1 Optimal Control Problem of Treatment for Obesity

2 in a Closed Population
3

4 D. Aldila , N.Rarasati, N.Nuraini, E.Soewono

5
1 Department of Mathematics, Faculty of Mathematics and Natural Sciences, Institut
6 Teknologi Bandung
7
aldiladipo@yahoo.co.id

8 Abstract Variety of intervention programs for controlling the obesity epidemic has
9 been done worldwide. However, there is not been found yet satisfactory scientific
10 tool to measure the effectiveness of the program. This is due to the difficulty in
11 parameterizing the human interaction and transition process of obesity. A dynami-
12 cal model for simulating the interaction between healthy people, overweight people,
13 and obese people in a randomly mixed population is discussed in here. Two sce-
14 narios of intervention programs were implemented in the model, dietary program
15 for overweight people with healthy life campaign and treatment program for obese
16 people. Assuming all control rates are constant, disease free equilibrium point, en-
17 demic equilibrium point and basic reproductive ratio (R0 ) as the epidemic indicator
18 were shown analytically. We find that the disease free equilibrium point is locally
19 asymptotical stable if and only if R0 < 1. From sensitivity analysis of R0 , we obtain
20 that larger rate of dietary program and treatment program will reduce R0 signifi-
21 cantly. With control rates are continuous in time, an optimal control approach was
22 applied in to the model to find the best way to minimize the number of overweight
23 and obese people. Some numerical analysis and simulations for optimal control of
24 the intervention were shown to support the analytical results.
25 Keywords : Obesity, basic reproductive ratio, optimal control.

26 1 Introduction
27 Obesity is an overweight situation in human body as a result of excessive accumula-
28 tion of fat situation. Every person need some calories to save them as the energy, as
29 well as to keep their body warm, and many other purpose. The high consumption of
30 high calories food, over nutrition and fast food combining with less physical activity
31 to burn the calories become the main factor that causing obesity. The normal com-
32 parison between body fat with peoples weight is 18-23 percents for man and 25-30
33 percents for women [7].
34 Obesity has reached the epidemic proportions since recent decades [3]. It be-
35 comes a worldwide problem as stated in the WHO report [8], there are more than 1
36 billion overweight adults and at least 300 million of them obese. Obesity could lead
37 to chronic diseases, including diabetes type 2, cardiovascular disease, hypertension,
38 stroke and certain forms of cancer [8].
39 Several factors have been identified in determining the susceptibility for obesity
40 such as human genes and also energy balance. Calory intake and also physical
41 activity are the main factors in energy balance.
42 Many programs have been initialized by WHO to solve the obesity pandemic
43 such as recognizing the heavy criteria and growing burden of noncommunicable
44 disease, developing global strategy on diet, physical activity and health through
45 broad consultation processes [7, 11]. Promoting healthy behaviours to encourage,
46 motivate and enable individuals to lose weight by eating more fruit and vegetables,
47 doing more physical activity, and reducing the consumption of sugars food have been
48 done around the world recently. The problem is how to calculate the effectiviveness
49 of these programs in long time period.
50 Some mathematical models have been developed to understand this disease like
51 in [10, 9]. Authors in [10] modeled that the obesity could spread by social interaction
52 among people with statistical approach. SI deterministic mathematical model has
53 been developed by author in [9]. Based in longitudinal data, the authors extract
54 and compare the rates of spontaneous versus contagious acquisition of a behavior to
55 predict the implications for future prevalence and control strategies.
56 Unlike the authors in [10, 9], the mathematical model of obese disease with ver-
57 tical transmission will be constructed in this paper. Intervention programs such as
58 healthy life campaign and rehabilitation program are applied in the model as control
59 variable which depends on time. Mathematical assumption and the construction of
60 the model will be shown in the next section. Mathematical analysis for equilibrium
61 points and basic reproductive ratio are shown in section 2. Optimal control prob-
62 lem and some numerical results are given in section 3 and 4, respectively. Some
63 conclusions are given in the last section.

