, hexameric, and bound in-
sulin concentrations, , , and , respectively, is derived. The
variations of these concentrations depend on different factors
that are calculated separately. Finally, the different contributions (4)
to these variations are assembled in a comprehensive system.
1) Diffusion: All the insulin formulations diffuse in the sub- where stands for the concentration of dimeric insulin
cutaneous tissue following the diffusion equations: and for the concentration of hexameric insulin. These
concentrations are initially defined as the mass of in-
jected dimeric and hexameric insulin, respectively, related to the
injection volume . Defining the parameters of
(3) as

(1)

where stands for the position vector and for the diffusion (5)
constant. is supposed to be the same for both the dimeric and
is interpreted as a production rate and as the equilib-
the hexameric insulin forms [10] while for the bound form it
rium constant, obtaining the following third-order dissociation
is reduced by the nondimensional factor . The re-
process:
duced diffusion rate of the bound form is motivated by the fact
that Glargine, that is only completely soluble at pH 4, forms
amorphous microprecipitates in the pH-neutral adipose tissue
[11]. Assuming homogeneity of the adipose tissue, the diffu-
sion process is considered to be isotropic, i.e., homogeneous and (6)
1996 IEEE TRANSACTIONS ON BIOMEDICAL ENGINEERING, VOL. 52, NO. 12, DECEMBER 2005

The chemical equilibrium is characterized by the fact that the tions, a system of partial differential equations that describes the
partial derivative with respect to time vanishes. From system (6), insulin infusion process is obtained
the following equilibrium condition is obtained:

(7)

where stands for the equilibrium concentration of

dimeric insulin and for the equilibrium concentration of
hexameric insulin. Due to the fact that the overall insulin con-
centration (e.g. U100) in the injection volume is given by the
manufacturers and that the injected volume is also known, these (10)
initial equilibrium concentrations can be easily calculated. As
suggested in [1], the dimeric production rate is From (9), the insulin concentration per unit time passing at
considered equal for all insulin preparations including insulin time to the blood stream at a concentric shell of radius is
Glargine. In accordance with Trajanoski [1], it was found that given by . Integrating this expression over the com-
even large variations of have no significant effect on the plete subcutaneous volume, denoted by , the exogenous in-
resulting exogenous insulin flow profile. sulin flow is obtained
3) Bound-Hexameric Conversion: For insulin Glargine it is
supposed that directly after the subcutaneous injection the com- (11)
plete injected insulin volume resides in the bound state. After-
wards, progressively small amounts of insulin undergo a state
This equation implies that the system of coupled partial differ-
transformation from the bound state into the hexameric state.
ential equations (10) shall be solved to obtain the time evolution
The rate, with which insulin of the hexameric association state
of the spatial distribution . This model allows the descrip-
disengages from the large bound complexes is modeled to be
tion of all insulin formulations through the adequate selection
proportional to the concentration of bound insulin with pro-
of the parameters , , , , , and . The values of
portionality factor . As due to the shifted pH of the Glargine
these parameters will be obtained by an iterative identification
molecules owing to the addition of positive amino acids the sol-
process and will vary from patient to patient.
ubility of the bound form is limited, a second term to model
The great advantage of the developed model is that it makes
this saturation process is required. By introducing the param-
it possible to predict the exogenous insulin flow of any injected
eter , that has to be identified, these considerations are
insulin class knowing only the amount of the injected insulin.
taken into account. The modeling idea is that disengagement is
Using the correct parameterization, the same model will be used
inhibited by high concentrations of the hexameric form. There-
to compute the exogenous flow of a basal-bolus insulin therapy.
fore, the disengagement rate is modeled to be proportional to
the difference between the maximum and current con- B. Concentration-Time Evolution of Exogenous Plasma
centration of hexameric insulin. These considerations have in- Insulin
fluence only on the variations of the hexameric and the bound
state yielding the following contributions to the variations of the The modeling approach of the concentration-time evolution
insulin concentrations: of plasma insulin is based on the mass balance of the different
substances and compartments involved in the process, thus
yielding a compartmental model. In general, the concentra-
tion-time evolution is obtained from the balance equation
between compartments. Based on this concept, the concentra-
(8) tion-time evolution of insulin is modeled in [12] describing
the concentration-time evolution of plasma insulin , hepatic
4) Absorption: Only the absorption of the dimeric insulin insulin and interstitial insulin by three coupled differential
form will be considered significant as proposed in [8]. The hex- equations
americ form, due to its molecule size, will have much more
difficulties passing through the capillary wall toward the blood
stream. The absorption rate of the dimeric insulin is considered
proportional to its concentration and it is modeled through the (12)
absorption rate constant obtaining the following contribu-
tion to the variation of the dimeric insulin: with

