Promotor: Prof. dr. P.J.F. Lucas
Copromotor: Dr. A.J. Hommersom
Prof. dr. T. Heskes – Radboud University
Prof. dr. S. Andreassen – Aalborg University
Prof. dr. M.G.M. Olde Rikkert – Radboud University Medical Center


The epidemiology of multiple chronic diseases present at the same time is referred to
as comorbidity or multimorbidity. With the ageing of people multimorbidity becomes
the rule rather than the exception, especially for the elderly. The human body is a complex
adaptive system and very often we only see a few symptoms as a tip of the iceberg.
Current statistical methodologies are not entirely suitable to analyse this phenomena as
they often consider only one (primary) disease. In this thesis we have explored the usefulness
of probabilistic network models in the field of multimorbidity. First we asked
ourselves the question how interactions between diseases, frequently present with multimorbidity,
can be best described. These interactions are often stochastic by nature and
it turns out that many of the interactions can be expressed very well by using probabilistic
networks, e.g., Bayesian networks. An important achievement of our research
is that learning the structure of a network from data can significantly contribute to unravelling
the intricate interactions that are hidden in clinical data. Another problem
we faced in this research is the fact that much of the clinical data comes from multiple
sources, e.g., from multiple general practices that use different kinds of electronic
health care systems. This introduces a certain bias, and to be able to deal with such data
we introduced a new concept called multilevel Bayesian networks. These networks can
deal with any big dataset that is hierarchically structured. We applied them by investigating
the simultaneous progression of chronic cardiovascular conditions, correcting
for both patient and practice-related variables. Because of the network structure the
progression is easier to understand. For example, it turned out that in the presence of
hypertension, the observed cumulative incidence rates of combinations of cardiovascular
disorders, i.e., multimorbidity, differ significantly from the expected rates. Another
aspect is that in many real-life systems, interactions often participate in feedback loops.
Here we adopted a qualitative viewpoint to model and understand such feedback loops.
Although qualitative reasoning has its limitations, we showed that without knowing
exact probabilities, we are still able to draw qualitative conclusions of the dynamics
that exist in a system. The ideas in this thesis are certainly generalizable to other areas
of scientific research. As an example we briefly discussed a simplified model of the
Arctic summer sea-ice decline and its regional effects on the polar bear populations