64 2 Formulation of the model

65 Mathematical model for obesity in a closed population will be constructed in this
66 section. We assume that the human population is divided in to three different
67 compartments, i.e healthy compartment (x), overweight compartment (y) and obese
68 compartment (z). The human population, denoted by K, is assumed to be constant
69 with the per capita mortality rate given by . We also assumed that obesity does
70 not cause death in human population.
71 According to [1, 2], obese disease could be transmitted vertically to new born
72 population through genetics. We assume that each compartments have same re-
73 cruitment rate for new born, denoted by . New borns from healthy people is 100
74 percent healthy, while if of the parents is overweight or obese, then some portion of
75 newborns will be healthy or overweight with the portion p and q, respectively.
76 Transitions from healthy to overweight compartment depends on daily social
77 interaction between healthy people with overweight and/or obese people with inter-
78 action coefficient given by b. With the same assumption, transition from overweight
 Table 1: Parameters description and value
Par. Description Value
1
natural recruitment rate (per day) 65365
1
natural mortality rate (per day) 65365
0.1
b interaction coefficient (per day) K
infection rate because bad life habit (per day) 0.05
natural recovery rate from overweight to healthy comp (per day). 0.05
K total of human population (individuals) 1000
p portion of health recruitment rate from overweight comp . 0.5
q portion of health recruitment rate from obese comp. 0.5
u1 health life campaign rate (per day) [0, 1]
u2 treatment rate for obese people (per day) [0, 1]

Figure 1: Transmission diagram for obese disease

79 to obese compartment also depend on daily social interaction with obese people with
80 interaction coefficient given by b and also transition due to unhealthy living habits
81 with rate of .
82 We assume that recovery rate from overweight compartment to healthy compart-
83 ment is while transition rate from obese compartment to overweight compartment
84 could not happen naturally. Intervention programs such as healthy life style cam-
85 paign is applied in the model to increase transition rate from overweight compart-
86 ment to healthy compartment with rate of u1 . Treatment intervention programme
87 is given to cure obese people with the rate of u2 .
From the assumptions above which is illustrated by the transmission diagram
given in Figure (1) and parameter description in table 1, we introduce the dynamical
system
dx b b


dt
= (x + py + qz) + ( + u1 (t)) y K

y + K
z x x
dy b b

= ((1 p)y + (1 q)z) + y + z x + u (t)z...

dt K K 2
(1)
( + u
1 (t)) y Kb z + y y




dz
= Kb z + y u2 (t)z z.

dt

88 The parameters of u1 and u2 applied in the model as the control variables which
89 depend on time. This should be done to accomodate the limitation of budget for
90 campaign and rehabilitation program. For mathematical analysis of the equilibrium
91 points and construction of the basic reproductive ratio which will be discussed in
 92 the next section, we first consider that these control variables u1 (t) and u2 (t) are
93 constant.

94 3 Analysis of the model

95 3.1 Equilibrium points
96 In order to obtain the equilibrium points of system 1, we assume all control variables
97 are constant and given by u1 and u2 . We also assume that the recruitment rate for
98 new born is same with natural mortality to avoid the blow up disappearance of
99 all compartments ( = ). Taking right hand side of equation 1 equal to 0, the
100 system 1 have two different equilibrium point i.e disease free equilibrium (DFE)
101 where all infected population is equal to 0 and endemic equilibrium (EE) where all
102 compartments exist in the field.
The disease free equilibrium point is given by
DF E = (x , y , z ) = (K, 0, 0) , (2)
103 with K is the total population. The endemic equilibrium point is given by
 
g zK ( + u2 )
EE = (x, y, z) = , , z ,
(u2 K + K + bz + K) b bz + K
104 with g = K (p Ku2 + p2 K + qbz + q K + K u2 + K +Ku1 u2 + Ku1 )
105 and satisfy x + y + z = K.