(9)

5) Resulting Model of the Infusion Process: Finally, assem-

bling all these contributions to the variations of the concentra- (13)
TARÍN et al.: COMPREHENSIVE PHARMACOKINETIC MODEL OF INSULIN GLARGINE AND OTHER INSULIN FORMULATIONS
TABLE I
VARIABLES AND PARAMETERS OF THE INSULIN
MODEL EQUATIONS [(12) and (13)]
The input to this physiological model is the exogenous insulin
flow , calculated in the previous section. In Table I, the
meaning, typical values and units of the variables and parame-
ters appearing in this insulin model defined by (12) and (13) are
summarized.
The values for these parameters are obtained from [10]. By
adapting these parameters, the model can be particularized to
1997
TABLE II
RESULT OF THE IDENTIFICATION PROCESS FOR THE
PATIENT-SPECIFIC PARAMETERS FOLLOWING [13]
each specific patient. The identification and validation processes
of the patient-specific parameters are reported in [13], where
tuning of turned out to be sufficient to match the available
experimental data. However, generally speaking, other param-
eters may need to be adapted. The results obtained in [13] for
five data sets of patients1 are reproduced in Table II.
Finally, the compact state-space representation of the insulin
model described by (12) and (13) yields
(14)
with the state matrix and the input and output vectors, and
respectively, defined as
(15)
The state-space vector is defined as , the input
is the exogenous insulin flow and the output is the
plasma insulin concentration .
III. DISCRETIZATION
In this section, the spatial and time discretization of the above
mentioned model for simulation purposes is derived. As for the
time-discretization, a step width of 0.01 min turned out to be
appropriate.
A. Exogenous Insulin Flow
The developed model of three coupled differential equations
described by system (10) unfortunately has no closed solution.
However, a time and space discretization allows the calculation
of the exogenous insulin flow. For the spatial discretization, a
spheric grid with 15 shells is proposed in [14] accounting to the
model that the insulin after the injection fills a spheric volume
in the adipose tissue and starts to diffuse symmetrically as rep-
resented in Fig. 2, which clearly represents a simplification.
1These data sets were supplied by the Institute for Diabetes-Technology, Ulm,
Germany.
1998 IEEE TRANSACTIONS ON BIOMEDICAL ENGINEERING, VOL. 52, NO. 12, DECEMBER 2005

to apply a simple balance equation which finally leads to the

following discrete diffusion equation (the proof can be found in
the Appendix):

(20)
where stands for the time variation of the concen-
tration in the shell . The volume denoted by repre-
sents the volume of the shell . The difference
describes the net difference between the substance
flow diffusing through the shell at discrete time . The
outgoing substance flow is subtracted from the incoming
substance flow , thus representing the complete substance flow
balance. Therefore, in this representation, the time and space
discretization of the diffusion process is represented by the bal-
ance between the incoming and outgoing substance flow dif-
fusing through the considered shell.
The substance flows and are calculated using Fick’s
first diffusion law [15]
Fig. 2. Spheric grid for the spatial discretization.
(21)
The volume of the injected insulin is denoted by and
stands for the radius of the innermost sphere. Considering this, where is the generic diffusion constant, the substance
is expressed as concentration inside the considered shell and the sub-
stance flow diffusing through the spherical area with radius .
Considering (21) at a fixed time yields
(16)

Analogously, the volume of the spheric shell can be ex- (22)

pressed as a function of the limiting radii and yielding
Hence, in order to calculate the complete substance flow
(17) from one spheric shell to the next at any discrete time instant ,
(22) is integrated from to obtaining

For further calculations, the radius that halves the volume

of the -shell is calculated as (23)

(18) The election of and as limiting radii for the inte-

gration is arbitrary; any other nonoverlapping shell subdivision
could have been chosen.
With these definitions the discretization of the diffusion
With these calculations and substituting and in (20)
process can now be tackled. Considering spheric coordinates
according to (23), the mass balance around the th spheric
and an isotropic diffusion process, Fick’s second diffusion law
shell leads to the discrete diffusion equation
is stated as

