Mathl Comput. Modelling, Vol. ! i, pp. 6-10, 1988 0895-7177/88 $3.00 + 0.

00
Printed in Great Britain Pergamon Press plc

Design of Thyroxine Replacement Therapy

with SCoP and SCoPfil

Michael C. Kohn
National Biomedical Simulation Resource
Box 3709 Duke (,'nivcrsit~" Medica/ Com:r
Dm'h~m, NC )7710

S00-672-2543 (NC)
S()0+334 2()$3 (USA)

The Resource offers periodic simulation training workshops

Abstract. The National Biomedical Simulation Resource on both introductory and advanced levels and organizes
(NBSR) is a federally funded computer installation which annual topical symposia on a subject of current interest in
provides nationwide access without charge to specialized biomedical simulation. The NBSR maintains a mailing list
hardware and software for computer simulation in of approximately 1000 researchers. People on this list are
biomedical research and offers training in their use. The kept informed of newly acquired hardware and software
NBSR staff has written a simulation program package, through a quarterly newsletter, Impulse. Resource staff
SCoP (Simulation Control Program), which is versatile and members may be contacted through voice only WATS lines
easy to use. SCoP is available for IBM PC, VAX VMS, for assistance with using the computers or programs.
and Unix systems. Instructions are given for obtaining an
account on the NBSR computer and a copy of the SCoP NBSR Hardware and Software
package. A model of thyroxine replacement therapy for a
patient with defective thyroid function is presented as an The Resource host computer is currently a VAX 11/750
example of the use of SCoP. running the VMS version 4.4 operating system. We expect
to add a VAX 8550 running the Ultrix operating system in
Keywords. computer resource; biomedical simulation; the near future. Peripheral processors on the VAX 750
parameter estimation; design of therapy. include a mini-MAP and a MAP 6430, a single-precision and
a double-precision array processor (CSP, Inc., Billerica,
MA), respectively. The NBSR also has an AD 10 and an
Need for Simulation AD 100, fixed-point and floating-point digital integrators
(Applied Dynamics International, Ann Arbor, MI),
Biological systems are complex owing to synergistic respectively. If the speed of an IBM PC is taken as unity,
interactions among their constituent elements. Because of the speed of a PC with an 8087 math processor, a VAX
this complexity such systems frequently exhibit nonintuitive 11/750, a MAP 6430, and an AD 100 are 8, 60, I000, and
or even counterintuitive behavior. This makes it difficult to 7000, respectively.
design definitive experiments and to interpret experimental
data objectively. Simulation is a research tool that can help
overcome these difficulties by predicting the logical Commercial simulation languages at the Resource include
CSSL-IV (Simulation Services, Chatsworth, CA) and ACSL
consequences of an investigator s hypotheses and identi- (Mitchell and Gauthier Assoe.,.Ine.,Concord, MA). Also_
fying thecritiealty important variables in his systen~ included isS I M N O N (ScientificSystems, Camlnidge, MA),
an interactivesimulation language. The Resource software
Untilrecently few biomedical researchersused simulation in librarycontains other mathematical programs to aid in the
theirwork because they lacked experience with computers constructionof simulationmodels. These include the I M S L
and mathematics. Also, most existing computer centers arc subroutine library(IMSL, Inc.,Houston, 'IX), O D E P A C K
not well suited to support simulation efforts. It is difficult to (Lawrence Livermore Laboratory, Berkeley, CA) for
obtain funding for large research computers, and historically solution of stiff differential equations, and the SAS statistical
has been little sharing of hardware and software mmong program (SAS Institute, Cary, NC). The Symbolic
existing installations. Manipulation Program (SMP) on the VAX 750 solves
equauons algebraically instead of numerically and is useful
One solution to this problem has been the creation of the in genca'ating the equations constituting a simulation model.
National Biomedical Simulation Resource (NBSR), a
federaily funded computer installationat the Duke University Graphics support at the NBSR includes PLOT I0, a popular
Medical Omter which has specializedhardware and software graphics package from Tektronix, and GRAPHLIB, a
for biomedical simulation. The Resource is accessible locally written package of graphics subroutines. The small
nationwide via the T Y M N E T network and outside North size of the NBSR staff makes plotting users' output
Carolina by a directW A T S line. Nonrenewable temporary impractical. However, a user can produce plots from his
(threemonths) accounts arc availablefor those who wish to own terminal if he has the appropriate hardware.
examine our software for possible use and annually
renewable project accounts for those with established
research effortsthat require computer tools. Collaborator
accounts are availablefor additionalinvestigatorson a single
project. There are no charges for any of the NBSR's
services,includingthe telephone link.
Simulation Control Program SCoP includes a library of numerical methods routines to aid
in the definition and solution of the model’s equations. The
al., 1986) a library includes eight integrators, four simultaneous
general simulation program, SCoP (Simulation Control equations solvers, five general modeling functions to define
Program), in the C language for IBM PC, and hyperbolic, sigmoidal, and cubic spline curves, and seven
Unix stochastic modeling functions. Nine of the most common
forcing functions are provided, including step, pulse, ramp,
and several ueriodic functions. The librarv includes routines
for steady-&ate and trajectory sensitivity analysis. These
routines may be called directly from within the template file.