106 3.2 Basic reproductive ratio

Using the next generation matrix operator approach in [5], we compute the basic
reproductive ratio R0 which is associated with the disease free equilibrium (DFE).
The basic reproductive ratio R0 represents the expected numbers of secondary cases
produced by a typical infected individual during the entire period of infectiousness
in a virgin population [4]. First of all, to construct the next generation matrix, we
define
A = (G + T )A (3)
where A = (y, z)transpose , G as transition matrix and T as transmission matrix.
Transition matrix describing change in state , including removal rate like death or
immunity while transmission rate describing the production of new infections. The
transition matrix G from the infected compartment (overweight and obese) is given
by " #
(1 p) u1 (1 q) + u2
G= , (4)
u2
while the transmission matrix T is given by
" #
b b
T = , (5)
0 0
 because in disease free equilibrium the whole population are healthy which is equal
to K. In [5, 6], the basic reproductive ratio is equal to spectral radius of the matrix
T G1 . It can be seen that the matrix T G1 which is called as the next generation
matrix is given by
" b(+u ) b(+ qu2 ) b( p+++u1 )
#
b
L
2
+ L
L
+ L
N GM = , (6)
0 0

where L = p u2 + 2 p + q + u2 + + u1 u2 + u1 . Further detail about

construction of next generation matrix can be seen in [5]. Recall that the spectral
radius of a matrix is equal to its largest eigen value, therefore the basic reproductive
ratio is given by
b ( + u2 + )
R0 = . (7)
2 p + (pu2 + q + + u1 ) + u2 ( + u1 )
107 We find that the disease free equilibrium will be locally asymptotically stable if and
108 only if R0 < 1.
109 Figures (2) are given to show the sensitivity between some parameters. The left
110 figure in Figures (2) shows the level set of R0 with respect to the value of healthy life
111 campaign u1 and treatment intervention u2 . It can be seen that larger the number
112 of these both intervention, then the value of basic reproductive ratio R0 will also
113 decrease significantly. Unfortunately, larger rate of u1 and u2 need more cost for
114 implementation. Therefore, we have to find an optimal way to implement these
115 intervention such that we can minimize number of overweight people and obese
116 people with cost as low as possible.
117 The right figure in Figures (2) is given to show the sensitivity between infection
118 rate because of bad life habit () and natural recovery rate (). It can be seen
119 that with constant value of natural recovery rate, the value of basic reproductive
120 ratio is increasing when transition to obese compartment because bad life habit
121 also increasing. Also, with constant value of transition because bad life habit rate,
122 the basic reproductive ratio will decrease as well as when natural recovery rate
123 increasing.
124 In the next section, characterization of optimal control problem to reduce the
125 number of overweight and obese compartments with intervention of health life cam-
126 paign and treatment intervention will be given.

127 4 Optimal control problem characterization

Our purpose in this model is to minimize number of overweight and obese compart-
ments and the corresponding control functions. Together with the mathematical
model of obese disease in equation 1, we consider an optimal control problem with
the objective functional given by
Z T
x x2 + y y 2 + z z 2 + u1 u21 + u2 u22 dt,


J(ui , ) = (8)
0
 Figure 2: Parameter sensitivity of R0 between u1 vs. u2 (left) and vs. (right).

128 with is the set of all compartment variables. x , y and z are weight constants
129 for healthy compartment, overweight compartment and obese compartment, respec-
130 tively. Here we want to minimize the number of infected population, therefore we
131 set y > 0 and z > 0 while x = 0. The weigth parameters u1 and u2 are weight
132 constants for health life campaign (u1 (t)) and treatment intervention (u2 (t)) which
133 regularize the optimal control condition. The cost for health life campaign could
134 come from advertising in the press media, electronic media or even from direct visit
135 in school or hospital. Similarly, the cost for treatment intervention coming from cost
136 for drug or any cost associated with treating the patients to get recovery as soon as
137 possible.
We would like to find the value of control variable u1 (t) and u2 (t) from t = 0 to
t = T such that