(19)

where stands for a generic concentration and for a

generic diffusion constant.
Taking into account the system given by (10), shall
be substituted by the dimeric, hexameric, and bound concen-
trations , , and and the corresponding diffusion con-
stants. This diffusion equation (19) cannot be solved analyti-
cally. Therefore, it has to be discretized in time and space. In
order to do this, a discrete calculation grid consisting of several
spheric shells with index is introduced, offering the possibility
(24)
This equation describes the diffusion process of the insulin
absorption in a general way, discrete in space and time. For
the different insulin states the variable shall be substituted by
the concentration of the actual insulin state, i.e. , , and ,
TARÍN et al.: COMPREHENSIVE PHARMACOKINETIC MODEL OF INSULIN GLARGINE AND OTHER INSULIN FORMULATIONS
as well as the diffusion constant. The complete insulin absorp-
tion model in discrete form for all insulin formulations listed in
Table IV is expressed as
(25)
(26)
(27)
The resulting exogenous insulin flow at time , , is
then calculated by collecting the absorbed amounts of insulin
in each shell after each discrete time step
over the whole grid volume
(28)
where represents the number of spheric shells that
are considered for the calculation. The model parameters for the
different insulin preparations are shown in Table IV.
With this discretization it is now possible to implement and
solve the developed model numerically. However, before being
able to fully implement the model there are some additional as-
pects that have to be considered. The boundary and initial condi-
tions shall be taken into account when solving (25)–(28). For the
innermost sphere it shall be considered that there is no diffusion
from the ( 1)st sphere. In contrast, for the outermost spheric
1999
shell it shall be considered that the insulin concentration disap-
pears for all shells with .
The initial conditions are provided by the amount and class
of the injected insulin. Commercially available insulin prepa-
rations are labeled according to their insulin concentration. An
insulin preparation labeled with U100-Insulin means that it con-
tains insulin with a concentration of 100 IU/ml. Taking this into
account the injected volume can be calculated knowing the in-
sulin concentration and the injected amount of insulin. Initially,
there is no insulin in the shells around the injected volume. For
insulin Glargine, the whole injected insulin volume firstly re-
sides in the bound state. After the injection, small amounts of
insulin are converted into the hexameric state. Therefore, the
initial concentration of the bound state in the innermost sphere
is equal to the concentration of the injected insulin. The ini-
tial concentrations of the hexameric and the dimeric states are
zero. For all other insulin classes, the initial concentration of
the bound state is zero while the concentration of the hexameric
and dimeric states corresponds with the insulin concentration of
the used preparation in equilibrium. The volume of each spheric
shell shall be equal to the injected volume.
B. Plasma Insulin Concentration
Having derived the evolution of the exogenous insulin flow,
the plasma insulin concentration can be obtained from the dis-
cretization of the linear state-space model with sample period
equal to the time step chosen above, . Considering that the
input is constant between discrete time instants and , the
resulting discrete system is [16]
(29)
where is the state vector (amount of plasma
insulin, hepatic insulin and interstitial fluid insulin) and matrices
, , and are the state, input, and output matrices of the
continuous model (15).
IV. SIMULATION RESULTS AND DISCUSSION
As insulin Glargine is comparably new, not many real data
sets have been published for type-1 diabetic patients so far.
Therefore, in this contribution we will focus on the exper-
imental results published by Lepore et al. [2]. Apart from
that, results from real experiments have also been reported for
healthy subjects [17], [18] and type-2 diabetic patients [19].
A. Concentration-Time Profile of Plasma Insulin
In order to validate the derived model, the experimental
results presented in Lepore [2] were tried to be reproduced
with our model. In this study, 20 patients obtained insulin
Glargine and NPH injections of 0.3 IU/kg subcutaneously at
time . Before , insulin was infused intravenously to main-
tain reasonable initial insulin plasma concentrations of about
. After , intravenous insulin infusion was
gradually decreased to zero such that mainly insulin from the
subcutaneous injection site appeared in plasma.
2000 IEEE TRANSACTIONS ON BIOMEDICAL ENGINEERING, VOL. 52, NO. 12, DECEMBER 2005

TABLE III
MODEL PARAMETERS FOR NPH AND GLARGINE

TABLE IV
MODEL PARAMETERS FOR DIFFERENT INSULIN FORMULATIONS
Fig. 3. Comparison of measured (see [2]) and simulated mean insulin plasma
concentrations.

