AN AGENT BASED SIMULATION OF THE DYNAMICS
IN COGNITIVE DEPRESSOGENIC THOUGHT
Azizi Ab Aziz and Michel C. A. Klein
Agent Systems Research Group, Department of Artificial Intelligence, Faculty of Sciences
Vrije Universiteit Amsterdam, De Boelelaan 1081, 1081 HV Amsterdam, The Netherlands
Keywords: Agent based Simulation, Affective Disorder, Cognitive Depressogenic Formation, Social Support
Feedbacks.
Abstract: Depression is a common mental disorder. Appropriate support from others can reduce the cognitive
distortion that can be caused by subsequent depressions. To increase our understanding of this process, an
agent model is presented in this paper in which the positive and negative effects of social support and its
relation with cognitive thoughts are modelled. Simulations show the effect of social support on different
personality types. A mathematical analysis of the stable situations in the model gives an additional
explanation of extreme cases. Finally, a formal verification of expected relations between support, risk
factors and depressive thoughts is performed on the simulation traces to check whether the simulations
describe realistic processes.
1 INTRODUCTION
Cognitive vulnerability is one of the main concepts
that play an important role to escalate the risk of
relapse in affective disorder (depression). In a
broader spectrum, it is a defect belief, or structures
that are persistently related for later emergent in
psychological problems. Before further reviewing
the underlying concepts of the vulnerability, it is
essential to understand its connection between
relapse condition in unipolar depression and social
support (Aziz et al., 2009). Unipolar depression is a
mental disorder, distinguished by a persistent low
mood and loss of awareness in usual activities
(Beck, 1987). Normally, under a certain degree of
stressors exposure, an individual with a history of
depression will develop a negative cognitive content
(thought), associated with the past losses. Such
cognitive content is often related to the maladaptive
schemas, which in a long run will cause individual’s
ongoing thought capability to be distorted and later
to be dysfunctional (Robinson and Alloy, 2003).
However, this cognitive distortion can be reduced
through appropriate supports from other members
within the social support network (Heller and Rook,
1997). Social support network is made up of friends,
family and peers. Some of it might be professionals
and support individuals in very specific ways, or
other people in this network might be acquaintances
in contact with every day. It has been suggested that
social support naturally can help to prevent and
decrease stress through positive inferences, which
later curbs the formation of cognitive vulnerability
(Alloy et al., 2004). However, some literatures have
shown that certain supports provide contrast effects.
Rather than attenuating the negative effects from
stressors, it will eventually amplify the individual’s
condition to get worse (Coyne, 1990).
In this paper, these positive and negative effects
from social support interaction and its relation with
cognitive thought are explored. To fulfil this
requirement, a dynamic model about cognitive
depressogenic thought is proposed. The proposed
model can be used to approximate a human’s
cognitive depressogenic thought progression
throughout time. This paper is organized as follows.
The first section introduces main concepts and
existing theory of cognitive depressogenic thought
and hopelessness. Thereafter, a formal model is
described and simulated (Section 3 and 4). The
model has been verified by a mathematical analysis
(Section 5) and by checking properties of simulation
traces (Section 6). Finally, Section 7 summarizes the
paper with a discussion and future work for this
model.
232
Ab Aziz A. and C. A. Klein M. (2010).
AN AGENT BASED SIMULATION OF THE DYNAMICS IN COGNITIVE DEPRESSOGENIC THOUGHT .
In Proceedings of the 2nd International Conference on Agents and Artificial Intelligence - Agents, pages 232-237
DOI: 10.5220/0002736102320237
Copyright
c
SciTePress
2 FUNDAMENTALS IN
COGNITIVE DEPRESSOGENIC
THOUGHT
People vary in their abilities to overcome stressful
life events and it allows them to manage their
troubles and not be overwhelmed. These variations
answer why the level of severity and duration among
different individuals can be diverse in nature. To
explain this mechanism, the Extended Hopelessness
Theory of Depression is used. In this theory, people
who exhibit a negative inferential style, in which
they describe, attribute negative events to stable
(likely to persist over time) and global (likely to
affect many aspects of life) will most likely to infer
themselves as fundamentally useless and flawed
(Abramson et al. 1999).