J(u1 (t), u2 (t)) = min {J(ui , )|(u1 , u2 ) } , (9)

138 subject to system 1 and where the control set is a piecewise continuous on [0,T]
139 and ai ui (t) bi for i = 1, 2. Tehese parameters, ai and bi are the lower and
140 upper bound of the control variables, respectively. The existence of optimal controls
141 (ui ), i = 1, 2, comes from convexity of the integrand of 8 with respect to controls
142 and regularity of the system 1.
To find the optimal control for u1 and u2 in order to minimize the cost function
in 8, we begin with defining a Lagrangian for the system 1 and 8. The Lagrangian
for the problem in here consists of the integrand of the objective functional and the
inner product of the right hand sides of the state equations and the adjoint variables
 = (x , y , z ). We define our Lagrangian as given by

L(, ) = x x2 + y y 2 + z z 2 + u1 u21 + u2 u22

+ x ( (x + py + qz) + ( + u1 (t)) y (by + bz) x x)



+ y ( ((1 p)y + (1 q)z) + (by + bz) x + u2 (t)z (10)
( + u1 (t)) y (bz + ) y y)




+ z ((bz + ) y u2 (t) z) .

The adjoint equation variables = (x , y , z ) correspond to the system of state

equations 1 is taken from partial derivative of the Lagrangian function respect to
each state variables and given by

x = L

x
= 2 x x + x ( ) + (x y )(by + bz)
L

y = y = 2y y x py + (y 1 )( + u1 (t))...



+ bx(x y ) + (y z )(bz + ) y ((1 p) ) (11)
z = z = 2z z 1 q + bx(x y ) y (1 q)...
L





+ by(y z ) + u2 (t)(z y ) + z .

143 The terminal condition of adjoint equation system 11 is equal to zero (x (T ) =

144 0, y (T ) = 0 and z (T ) = 0).
To obtain the optimality condition in order to minimize the cost function 8, we
also will differentiate the lagrangian L respect to each control variables and set it
equal to zero. Therefore, we have
L 
u1
= 2u1 u1 + y(x y ) = 0
L . (12)
u2
= 2u2 u2 + z(y z ) = 0

Solving the equations in 12, we obtain that

o
y(y x ) z(z y )
u1 = 2u1
, u2 = 2u2 . (13)

Consider the lower bound ai and upper bound bi for i = 1, 2 respect to each control
variable, the optimal control now characterized as
  
y(y x )
u1 = max a1 , min b1 , 2u
  1  . (14)
u2 = max a2 , min b2 , z(2z
u
y)
2

145 Now, we point out that our optimal control system is consist of the state system
146 1 with initial condition given for t = 0, the adjoint system 11 with final condition
147 given for t = T and the optimality condition in equation 14. The optimal control
148 problem will be solved with iterative method forward in time for state equation
149 and backward in time for adjoint equation. The controls are updated in each itera-
150 tion until the convergence condition reached. Further application of optimal control
151 theory in contagious disease could be seen in [12]. In the next section, some numer-
152 ical simulation for optimal control problem will be shown for different case which
153 represent the possibility condition in the field.
 300 1000 0.08
Overweight before treatment Obesse before treatment u1, u1=u2
u1=u2 u1=u2
u1, u1<u2
u1<u2 u1<u2
900 u2, u1=u2
0.07
u2, u1<u2
250
800

0.06

700
200

0.05
600

150 500 0.04

400
0.03

100
300

0.02

200
50

0.01
100

0 0 0
0 50 100 150 200 0 50 100 150 200 0 50 100 150 200

Figure 3: Numerical simulations of overweight (left), obese (middle) compartments

and control variable (right) with different condition of u1 and u2 .

154 5 Numerical simulation

155 The dynamical model in this paper considered the obese spread via daily social
156 interaction and the data for simulation is given in table 1. In the following three sub
157 sections, we discuss the simulation results for different cases i.e for different weight
158 cost for each control variable, different initial condition for each compartments and
159 different value for some parameter in autonomous system 1.