As in [2] mean profiles as opposed to profiles of single

patients are shown, simulations were carried out for the same
five patients as in Section II-B with the parameters depicted in
Table II.
The parameters of the absorption model for insulin Glargine
were found through an iterative identification process whereas
the parameters for NPH insulin have been taken from literature
(see Table III and [1]). Subsequently, from the time-evolution
of for each of the five patients the mean has been calculated
and displayed in Fig. 3.
From Fig. 3 it can be observed that in the simulation NPH
shows a more rapid plasma insulin increase than insulin Fig. 4. Original (top) and changed therapy scheme (bottom) due to lack of
basal insulin in the evening.
Glargine. Also, the simulated NPH curve shows a peak occur-
ring about 6.7 h and returns to the initial value about 16 h after
subcutaneous injection. These observations coincide remark- related to the NPH injection is much lower than that related to
ably well with results cited by Owens [17, p. 813]. Moreover, the Glargine injection.
simulated and measured curves of insulin Glargine equal each Trying to suggest a possible application of the insulin absorp-
other surprisingly well. Interpreting qualitatively the results tion model in therapy, we used two Intensified Conventional
depicted in Fig. 3, the most important observation is that 16 h Therapy (ICT) schemes of a real patient as model inputs taking
after subcutaneous injection, the plasma insulin concentration into account basal and bolus insulin injections (see Fig. 4). The
TARÍN et al.: COMPREHENSIVE PHARMACOKINETIC MODEL OF INSULIN GLARGINE AND OTHER INSULIN FORMULATIONS 2001

Fig. 5. Simulation results of the absorption process of all modeled insulin

classes.

patient eats three meals every day with equal amounts of carbo-
hydrates. Both parts of the figure show the simulated with its
basal fraction emphasized and resulting from the ICT scheme.
Additionally, the basal infusion rate profile, with which an in- Fig. 6. Exogenous insulin flow corresponding to 8 IU for different insulin
sulin pump would be initially operated according to Wizemann formulations.
and Renner [20] given the total insulin amount needed per day,
is depicted. As can be observed, the third bolus peak in the top
part of Fig. 4 is much larger than the other ones. According to
the supervising physicist, the patient frequently suffered high
blood glucose concentration readings at midnight and related
this to a higher bolus insulin need to compensate dinner. Thus,
the patient gradually increased the bolus dose applied for dinner.
However, doing this, frequent severe hypoglycemia following
the meal appeared.
Now, comparing the basal insulin simulated based on the
ICT scheme with the basal rate suggested by Wizemann and
Renner [20] for an insulin pump it can be seen that the reason
for high blood glucose concentration readings at midnight is not
the bolus dose to compensate dinner but a lack of basal insulin
in the evening. Therefore, the basal insulin injection amount at
noon is increased. In the bottom part of Fig. 4 the results of the
modified ICT scheme are shown. Now, the simulated basal frac-
tion of more closely follows the Wizemann–Renner-profile
eliminating the need for very high dinner bolus dosages. Fig. 7. Exogenous insulin flow I for 8 and 16 IU of Glargine in comparison
to 16 IU of NPH.