Although it is well documented that social support
mitigates a risk of relapse, but there is a condition
where feedbacks from the social support members
may indirectly escalate the risk of relapse. Such
feedbacks are considered as “maladaptive inferential
feedback” (MIF), and normally increase the negative
thought formation (Alloy et al., 2004). Contrary to
this, an adaptive inferential feedback (AIF) provides
a buffer to reduce the threat, by countering negative
inferences for negative event. AIF asserts that when
a social support member offers comfort by
attributing the source of negative event to be
unstable, it will later diminish the risk of creating
maladaptive inferences (Dobkin et al. 2004).
In addition, the Extended Hopelessness Theory
of Depression relates the development cognitive
depressogenic thought through previously described
two precursors. First, the present of positive social
support feedback (AIF) acts as a buffer to decrease
individuals’ possibility of having cognitive
depressogenic thought over time. Second,
individuals with cognitive depressogenic thought
will make negative inferences when facing negative
events. This condition is also associated with less
AIF from the social support members. Moreover,
both of these conditions capable to predict changes
in stressful events. Therefore, it can be further used
to elaborate the immunity level of individuals (as
contrast in vulnerability concept). In addition, many
studies have also associated the lower risk of
depression with the presence of AIF (Coyne, 1990).
As indicated in several previous works,
inferential feedbacks provide one of the substantial
factors towards the development of cognitive
depressogenic thought over time. By combining
either one of these two factors together with
situational cues, it leads to the formation of either
cognitive depressogenic inference or positive
attributional style. Situational cues refers to a
concept that explains individuals’ perception that
highly influenced by cues from events
(environment). Individuals under the influence of
negative thought about themselves will tend to
reflect these negative cognitions in response to the
occurrence of stressors. These later develop the
conditions called “stress-reactive rumination” and
“maladaptive inference” (Spasojevic and Alloy,
2001).
Stress reactive rumination reflects a condition
where individuals have difficulty in accessing
positive information, and further develop a negative
bias towards inference (maladaptive inference). This
process is amplified by previous exposures towards
cognitive depressogenic thought episode. After a
certain period, both conditions are related to the
formation of hopelessness. Hopelessness is defined
by the expectation that desired outcome will not
occur, or there is nothing one can do to make it right
(Panzarella et al. 2006). Prolong and previous
exposure from hopelessness will lead to the
development of cognitive depressogenic thought.
However, this condition can be reduced by having a
positive attributional style, which normally existed
during the presence of AIF and low situational cues
perception (Crossfield et al. 2002).
In short, the following relations can be identified
from the literature: (1) prolong exposure towards
MIF, negative events, and high-situational cues can
lead to the development of cognitive depressogenic
thought. (2) a proper support (AIF) will reduce the
risk of further development of future cognitive
depressogenic thought. (3) Individuals with high
situational cues and proper support will be less
effective in reducing the progression of cognitive
depressogenic thought, compared to the individuals
with less situational cues.
3 FORMAL MODEL
This section discusses the details of the dynamic
model. In this model, three major components
namely; environment, inferential feedbacks, and
thought formation will represent the dynamic of
interactions between social support feedback and
individuals involved in negative thought formation
during the beginning of relapse and recurrence in
depression. In the formalization, those important
concepts are translated into several interconnected
nodes. These nodes are designed in a way to have
values ranging from 0 (low) to 1 (high). Figure 1
AN AGENT BASED SIMULATION OF THE DYNAMICS IN COGNITIVE DEPRESSOGENIC THOUGHT
233
depicts the global interaction between these nodes.
3.1 Temporal Specification
In order to develop a model, a temporal specification
language called LEADSTO and its supporting
software environment has been used. LEADSTO
enables one to model direct temporal relationship
between two state properties (dynamic properties).
Consider the format of α→
e,f,g,h
β, where α and β
are state properties in form of a conjunction of atoms
(conjunction of literals) or negations of atoms, and
e,f,g,h represents non-negative real numbers. This
format can be interpreted as follows;
If state α holds for a certain time interval with
duration g, after some delay (between e and f),
state property β will hold a certain time interval of
length h.
For a more detailed discussion of this language, see
(Bosse et al., 2007). To formalize the concepts of
properties on dynamics relationship introduced in
the previous section (Section 2), for each of them, a
logical atom using predicate calculus is introduced.