160 5.1 Variation of weight cost for control variable

161 This condition describe the situation where the cost for application in the field is
162 different between campaign and rehabilitation intervention. We balance the human
163 populations and control functions in cost functional with choosing the weight cost
2
164 such that ux fu2 with f because the magnitudes of human populations and
165 the control variables are on different scales.
166 The numerical results for this subsection are obtained for different values of ui
167 while keeping the other parameters and initial conditions for state and adjoint vari-
168 ables remain same. From Figure (3) (middle), it can be seen that intervention of
169 health life campaign and treatment intervention success to minimize the number
170 of obese people. The number of overweight people increase from case before in-
171 tervention of rehabilitation programme was applied (see left figure in Figure (3)).
172 This phenomenon appears because some people from obese compartment remove to
173 overweight compartment as a result of rehabilitation programme. From Figure (3)
174 (right), it is shown that if u1 < u2 , then the rate of u1 will be much more larger
175 than when u1 = u2 . As a result, it can be seen that the number of obese people
176 is become smaller and overweight increase (not significantly) and also with smaller
 300 700 0.06
overweight, prevention scenario
obesse, endemic reduction scenario overweight, prevention scenario
obesse, endemic reduction scenario

600
250 0.05

500

200 0.04

400

150 0.03

300

100 0.02

200

50 0.01 u1, prevention scenario

100
u1, endemic reduction scenario
u2, prevention scenario
u2, endemic reduction scenario

0 0 0
0 50 100 150 200 0 50 100 150 200 0 50 100 150 200

Figure 4: Numerical simulations of overweight (left), obese (middle) compartments

and control variable (right) for prevention and endemic reduction scenario.

177 cost, i.e from 388611 become 195520.

178 5.2 Variation of the initial condition for the state equation
179 Simulations in this sub section are given to accommodate the different initial condi-
180 tion for state equation, let call it as prevention scenario where number of obese people
181 is relatively low (x(0) = 970, y(0) = 20, z(0) = 10) and endemic reduction scenario
182 where number of obese people is relatively high (x(0) = 830, y(0) = 20, z(0) = 150)
183 . The simulation is given with same value for all parameters and also u1 = u2 .
184 From Figure (4), it can be seen that endemic reduction case need larger rate of u1
185 and u2 to reduce the number of obese people in the final time (t=T) as same as like in
186 the prevention scenario. As a result, the cost function become much more larger from
187 388611 (prevention scenario) to 514940 (endemic reduction scenario). Therefore, to
188 make the programme of reducing number of obese incidence more success, it will be
189 better to give the intervention of health life campaign and treatment programme in
190 beginning of endemic (prevention scenario) because the cost is much lower than if
191 we give the interventions when endemic already happen. This is often overlooked
192 by the policy makers in terms of mass action for obesity management process.

193 5.3 Variaton of interaction coefficient parameters

194 We perform the last simulations in this sub section for the autonomous system 1
195 for case when R0 < 1 and R0 > 1. For case when R0 = 1.58, all parameters are
196 same with description in table 1 while for R0 = 0.79, the different only value of b
0.05
197 which is equal to 1000 . It can be seen in Figure (5) that if R0 < 1 then the dynamic
198 of system 1 will tends to disease free equilibrium while if R0 > 1 goes to endemic
 1000 35 250

health, Ro<1 overweight, Ro<1 obesse, Ro<1

health, Ro>1 overweight, Ro>1 obesse, Ro>1
30

950 200

25

20
900 150
human

human

human
15

850 100
10

800 50

750 5 0
0 500 1000 1500 2000 2500 3000 3500 4000 0 500 1000 1500 2000 2500 3000 3500 4000 0 500 1000 1500 2000 2500 3000 3500 4000
time time time

Figure 5: Numerical simulations of healthy (left), overweight (middle) and obese

(right) compartments when R0 < 1 and R0 > 1 .