B. Exogenous Insulin Flow Simulations

The absorption profiles for NPH and Glargine with the set of
parameters in Table III are shown in Fig. 5. These profiles are
coherent with the experiments shown in [17] and [19] regarding
the order of absorption velocities. Additionally, in Fig. 5 the
absorption profiles for other insulin formulations are depicted.
The absorption process parameters for these insulin prepara-
tions were taken from literature as shown in Table IV.
Fig. 6 shows the resulting exogenous insulin flow profile
over a time horizon of 24 h after the injection of 8 IU for dif-
ferent modeled insulin formulations. The profile varies substan-
tially depending on the injected insulin formulation. The pres-
ence of both basal insulin Glargine and NPH is longer than half
a day for 8 IU. However, NPH possesses a pronounced peak in
its exogenous insulin flow in contrast to Glargine that shows an
almost flat form. On the other extreme, the rapid insulin formu-
lations Actrapid and NovoRapid/Humalog show a very high and
pronounced peak but last only during a few hours. The exoge-
nous insulin flow of Semilente lies between the characteristics
of the basal and the bolus insulin formulations.
As can be observed from Fig. 6, all exogenous insulin flows
are smooth and without oscillations. Moreover, the area under
the curves from to , which measures the injected
insulin amount (in this case 8 IU), is equal for all insulin formu-
lations analogues, i.e. it does not depend on the model param-
eters of Table IV. This represents an excellent indicator for the
correct implementation of the model.
Fig. 7 shows a more detailed comparison between an exoge-
nous insulin flow of Glargine and NPH. The profiles of Glargine
injections with 8 and 16 IU clearly show the nonlinearity of the
model: doubling the injection dose does not lead to doubling the
maximum value. In contrast, the double insulin dose leads to a
prolongation of the effect. Furthermore, it can be observed that
NPH shows a pronounced peak in its exogenous insulin flow (at
2002 IEEE TRANSACTIONS ON BIOMEDICAL ENGINEERING, VOL. 52, NO. 12, DECEMBER 2005
Fig. 8. Exogenous insulin flows for basal injections.
.) in contrast to the almost flat profile of Glargine. Ob-
serving the rising and falling profiles of Glargine it can be said
that they are less curved than for NPH. Therefore, considering
the superposition of different injections with a period of 24 h,
the overall exogenous insulin flow for Glargine will be almost
flat compared to the peaks that appear with NPH injections every
12 h as depicted in Fig. 8. The basal injection regime consists
of injections of 14 IU of NPH at time 0, 12, 24, and 36 h and
one injection of 28 IU of Glargine at time 0 and 24 h. In Fig. 8,
only the resulting data from time 24 h to 48 h are shown. The
possible accumulation effects are, thus, taken into account.
C. Spatial Diffusion Simulations
In the following, the correctness of the time and space dis-
cretization and its correct implementation for the obtained in-
sulin absorption model is shown. The time and space discretiza-
tion is analyzed contemplating the implemented grid structure.
In Fig. 9, the state of the concentrations in the discrete calcu-
lation grid after an injection of 8 IU insulin Glargine at time
is represented at different time steps from the injection
time. The -axis represents the number of the spheric shell (see
Fig. 2) considering that the injected volume bears cell number
0. The -axis shows the concentrations of dimeric, hexameric
and bound insulin in the grid cells. The time annotations indi-
cate the time in minutes that has passed between the injection
and the current grid state.
As can be observed from Fig. 9, even 2 h after the injec-
tion most of the injected insulin resides in the bound state, i.e.
the concentration of the bound state in the innermost shell (cell
number 0) is much higher than the concentrations of the dimeric
and hexameric states. The bound state hardly diffuses to the ex-
terior corresponding perfectly with the modeling considerations
about the reduced diffusion rate. For insulin Glargine, it is con-
sidered that directly after the subcutaneous injection all the in-
jected insulin volume appears in the bound state. Afterwards,
small amounts of the insulin in the bound state are converted to
insulin of the hexameric state. This process determines the time
evolution of the insulin concentrations in hexameric, dimeric
and bound state, which corresponds exactly to the model con-
cept of Section II. Moreover, the concentrations of the dimeric,
Fig. 9. Grid state at the indicated time instants after an injection of 8 IU
Glargine.
hexameric and bound state gradually diminish from the inner-
most shell to the outer shells as expected. Although this process
is not continuous but in steps, as depicted in Fig. 6, the resulting
exogenous insulin flow yields smooth provided an ade-
quate time discretization step width of 0.01 min. Experiments
also have demonstrated that even if the grid structure contains
four times more cells the resulting profile is barely modified.
Therefore, with an appropriate time discretization, this space
discretization is considered to be correct and valid.
D. Absorption Evaluation
In the following, experimental results published in [17] and
[19] about Glargine and NPH subcutaneous absorption are used
to evaluate the absorption model. In both cases, the insulin ab-
sorption was measured indirectly monitoring the disappearance
of radioactivity at the injection site. This was done by means of
a thallium-activated sodium iodide gamma counter positioned
over the injection site. It has to be noted that these profiles rep-
resent rather than insulin and therefore, the extent to which
insulin or a degradation product containing was measured