To formalize the dynamic relationship between these
concepts, the following temporal relationships are
used.
NEVT: Negative Events
A set of generated events is experienced by an agent
X through simulation of several conditions using
weighted sum w (where
w
=1) of life L, chronic C,
and daily D events.
X:AGENT
life_event(X,L) chronic_event(X,C) daily_event(X,D)
neg_event(X, w
1
.L+ w
2
.C+ w
3
.D)
PTS: Positive Attributional Style
If the agent X faces bad situational cues B, negative
events Ne, cognitive depressogenic thought Cd,
adaptive inferential style AiF, and has a proportional
contribution towards positive attributional style
η
then the positive
attributional style level is
η
*AiF+(1-
η
).(1-(B*Ne*Cd)) *AiF
X:AGENT
sit_cues(X, B) neg_event(X, Ne) adapt_inf(X, AiF) η
cog_dep_tgt(X, Cd)
pos_att_style(X, η*AiF+ (1-η).(1-(B*Ne*Cd))*AiF )
CDI: Cognitive Depressogenic Inferences
If the agent X experiences the intensity levels of
experiences negative inferential style MiF,
situational cues B, cognitive depressogenic thought
Cd, negative events Ne and has a proportional
contribution towards inferences
α
then the cognitive
depressogenic inferences level is
α
*MiF + (1-
α
).(B*Ne*Cd))*MiF
X:AGENT
sit_cues(X, B) neg_event(X, Ne) maladap _fb(X, MiF)
α cog_dep_tgt(X, Cd)
cog_dep_inf (X, α*MiF + (1-α).(B*Ne*Cd))*MiF)
STR: Stress Reactive Rumination
If the agent X experiences the intensity levels of
cognitive depressogenic thought Cd, and cognitive
depressogenic inference CDi and has a proportional
regulator
β
then the stress reactive rumination level
is
β
* CDi + (1-
β
)* Cd
X:AGENT
cog_dep_inf (X, CDi) cog_dep_tgt(X, Cd) β
sts_reactive(X, β* CDi + (1-β)* Cd )
MDI: Maladaptive Inference
If the agent X faces stress reactive rumination in SR
level and perceives positive attributional style PS
level and has a proportional contribution regulator γ
then the maladaptive inference level is
γ
*SR *(1-PS)
X:AGENT
sts_reactive(X, SR) cog pos_att_style(X, PS) γ
maladap _inf(X, γ*SR *(1-PS))
IMT: Immunity
If the agent X experiences the intensity levels of
cognitive depressogenic thought Cd, and initially has
BiM level of base immunity and has a proportional
Adaptive
inferential
feedback
Maladaptive
inferential
feedback
Negative life
events
Positive
attributional
styles
Cognitive
depressogenic
inferences
Hopelessness
Stress-reactive
rumination
Dysphoric
depressogenic thought
Immunity
Maladaptive
inference
Situational
cues
Social support
provision
Figure 1: Overview of the Cognitive Depressogenic Thought Model.
ICAART 2010 - 2nd International Conference on Agents and Artificial Intelligence
234
regulator
λ
then the immunity level (IM) is
λ
*BiIM+(1-
λ
)*(1-Cd)*BiM
X:AGENT
cog_dep_tgt (X, Cd) base_im (X, BiM) λ
immunity(X, λ*BiM+(1-λ)*(1-Cd)*BiM)
HPS: Hopelessness
If the agent X faces level of maladaptive inference
MDi and has previous level of hopelessness Hp and
has adaptation rate
ψ
then the hopelessness level for
agent X after
Δ
t is Hp + (1-Hp)*
ψ
*(MDi-Hp)*Hp*
Δ
t
X:AGENT
maladap _inf(X, MDi) hoplness(X, Hp) ψ
hoplness(X, (1-Hp)*ψ*(MDi-Hp)*Hp*Δt)
CDT: Cognitive Depressogenic Thought
If the agent X faces level of hopelessness Hp and
has previous level of cognitive depressogenic
thought Cd and has adaptation rate
ϕ
then the
cognitive depressogenic thought level for agent X
after
Δ
t is Cd + (1-Cd)*
ϕ
*(Hp-Cd)*Cd*
Δ
t
X:AGENT
hoplness(X, Hp) cog_dep_tgt (X, Cd) ϕ
cog_dep_tgt (X, Cd + (1-Cd)*ϕ*(Hp-Cd)*Cd*Δt)
4 SIMULATION TRACES
In this section, the model was executed to simulate
several conditions of agents with the respect of
exposure towards negative events, feedbacks from
the social support members, and situational cues.
