Mediastinet was originally an influence diagram. It has been converted into a decision analysis network (DAN) by removing the total ordering of decisions and some dummy states, and by declaring always-observed variables, revelation conditions, and restrictions.

References

M. Luque. Probabilistic Graphical Models for Decision Making in Medicine. PhD thesis, UNED, Madrid, 2009.

M. Luque, F. J. Díez and C. Disdier. A decision support system for the mediastinal staging of non-small cell lung cancer. In Proceedings of the 8th Bayesian Modelling Applications Workshop, (Barcelona, Spain).

F. J. Díez, M. Luque, C. L. König and I. Bermejo. Decision analysis networks. Technical Report CISIAD-14-01, UNED, Madrid, 2014.

Description of the problem

A doctor has to decide how to treat a patient that suffers from non-small cell lung cancer. The key to select a treatment is whether there is metastasis at the mediastinum or not. A CT scan is always performed. Additionally, the doctor can perform other tests to study the mediastinum: TBNA, PET, mediastinoscopy (MED), EBUS and/or EUS. The physician can also decide the order of the tests, with the following constraints: no test be performed before the TAC; an EBUS or an EUS can be performed only after a PET. Then the physican has to decide the treatment. The decision must take into account the morbidity and economic cost of each test and each treatment, the quality adjusted life expectancy, the probabilities of survival, and the willingness-to-pay (lambda).

Structure of the network

The structure of chance and decision variables reflects that the target and unknown variable is N2N3, which indicates whether there is metastasis or not. Chance variables representing the results of tests have a probabilistic relation with the N2N3 variable and result of the TAC. We must also note that the results of MED, EBUS and EUS depend on the result of the PET.

The structure of utility nodes has been built by using super-value nodes [Tatman and Shachter, 1990]. It consists of two main parts. The node Total_Economic_Cost and its ancestors represent the economic cost of the medical tests and treatments. The node Total_QALE and its ancestors represent the quality adjusted life expectancy of the patients. Both parts are summed by weighting the Total_Economic_Cost by the node C2E (cost to effectiveness), which represents the inverse of the lambda parameter (willingness to pay) used in cost-effectiveness analysis. The node Net_Effectiveness represents the global utility.

The Quality Adjusted Life Expectancy (QALE) in utility nodes is measured in Quality Adjusted Life Years (QALYs). Costs are in euros.