After creating the corresponding parameter database file with

Paredit, the SCoP model is compiled and linked to a library
of SCoP routines. Command files for assembling SCoP and
SCoPfit models are supplied with SCoP to simplify this
step. If no compiler errors are detected the model will
commence execution.

SCoP includes command-line and full-screen editing

functions to enable the user to interactively alter the values of
model parameters. The results of a simulation can be saved
in a data file for subsequent output in graphical and/or
tabular format. The user can interactively select parameter
files, reference data files, or saved output data files for
processing by SCoP.

Thyroxine Regulation Model

Spain (1982) reviewed the observed regulation of blood

thyroxine (TH) level by circulating thyroid stimulating
hormone (TSH) and the feedback inhibition of TSH release
from the pituitary by circulating TH. Figure 1 shows a
block diagram of this system. His model of this process is
outlined in Fig. 2. The main features of this model are
Michaelian kinetics for TSH-stimulated release of TH from
the thyroid gland, sigmoidal kinetics for inhibition by TH of
TSH release from the pituitary, and linear kinetics for the
clearance of the two hdmrones from the circulation. Fig. 2
also gives the parameters Spain reported for the normal
condition.

A patient was examined by a physician who found an

abnormally low level of serum TH. We created a SCoP
model based on Spain’s equations by completing the
template for time-dependent models and selecting the fourth-
order Runge-Kutta algorithm to perform the integration, We
simulated the progress of the patient over an interval of 30
days. Figure. 3 shows the predictions of the model for the
time course of the recovery of normal hormone levels,
assuming that the patient’s TH deficiency is temporary. The
data points are the hormone levels actually observed by the
physician.

The defect in regulation of the hormones could be due to a

toxin which competes with TSH for its receptor on the
plasma membrane of thyroid cells, inhibits the transport of
TH across the cell membrane, or blocks the expression of
the gene for either the TSH receptor or the TH transporter.
Note that the toxin would not be expected to interfere with
either the binding of TH to the pituitary receptor or the
exocytosis of TSH-containing vesicles because serum TSH
is actually elevated.

Changes in the uarameters in the model which corresuond to

the possible defects are an increase in the Ko.5 for TSH
binding and a decrease in the Vmax of TH release. We
for attempted to fit the observations by altering these two
The user creates a modelby selecting the appropriate oarameters. The best fit (Fig. 4) was obtained with a 60%
,Y,

template and using a text editor to insert the information reduction in Vmax to 4 l@L/day. Because the rate of release
requested by the detailed comments in the template file. of TH is limited by the fraction of the TSH receptors
occupied, this purely kinetic model cannot distinguish
between interference with TSH binding and impairment of
TH transport.
8 Proc. 6th Int. Conf. on Mathematical Modelling