With variation of these conditions, some interesting
patterns can be obtained, as previously defined in the
earlier section. For simplicity, this paper shows
several cases of cognitive depressogenic thought
levels formation using three different agent
attributes. These cases are; (i) an agent Heidi with a
good feedback from the social support members, and
using a good judgment about the situation (B=0.2,
MiF=0.1, AiF=0.8), (ii) an agent Kees that receives
good feedbacks but with bad judgment about the
situation (B=0.8, MiF=0.1, AiF=0.9), and (iii) an
agent Piet with bad feedbacks from the social
support, and bad judgment about the situation
(B=0.9, MiF=0.8, AiF=0.1). The duration of the
simulated scenario is up to t = 1000 (to represent the
conditions within 42 days) with two negative events.
The first event consisted of the prolonged and
gradually decreased stressors, while the second
event dealt with the decreased stressor. For all
conditions, the initial cognitive depressogenic
thought was initialized as 0.5.
Case #1: Prolonged Repeated Stressor with
Different Individuals Inferential Feedback
and Situation Cues
During this simulation, each type of individual
attribute has been exposed to a prolonged stressor
condition. The result of this simulation is shown in
Figure 2.
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
1.1
1 101 201 301 401 501 601 701 801 901
Stressors
Piet
Kees
Heid
i
Figure 2: Cognitive Depressogenic Level for Each
Individual during Prolonged Stress Events.
In this simulation trace, it shown that Piet (high
situational cues, and negative inferential feedback)
tends to develop a cognitive depressogenic thought,
in contrast with the others. Heidi (low situational
cues, and positive inferential feedback) shows a
rapid declining pattern in developing the cognitive
condition. Note that Kees (high situational cues and
positive inferential feedback) has also developed a
decreasing pattern towards the cognitive condition.
However, Kees has a lesser decreasing effect
towards a negative thought despite a high positive
support, given that this individual tends to perceive
negative view about the situation. Persistent positive
support from the social support members helps each
agent to reduce the development of cognitive
thought throughout time
Case #2: Decreased Stressor with Different
Individual Inferential Feedback and
Situational Cues
In this simulation trace, there are two conditions
were introduced, one with a very high constant
stressor, and with no stressor event. These events
simulate the condition of where agents were facing a
sudden change in their life, and how inferential
feedbacks and perceptions towards events play
important to role towards the diminishing of
cognitive thought. The result of this simulation is
shown in Figure 3.
AN AGENT BASED SIMULATION OF THE DYNAMICS IN COGNITIVE DEPRESSOGENIC THOUGHT
235
0
0.2
0.4
0.6
0.8
1
1.2
1 101 201 301 401 501 601 701 801 901
Stressors
Heidi
Kees
Piet
Figure 3: Cognitive Depressogenic Level for Each
Individual during Fluctuated Stressors.
A comparison for each agent shows that Piet gets
into a sharp progression towards a high cognitive
thought after direct exposure towards a heighten
stressor. At the start of a high constant stressor, both
individuals Heidi and Kees develop cognitive
thought. However, after certain time points, those
progressions dropped and reduced throughout time.
As for Piet, even the stressors have been diminished,
the level cognitive depressogenic thought was still
high for several time points until it decreased.
5 MATHEMATICAL ANALYSIS
By a mathematical formal analysis, the equilibria of
the model can be determined. The equillibria
explains condition where the values for the variables
which no change occur. Assuming all parameters are
non-zero, the list of LEADSTO specifications for
the case of equilibrium for the agent X are:
dCd(t)/dt=(1-Cd)*
ϕ
*(Hps-Cd)*Cd (1)
dHp(t)/dt = (1-Hp)*
ψ
*(MDi-Hp)*Hp (2)
Assuming both adaptation rates are equal to 1,
therefore, these are equivalent to;
Cd=1 or Hp=Cd or Cd=0 (3)
Hp =1 or MDi=Hp or Hp=0 (4)
From here, a first of conclusions can be derived
where the equilibrium can only occur when the
Cd=
1, Hp=Cd
, or Cd=0 (refer to Equation 3). In this
paper, only condition
Cd=1, has been chosen for the
discussion. From this case (
Cd=1), it can be further
derived that respective values for the equilibrium
condition to take place. These values can be
calculated from the following formulae.