199 equilibrium where all compartments exist in the field. From this simulation also, we
200 can conclude that minimizing intense of social contagion will decreasing number of
201 overweight people and obese people. This results show that social interaction with
202 overweight and obese people could increase number of new incident of obesity. It
203 has been suggested that changing this social interaction rate may lead to decline of
204 obesity.
205 Other alternative way to reduce R0 could be choosen according to the form of
206 R0 in equation 7. For an example, maximizing portion of healthy recruitment rate
207 (p and q), maximizing healthy life campaign (u1 ), maximizing recovery rate from
208 obese compartment to healthy compartment () and so on.

209 6 Conclusion
210 Mathematical model for obesity in a closed population has been constructed in
211 this paper. Vertical transmission from infectious parents has been accommodated
212 in to the model as well as the intervention program like healthy life campaign and
213 treatment/rehabilitation program. Disease free equilibrium and endemic equilibrium
214 have been shown analytically. Basic reproductive ratio as the endemic indicator has
215 been obtained for local stability criteria of disease free equilibrium.
216 With the optimal rate, it is shown that the intervention program with healthy
217 life campaign and rehabilitation has significantly reduced number of obese people.
218 The numerical results show that it is much better to start the program in the early
219 situation where number of obese people is relatively low rather than to wait until
220 the number of obese people is relatively high.
221 Further work should be done to include different social interaction rate for each
222 compartment, different age class as well as to accommodate the cross marriage
223 among the parents.

224 7 Acknowledgements
225 We thanks to the reviewer for all the constructive comments. This research is funded
226 by the Research Grant of the Indonesian Directorate General for Higher Education.

227 References
228 [1] C. William, et al, Determinants of Diabetes Mellitus in the Pima Indians, Dia-
229 betes care, Volume 16, suplement 1, January 1993, 216-227.
230 [2] C.Robert et al, Predicting Obesity in Young Adulthood From Childhood and
231 Parental Obesity, The New England Journal of Medicine, Vol 337, 869-873 (1997)
232 [3] Susan S Deusinger, Robert H Deusinger, Susan B Racette. The Obesity Epi-
233 demic: Health Consequences and Implications for Physical Therapy, hys Ther.
234 2003;83:276-288.
235 [4] O. Diekmann, J. A. P. Heesterbeek, Mathematical Epidemiology of Infectious
236 diseases, Model Building, Analysis and Interpretation, John Wiley and Son,
237 Chichester (2000).
238 [5] O. Diekmann, J. A. P. Heesterbeek and M. G. Roberts, The Construction of
239 Next-Generation Matrices for Compartmental Epidemic Models, J. R. Soc. Inter-
240 face 2010 7, 873885.
241 [6] A.Hurford, D. Cownden, T. Day, Next-generation Tools for Evolutionary Inva-
242 sion Analyses, J. R. Soc. Interface 2010, 561-571.
243 [7] World Health Organization: Obesity: Preventing and Managing the Global Epidemic.
244 Report of a WHO Consultation. WHO Technical Report Series no. 894. Geneva: WHO;
245 2000. available from: http://www.who.int/nutrition/.

246 [8] World Health Organization: Obesity and Overweight: Media Center: Keys Fact. avail-
247 able from: http://www.who.int/mediacentre/.

248 [9] Hill AL, Rand DG, Nowak MA, Christakis NA: Infectious Diseas Modeling of Social
249 Contagion in Networks, PLoS Comput. Biol., 2010, 6:e1000968.

250 [10] Christakis NA, Fowler JH: The Spread of Obesity in a Large Social Network Over 32
251 Years. N Engl J Med 2007, 357:370379.

252 [11] Taylor RW, et al , APPLE project: Two-Year Follow-up of An Obesity Prevention
253 Initiative in Children:The APPLE Project, AM. J. Clin. Nutr., 2008, 88:1371-1377.

254 [12] D.Aldila, T.Gotz, E.Soewono, Optimal Control Problem Arising from Dengue Disease
255 Transmission Model, Mathematical Biosciences, 242 (2013) 916.