Design of Optimal Therapy
To correct for the low TH level, the physician decided to
administer thyroxine replacement therapy. However, to
optimize the therapy it is important to predict the patient's
response to variations in the dose of TH given and the
frequency of administration. We have modified Spain's
model to include periodic doses of TH by using SCoP's
perpulse (periodic pulses) function.
TH = TH + perpulse(time, lag, dose, duration, delay);
The lag time was taken as 0 days. As the integration time
step in the corresponding SCoP model was about 15 min,
we used with a pulse duration of 10 min (less than one
integration step) to ensure that the model would treat the
dose as a bolus of hormone. The model was implemented in
SCoPfit, and the dose and delay were selected as the
adjustable parameters. Initial estimates for the parameter
values were 3 ~tg/L of blood for the dose and 1 day for the
delay between consecutive doses. The values of serum TSH
and TH levels to be matched were taken from the predictions
of Spain's model with the parameter values corresponding to
the normal condition.
The initial parameter values gave a root-mean-square
deviation of about 12 gg/L of the hormone concentrations
from their desired values (coefficient of variation = 25%).
After about 100 passes through the model (each iteration of
the Praxis algorithm requires several integrations of the
equations), the optimal value of the dose was 5.5 gg/L and
the delay was unchanged. The least-squares error was
reduced to less than 1.2 t.tg/L, a coefficient of variation of
2% (Fig. 5). Note the oscillations in the predicted TH serum
concentration. The reduced Vmax for TH release is
insufficient to compensate for the rate of clearance of the
bolus of hormone administered to the patient, resulting in a
decline in the TH level until the next dose of hormone.
Although the optimization required an hour on the IBM PC,
it was completed in 7 minutes on the NBSR VAX 750 and in
less than 2 min on a Convex supercomputer.
Conclusions
SCoP is easy to learn and easy to use while still offering
powerful modeling functions. NBSR clients have found the
SCoP package of programs to be valuable in many areas of
research, including design of therapy, pharmacokinetic and
pharmacodynamic model~g, enzyme kinetics, membrane
transport phenomena, and electrophysiology.
This work was supported by USPHS grant RR10693.
References
Brent, R. P. (1973). Algorithms for Minimization without
Derivatives. Prentice-Hall: Englewood Cliffs, NJ.
Kootsey, J. M., M. C. Kohn, M. D. Feezor, G. R.
Mitchell, and P. R. Fletcher (1986). SCoP: a Simulation
Control Program for Micro- and Minicomputers. Bull.
Math. Biol., 48, 427~,41.
Spain, J. D. (1982). Basic Microcomputer Models in
Biology. Addison-Wesley: Reading, MA.

Figure 1
Thyroxine Regulation Model

I Clearance
Pituitary

TSH Release
~,~ TSH

TH

TSH
Binding

Binding

Th,roid

I TH / ~

~ Clearance
TH Release

TSH
 Figure 2

Thyroxine Regulation Model Equations

TH = [Thyroxine] TSH = [Thyroid Stimulating Hormone]

Rate of TSH Release Rate of TH Release

Vmax Vmax
v= ” =
1 + (TH/K; 13 I+ K;y/TSH
Vmax = Spg/L/day Vmax = lOpg/L/day
TSH
KF5 = SOpg/L K OS = l.$lg/L

Rate of TSH Clearance Rate of TH Clearance

v = CTSH * TSH v=CTH*TH
CTSH = 0.25 day- 1 CTH = 0.1 day- 1

DTSH = v_TSH-release v_TSH_clearance

DTH = v_TH_release - v_TH_clearance

Figure 3

Hormone Levels with Normal Parameters

100 -

60 -

$ 60-

40
. .
. .
.
.

20 .
l . . . .

Ol..~..,.....,.....,
0 10 20 30

Time, days

Fig. 3. The curyes are the predictions of the model and the data points are
the observed hormone concentrations. In this and the two following figures
diamonds are TSH concentration and squares are TH concentration.
 Proc. 6th Int. Cmf on Mathrmutical Meddling

Figure 4

Hormone Levels with Abnormal Parameters

100

90 1
80
1

70 -

Jo GO-

2 50 -

E 40 - .
s?
30 --

o!. . . . . I.. -. I., . . I

0 10 20 30

Time, days

Fig. 4. The Vmax for TH release was reduced 60 c/o to 4 kg/L/day.

Figure 5

Hormone Levels after Treatment

ol 0 10 20 30

Time, days

Fig. 5. Response of the model with thyroxine dose