CDi =
α
*MiF + (1-
α
)*(B*Ne*Cd)*MiF
PS =
η
*AiF + (1-
η
)* (1-(B* Ne*Cd)).AiF
SR =
β
*[
α
*MiF + (1-
α
)*(B*Ne*Cd)*MiF] + (1-
β
)
MDi =
γ
*[
β
*(
α
*MiF + (1-
α
)*(B*Ne*Cd)*MiF]
+(1-
β
))*(1-(
η
*AiF+(1-
η
)*(1-(B*
Ne*Cd)))*AiF)]
IM =
λ
* BiM
This equillibria describes the condition when
agents are experiencing an intense negative
cognitive thought throughout time will eventually
have their level immunity reduced to the lowest
boundary of agents’ limit. This condition creates
higher vulnerability towards the development of
onset during the present of negative events.
Simulation trace from the experiment #1 confirms
this condition
6 AUTOMATED VERIFICATION
This section deals with the verification of relevant
dynamic properties of the cases considered in the
human agent model, which coherence with the
literatures. The Temporal Trace Langue (TTL) is
used to perform an automated verification of
specified properties against generated traces. TTL is
designed on atoms, to represent the states, traces,
and time properties. This relationship can be
presented as a
state(
γ
, t, output(R)) |= p, means that
state property p is true at the output of role R in the
state of trace
γ
at time point t (Bosse et al., 2009).
Based on that concept, several dynamic properties
can be formulated using a sorted predicate logic
approach. Below, a number of them are introduced
in semi formal and in informal representations.
VP1: Positive Supports will Reduce the Risk
in Developing Future Depressogenic Thought
When an agent X received more positive supports
from its social support networks, then the agent will
unlikely to develop further hopelessness in future.
∀γ:TRACE, t, t’:TIME, R1,R2,R3,MIN_LEVEL:REAL,
X:AGENT
[ state(γ, t) |= adapt_inf (X, R1) & R1 > MIN_LEVEL
state(γ, t) |= cog_dep_tgt (X,R2) & R2 > 0]
t’:TIME > t:TIME
[state(γ, t’) |= cog_dep_tgt (X,R3) & R3 < R2]
This property can be used to verify future condition
of an agent if the agent receives positive supports
from its social support members throughout time.
Many research works have maintained that positive
supports from members will decrease possibilities of
having further negative thought in future (Heller and
Rook, 1997).
VP2: Negative Perception towards Situation
and Bad Support received from the Social
Support Networks will Increase the Risk of
Further Depressogenic Thought
When an agent X perceives all situations will give
negative impact and an agent X receives bad support
from its social support networks, then the agent X
ICAART 2010 - 2nd International Conference on Agents and Artificial Intelligence
236
will almost likely to develop future depressogenic
thought.
∀γ:TRACE, t, t’:TIME, R1,R2,R3,R4, MIN_MLD_LEVEL,
MIN_SC_LEVEL, MAX_CDT_LEVEL:REAL, X:AGENT
[state(γ, t) |= maladap_bf (X, R1) &
R1 > MIN_MLD LEVEL &
state(γ, t) |= sit_cues(X, R2) & R2 > MIN_SC_LEVEL &
state(γ, t) |= cog_dep_tgt (X,R3) & R3 <
MAX_CDT_LEVEL]
t’:TIME > t:TIME
[state(γ, t’) |= cog_dep_tgt (X,R4) & R4 > R3]
By checking property VP2, one can verify whether
negative perception (situational cues) and bad
support will influence the rise of depressogenic
thought. It is particularly significant to observe this
property in the model given that bad support and
negative perception is highly correlated towards the
development of depressogenic thought (Crossfield et
al., 2002).
