[Zurück]

P. Einzinger, N. Pfeffer:
"Modeling Health Care Systems - An Approach Using Routine Health Care Data";
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. 181 - 182.

Introduction. The use of dynamic models is already quite common in the field of health technology assessment.
However, there has been less research on modeling and simulation addressing global questions of health care
systems, for example which consequences a change of the reimbursement system could have. In health economics
formal models do exist, but in most cases they provide just qualitative results without incorporating quantitative
data. On the contrary, statistical methods are often used when analyzing large data sets. This approach is useful but
cannot give insights into the behavior of the system at a different operating point where data do not exist, because
its purpose is modeling the data and not the structure of the system.
In Austria there exist comprehensive routine health care data from reimbursement claims. Simulation models that
can incorporate these data are therefore very beneficial. We present an approach that was used for building and
parameterizing a model of extramural health care and its reimbursement system.
Approach. The model is agent-based and includes patients and medical service providers - physicians and other
medical institutions of extramural health care - as agent types. Patients may develop one or many of several
different chronic diseases that are incorporated into the model. These diseases provoke a certain service need in
the patients. Therefore they consult the medical service providers in order to fulfill the service demand of all their
illnesses. These providers get payment for the visits of the patients, where the actual reimbursement depends on
the reimbursement system.
Some parts of the model are difficult to parameterize, especially what patients actually do if they have a certain
disease (i.e. which treatment pathways they take and which services they get). This could be provided by expert
opinion, but our approach tries to parameterize as much as possible from available routine health care data. The
basis is probabilities for the association of medical diagnoses and prescribed drugs derived in [1], as the data in
extramural care do not contain diagnosis codes. Thus it is possible to link patients and their medical claims to
diseases.
The model uses prevalence of diseases (for determining how many patients have a certain disease at simulation
start) and incidence rates (for determining at which rate patients get new diseases). Incidence rates are calculated
from the data by counting patients that have a link to the corresponding diagnosis in one year but not before and
dividing that by the time these patients were at risk. Prevalence is calculated by counting all the cases of a disease
in one year, correcting it for the incidence of one year and dividing that by the total number of patients in the data.
The model input for the service need of the diseases is given by frequency distributions of services per quarter of
year obtained from samples of patients with only one disease at a time. Every disease of a patient draws its service
need from these frequency distributions every 90 days of simulation time. Each medical provider has their own
portfolio of offered services. Probabilities for providers to offer a certain service, depending on specialty, equal the
fractions of providers in the database who accounted the service at least once. When the medical problem has
determined its new service need a provider search is performed. The chosen providers must be in a certain range of
the patient, have acceptable waiting times (i.e. not too many patients in their queues) and must cover the service
need in an optimal way - it has to be covered with as few providers as possible. Formally this is a minimum set
cover problem with the service need as a universe and the subsets of offered services of each provider as the
subsets from whom the covering family of sets is taken [2]. Patients consult the providers found regularly until
their service need is fulfilled and thus treatment pathways emerge automatically.
Conclusion. The approach shows how analyses of routine health care data can serve as input for a complex and
comprehensive model structure where it is not possible to do extensive literature searches and gathering of expert
opinion as there are many different diseases to consider. Derivation of prevalence and incidence rates for diseases
from the data is straightforward. For other parts such as treatment pathways additional assumptions are necessary,
but for example the optimal selection of providers according to their service portfolio leads to at least plausible
results.