Contradictory results in decision tree

[Manuel Luque]

When showing the decision tree (for the attached network) for lambda = 30,000 €/QALY, OpenMarkov says that the utility of doing first test E is the same as when doing first test C. This contradicts the results shown on the CEP.

[Francisco Javier Díez]

I attach another DAN leading to an incorrect decision tree for the unicriterion case: it says that the NMB of doing test B is higher than when not doing it, which is clearly wrong because the result of this test does not affect the decision about therapy.

Given that the increase in utility for this branch is almost the same as the cost of the test, I conjecture that the problem might have to do with the sign of the weights used for unicriterization.

The results of cost-effectiveness analysis seem to be correct.

[Francisco Javier Díez]

The problem was not in the inference algorithm, but in the coding of the network.