7 CONCLUSIONS
In this paper, the assumed role of negative cognitive
content in depression is explained. Based on this, a
agent-model is presented that describes the temporal
relation between personal characteristics, negative
life events and social support. This model is used in
a small simulation to investigate the effect of
different types of support on different persons that
undergo similar life events. The mathematical
analysis of the model and the verification of
expected behaviour of the modelled agents in the
simulation traces give some evidence for the
appropriateness of the model.
In the future, we would like to extent the model
with the effect of negative thoughts and a bad mood
on the willingness to offer support. Together with
the existing elements of the model, this would allow
for a multi-agent simulation of a larger community,
in which different persons interact with each other
by giving and receiving support. Such analysis
would make it possible to investigate the
consequences of depressive persons in a small
community.
ACKNOWLEDGEMENTS
The preparation of this paper would not have been
possible without the support and ideas of Prof. Dr.
Jan Treur. Both authors would like to thank him for
ideas, and refinement of this paper.
REFERENCES
Abramson, L.Y., Alloy, L.B., Hogan, M.E., Whitehouse,
W.G., Donovan, P., Rose, D.T., Panzarella, C.,
Raniere, D. 1999. Cognitive vulnerability to
depression: Theory and evidence. Journal of Cognitive
Psychotherapy, 13, 5-20.
Alloy, L. B., Abramson, L. Y., Gibb, B. E., Crossfield, A.
G., Pieracci, A. M., Spasojevic, J., Steinberg, J. A.
2004. Developmental antecedents of cognitive
vulnerability to depression: Review of findings from
the cognitive vulnerability to depression project.
Journal of Cognitive Psychotherapy, 18(2), 115-133.
Aziz, A.A., Klein, M.C.A., Treur, J. 2009. An agent model
of temporal dynamics in relapse and recurrence in
depression. In: Ali, M., Chen, S.M., Chien, B.C.,
Hong, T.P. (eds.), IEA-AIE 2009. LNAI, Springer
Verlag, pp. 36-45.
Beck, A.T. 1987. Cognitive models of depression, Journal
of Cognitive Psychotherapy 1, pp. 5–37.
Bosse, T., Jonker, C.M., Meij, L. van der, Treur, J. 2007.
A language and environment for analysis of dynamics
by simulation. International Journal of Artificial
Intelligence Tools, vol. 16, pp. 435-464.
Bosse, T., Jonker, C.M., Meij, L. van der, Sharpanskykh,
A., Treur, J. 2009. Specification and verification of
dynamics in agent models. International Journal of
Cooperative Information Systems, vol.18, pp.167 - 193
Coyne, J.C. 1990. Interpersonal process in depression. In:
G.I. Keitner, Editor, Depression and families: Impact
and treatment, American Psychiatric Press,
Washington, DC , pp. 31–53.
Crossfield, A. G., Alloy, L. B., Gibb, B. E., Abramson, L.
Y. 2002. The development of depressogenic cognitive
styles: The role of negative childhood life events and
parental inferential feedback. Journal of Cognitive
Psychotherapy, 16(4), 487–502.
Dobkin, R.D, Panzarella, C., J. Fernandez, Alloy, L.B.,
Cascardi, M. 2004. Adaptive inferential feedback,
depressogenic inferences, and depressed mood: A
laboratory study of the expanded hopelessness theory
of depression, Cognitive Therapy and Research, pp.
487–509.
Heller, K., Rook, K.S. 1997. Distinguishing the theoretical
functions of social ties: Implications for support
interventions. In: S. Duck, Editor, Handbook of
personal relationships: Theory, research, and
interventions, John Wiley and Sons, Chichester,
England, pp. 649–670.
Panzarella, C., Alloy, L.B., Whitehouse, W.G. 2006.
Expanded hopelessness theory of depression: On the
mechanisms by which social support protects against
depression, Cognitive Therapy and Research 30 (3),
pp. 307–33
Robinson, M. S., Alloy, L. B. 2003. Negative cognitive
styles and stress-reactive rumination interact to
predict depression: A prospective study. Cognitive
Therapy & Research, 27(3), 275-292.
Spasojevic, J., Alloy, L. B. 2001. Rumination as a
common mechanism relating depressive risk factors to
depression. Emotion, 1(1), 25-37.
AN AGENT BASED SIMULATION OF THE DYNAMICS IN COGNITIVE DEPRESSOGENIC THOUGHT
237