[Zurück]

S. Parragh, P. Einzinger:
"A Comparison of System Dynamics and Agent-Based Modeling on a Model of Health Care Utilization";
Vortrag: MATHMOD 2012 - 7th Vienna Conference on Mathematical Modelling, Wien; 14.02.2012 - 17.02.2012; in: "Preprints Mathmod 2012 Vienna - Full Paper Volume", F. Breitenecker, I. Troch (Hrg.); Argesim / Asim, 38 (2012), S. 399 - 400.

Introduction. The choice of an appropriate method for modeling the utilization of health care services is affected
by the availability of data and its level of detail. Therefore the question arises whether or not and if so, how the
results differ for different methods.
System Dynamics and Agent-based Modeling. System dynamics (SD) and agent-based modeling (ABM) are
two entirely different approaches to model a system. SD is a top-down modeling method, which means that the
system is described from a global perspective, requiring knowledge of the global relations and causalities (see e.g.
[2]). ABM on the contrary is a bottom-up approach, where the single acting entities of the system, i.e. the agents,
and their behavior are modeled. The global behavior of the system then results from the agents´ interactions during
simulation. Regarding the simulation of human systems, which are based on individual preferences and decisions,
ABM provides a natural description. But it requires individual disaggregated data and soft factors concerning the
human psychology for parametrization. Furthermore, depending on the complexity of the system, the simulation
can be extremely computation intensive [1]. SD simulation needs less computational power and also the data
requirements are often easier to meet. However, it can be a very difficult task to understand and to quantify the
causalities needed for the implementation. For comparison a demonstration model of health care utilization is
implemented as both an SD and an ABM model with the same parametrization.
A Model of Health Care Utilization. The system consists of patients, who can be healthy or sick, and medical
providers. To keep the model as simple as possible, only two degrees of severity of the same disease and two
corresponding treatments are considered. The patients try to maximize their life quality, which depends on their
health state and consequently on the received treatment. The goal of the medical providers is to obtain an optimal
mix of achieved income and effectiveness.
Both methods are implemented in AnyLogic, a multi-method simulation modeling tool by XJ Technologies. The
agent-based model consists of the agents ´patient´ and ´medical provider´. The patient is ruled by two statecharts,
one specifying his current health state, the other one depicting his satisfaction with his attending physician, i.e. the
medical provider he consults in case of illness. If he is not satisfied, he can change provider in order to achieve
a higher life quality. The agent medical provider is able to diagnose the degree of severity (with a certain error
rate), to decide on the therapy and to treat a patient. He aims to optimize his performance with respect to his
income, the utilization ratio of his working hours, and the effectiveness of his treatments. To attain this goal, he
adjusts the criterion, on which his treatment decision is based. In the SD implementation three state variables
represent the patients according to their current health state. The medical providers influence the healing rates by
prescribing the right therapy. With the help of averaged illness durations and prevalences, a link between the goal
of the patients and the resulting behavior of the providers is established. Aggregated income, effective treatments
and the influence of the patients then control treatment decisions.
Results. First simulation experiments have shown similar behavior of both models with the default parametrization.
Also the adaptions of the system to small parameter variations resemble each other. But reactions on drastic
changes, like for example a sudden undersupply of medical providers, differ strongly regarding the quantitative
outcome, even though the overall trends are the same. Further analyzes and alterations of the models are needed to
determine if those differences are inherent in the chosen approach.
Conclusion. Both techniques provide the means to describe the dynamics of health care utilization in the specified
system. ABM offers the more intuitive approach plus the possibility to easily add or change factors on an
individual level, like including effects of distance or social networks on the provider choice (see e.g. [3]). For
the implementation of the SD model, the main system components and global feedback mechanisms have to be
identified. On the one hand, this requirement helps to gain a better insight into the system on a global level. On the
other hand, dynamics due to individual actions or heterogeneities are difficult to describe and to capture.