is not known, as reported in [19]. This is also stated in [22, p. 6],
where the intrinsic instability of the radio-labeled compounds is
pointed out as the cause of incoherent conclusions drawn from
residual radioactivity measurements. Thus, the data published in
[17] and [19] have to be carefully interpreted, i.e. qualitatively
rather than quantitatively.
In [17], two studies were conducted for healthy subjects.
Study 1 compared the subcutaneous abdominal injection of
0.15 IU/kg of -labeled insulin Glargine[15] (zinc content
of 15 ), insulin Glargine[80] (zinc content of 80 ),
NPH insulin, and placebo. In study 2, 0.2 IU/kg of insulin
Glargine[30] (zinc content of 30 ) was injected into
the arm, leg, and abdominal regions. Both studies had a ran-
domized crossover design and enrolled 12 healthy men, aged
18–50 years. As the commercial formulation of Glargine is
Glargine[30], and to enable a comparison with the data of
[19], only study 2 will be considered for the evaluation of
the absorption model. More precisely, the data obtained after
abdominal injection will be chosen for the comparison. This
selection is arbitrary but, as reported in the article, the injection
site has a minor influence on the absorption velocity.
In [19], the subcutaneous absorption characteristics of insulin
Glargine and NPH insulin in patients with type-2 diabetes were
TARÍN et al.: COMPREHENSIVE PHARMACOKINETIC MODEL OF INSULIN GLARGINE AND OTHER INSULIN FORMULATIONS
TABLE V
ADAPTED MODEL PARAMETERS FOR NPH AND GLARGINE
Fig. 10. Glargine/NPH absorption: simulation versus experimental data.
compared. In this study, 14 patients with type-2 diabetes (aged
40–70 years) previously untreated with insulin were randomized
to receive in a fasting state either a single subcutaneous injection
of 0.3 U/kg -insulin Glargine or 0.3 U/kg -NPH insulin.
In order to fit the experimental data, both the NPH param-
eters taken from literature and the Glargine parameters previ-
ously identified (see Table III) had to be changed. For the param-
eter set in Table V, the results shown in Fig. 10 were obtained.
While it is not impossible to adapt the model to the arguable
data, the large deviation of the parameters shows that the model
had to be forced to fit the data. This can also be concluded com-
paring the experimental results with Fig. 5. According to the
absorption profile, a subcutaneous NPH injection is expected to
be completely absorbed within about 24 h. This coincides with
experiences from therapy and is reflected in Fig. 5 but not in
the data resulting from residual radioactivity measurements in-
dicating that the right parameter values are those of Table III.
2003
V. CONCLUSION
In this contribution, a comprehensive model for the pharma-
cokinetics of most commercially available insulin formulations
is developed, including the recently introduced insulin Glargine
(tradename Lantus).
The developed model is based on the model proposed by [1]
that is one of the most complete models currently available [3].
In the Trajanoski model, only two insulin states are considered.
In the model developed here, a new virtual insulin state, namely
the bound state, is introduced. Immediately after the subcuta-
neous injection all the injected insulin resides in the bound state
and only then small amounts of insulin of the hexameric state
disengage from the bound state. The model parameters have
been obtained for most of the commercially available insulin
formulations. Therefore, the developed model serves for all pre-
diction purposes selecting the correct parameterization. Unfor-
tunately, the model of three coupled differential equations has
no closed solution and therefore, a time and space discretiza-
tion is implemented for the calculation of the insulin concentra-
tion-time profile.
The insulin absorption model is extended to describe the de-
composition process of insulin as suggested in [12]. Thus, the
concentration-time evolution of exogenous plasma insulin after
a subcutaneous injection is calculated. Using this comprehen-
sive model, the experimental results published in [2] could be
reproduced. Comparing the resulting plasma insulin concen-
trations following the test injections, it can be concluded that
our model delivers accurate qualitative predictions. The simu-
lated plasma insulin concentration curve after an NPH injection
coincides very well with qualitative observations known from
therapy: the simulated NPH shows a more rapid plasma insulin
increase than insulin Glargine, a peak occurring about 6–7 h
and a return to the initial value about 16 h after injection. More-
over, simulated and measured curves of plasma insulin concen-
tration after the Glargine injection resemble each other as well
indicating the prediction capability of the developed model. For
further evaluation of these prediction capabilities, ICT schemes
from a real type-1 diabetic patient before and after a regimen
change due to a dosage error of basal insulin were used as in-
puts to the model. It can be concluded that the developed model
indeed is able to show fundamental shortages in the basal insulin
profile resulting from ICT scheme errors when is compared
to a standardized insulin pump infusion profile.
Finally, analysing the time and space discretization and the
implementation of the obtained insulin absorption model, it can
be concluded that both discretizations as well as the implemen-
tation are correct. The curves show no oscillations, the AUC
considerations support the implementation correctness and the
simulation results coincide with expectations and modeling
considerations.
APPENDIX
Lemma 1: Let the diffusion process be described according
to Fick’s second diffusion law yielding
(30)
2004 IEEE TRANSACTIONS ON BIOMEDICAL ENGINEERING, VOL. 52, NO. 12, DECEMBER 2005

where denotes the concentration as a function of time where (17) has been used for . Now, all that remains, is to
and space depending on the isotropic diffusion constant and show that
the concentration gradients in space abbreviated by the Laplace
operator . This equation can be approximated by the discrete
diffusion equation
(40)
(31)
for this, consider the definitions of and and
with denoting the volume of the st shell, repre-
senting the flux into this shell and describing the flux out
of this shell respectively.
Proof: Equation (30) can be rewritten expanding the
Laplace operator, considering rotational symmetry, in spherical (41)
polar coordinates as
Therefore
(32)

The proof consists of showing that (32) is the limiting case for
(31). For the sake of later convenience the function is
defined as

(33) (42)

Equation (32) can be expressed as

ACKNOWLEDGMENT
The authors are very grateful to one of the reviewers for his
(34) constructive incisive comments.

Now define REFERENCES

[1] Z. Trajanoski, P. Wach, P. Kotanko, A. Ott, and F. Skraba, “Pharmacoki-
netic model for the absorption of subcutaneously injected soluble in-
sulin and monomeric insulin analogues,” Biomedizinische Technik, vol.
38, no. 9, pp. 224–231, 1993.
(35) [2] M. Lepore, S. Pampanelli, C. Fanelli, F. Porcellati, L. Bartocci, A. di
Vincenzo, C. Cordoni, E. Costa, P. Brunetti, and G. Bolli, “Pharmacoki-
netics and pharmacodynamics of subcutaneous injection of long-acting
so that (34) becomes human insulin analog glargine, nph insulin, and ultralente human insulin
and continuous subcutaneous infusion of insulin lispro,” Diabetes, vol.
49, pp. 2142–2148, 2000.
(36) [3] G. Nucci and C. Cobelli, “Models of subcutaneous insulin kinetics.
A critical review,” Comput. Meth. Programs Biomed;, vol. 62, pp.
249–257, 2000.
On the other hand, the fluxes and in (31) quantifying [4] L. Reinhart and C. Panning, “Insulin glargine: a new long-acting insulin
the amount of insulin entering and leaving the considered shell product,” Am. J. Health Syst. Pharm., vol. 59, no. 7, pp. 643–649, Apr.
respectively are expressed depending on the concentration gra- 2002.
[5] L. Heinemann, “Variability of insulin absorption and insulin action,” Di-
dient by means of Fick’s first diffusion law [see also (22)] abetes Care, vol. 4, pp. 673–682, 2002.
[6] T. Levien, D. Baker, J. J. White, and R. Campbell, “Insulin glargine: a
new basal insulin,” Ann. Pharmacother., vol. 36, pp. 1019–1027, 2002.
(37) [7] R. Rosskamp and G. Park, “Long-acting insulin analogs,” Diabetes
Care, vol. 22, no. suppl. 2, pp. B109–B113, 1999.
Considering definition (33), (37) is rewritten as [8] E. Mosekilde, K. Jensen, C. Binder, S. Pramming, and B. Thorsteinsson,
“Modeling absorption kinetics of subcutaneous injected soluble insulin,”
J. Pharmakokinet. Biopharm., vol. 17, pp. 67–87, 1989.
(38) [9] E. Teufel, M. Kletting, H.-J. Pfleiderer, and C. Tarín, “Glucose metab-
olism modeling with BPTT-trained Elman nets,” in Proc. IEEE Neural
Therefore, (31) can be rewritten as Network for Signal Processing Workshop 2003, vol. 1, 2003, pp. 1–2.
[10] B. Höfig, “Physiologische Modellierung des menschlichen Glukose-
Metabolismus für die simulationsgestützte Therapie des insulinab-
hängigen Diabetes mellitus,” Ph.D. dissertation, Universität Stuttgart,
Institut A für Mechanik, Stuttgart, Germany, 1996.
[11] C. Dunn, G. Plosker, G. Keating, K. McKeage, and L. Scott, “Insulin
(39) glargine: an updated review of its use in the management of diabetes
mellitus,” Drugs, vol. 63, no. 16, pp. 1743–1778, 2003.
TARÍN et al.: COMPREHENSIVE PHARMACOKINETIC MODEL OF INSULIN GLARGINE AND OTHER INSULIN FORMULATIONS
[12] C. Cobelli, G. Federspil, G. Pacini, A. Salvan, and C. Scandellari, “An
integrated mathematical model of blood glucose and its hormonal con-
trol,” Math. Biosci., 1982.
[13] E. Teufel, C. Tarín, J. Picó, J. Bondia, G. Freckmann, and H.-J. Pflei-
derer, “Parameter identification of the endogenous glucose metabolism
model for diabetes mellitus I patients,” presented at the Proc. Int. Wis-
senschaftliches Kolloquium, Ilmenau, Germany, Sep 2004.
[14] P. Wach, Z. Trajanoski, P. Kotanko, and F. Skrabal, “Numerical approxi-
mation of mathematical model for absorption of subcutaneously injected
insulin,” Med. Biol. Eng. Comput., vol. 33, no. 1, pp. 18–23, Jan. 1995.
[15] A. Fick, “Fick’s first law of diffusion,” Ann. Physik, vol. 170, p. 59, 1855.
[16] C.-T. Chen, Linear System Theory and Design. Oxford, U.K.: Oxford
Univ. Press, 1999.
[17] D. Owens, P. Coates, S. Luzio, J. Tinbergen, and R. Kurzhals, “Pharma-
cokinetics of 125I-labeled insulin glargine (HOE 901) in healthy men:
comparison with NPH insulin and the influence of different subcuta-
neous injection sites,” Diabetes Care, vol. 23, pp. 813–819, 2000.
[18] G. Kuerzel, U. Shukla, H. Scholtz, S. Pretorius, D. Wessels, C. Venter,
M. Potgieter, A. Lang, T. Koose, and E. Bernhardt, “Biotransforma-
tion of insulin glargine after subcutaneous injection in healthy subjects,”
Curr. Med. Res. Opinion, vol. 19, pp. 34–40, 2002.
[19] S. D. Luzio, P. Beck, and D. R. Owens, “Comparison of the subcutaneous
absorption of insulin Glargine (Lantus) and NPH insulin in patients with
type 2 diabetes,” Horm. Metab. Res., vol. 35, pp. 434–438, 2003.
[20] E. Wizemann, R. Renner, U. Reitberger, and K. D. Hepp, “Prospektive
Evaluation einer standardisierten Basalratenverteilung für die CSII bei
Typ 1 Diabetes über 6 Monate,” presented at the Deutsche Diabetes-
Gesellschaft, Aachen, Germany, May 2001.
[21] R. Gessler, “Ein portables System zur subkutanen Messung und
Regelung der Glukose bei Diabetes mellitus Typ-I,” Ph.D. dissertation,
Abteilung Allgemeine Elektrotechnik und Mikroelektronik an der
Universität Ulm, Ulm, Germany, 2000.
[22] (2005) Lantus—European Public Assessment Report. European
Medicines Agency. [Online]. Available: http://www.emea.eu.int/htms/
aboutus/emeaoverview.htm
[23] C. Tarín, E. Teufel, G. Freckmann, and H.-J. Pfleiderer, “Pharma-
cokinetic model for the absorption of subcutaneously injected insulin
Glargine,” in Proc.AIDPIT Workshop3, vol. 1, 2004, pp. 1–2.
[24] J.-W. Chen, J. S. Christiansen, and T. Lauritzen, “Limitations to sub-
cutaneous insulin administration in type 1 diabetes,” Diabetes, Obesity,
Metab., vol. 5, pp. 223–233, 2003.
C. Tarín received the M.S.Eng. degree from the
Technical University of Valencia, Valencia, Spain,
in 1996 and the Ph.D. degree from the University of
Ulm, Ulm, Germany, in 2001.
She is currently with the Department of Systems
Engineering and Control of the Technical University
of Valencia. Her main research interests are in control
of biologic systems, nonlinear control, and intervalar
analysis applied to robust control.
2005
E. Teufel received the M.S.Eng. degree from the
Technical University of Karlsruhe, Karlsruhe,
Germany, in 1999 and the Ph.D. degree from the
University of Ulm, Ulm, Germany, in 2004.
He is currently with the Department of Systems
Engineering and Control of the Technical University
of Valencia, Valencia, Spain. His main research inter-
ests are in modeling of biological and technical pro-
cesses, bio-sensors, and diabetes care.
J. Picó (M’90) received the M.S.Eng. and Ph.D.
degrees from the Technical University of Valencia,
Valencia, Spain, in 1989 and 1996, respectively.
He is currently Associate Professor with the
Department of Systems Engineering and Control
of the Technical University of Valencia. His main
research interests are in nonlinear and intelligent
control, modeling and control of biologic and
biotechnological processes, and application of
intervalar arithmetic to robust control.
J. Bondia received the M.S. and Ph.D. degrees
from the Technical University of Valencia, Valencia,
Spain, in 1994 and 2002, respectively.
He is currently Senior Lecturer at the Department
of Systems Engineering and Control of the Technical
University of Valencia. His main research interests
are in real-time systems, intervalar systems, and ro-
bust control of systems with parametric uncertainty.
H.-J. Pfleiderer (M’83–SM’83–F’90) received the
M.S.Eng. and Ph.D. degrees from the Technical Uni-
versity of Stuttgart, Stuttgart, Germany, in 1967 and
1971, respectively.
As a Postdoctoral Fellow at IBM’s T. J. Watson
Research Laboratory, Yorktown Heights, NY, he
worked on systems for magnetically recorded sig-
nals. Since 1973, he has been with the Research
Laboratories, Siemens AG, Munich, Germany,
engaged in the design of silicon circuits, especially
charge coupled devices. When he left the Research
Lab in 1989 to join the Microelectronics Department at the University of
Ulm, he was Director of the Silicon Microelectronics Laboratory including
processing and design. His interests are the design of silicon circuits, ranging
from technology oriented BiCMOS-circuits to systems oriented asynchronous
circuits.
Dr. Pfleiderer is a member of the ITG board of directors.