This problem was proposed by Demirer and Shenoy (2006) and implemented as a DAN in OpenMarkov by Caroline König, who added the variable Symptom as an example of an always-observed variable. Revised by Javier Díez in November 2015.

References

R. Demirer and P. P. Shenoy. Sequential valuation networks for asymmetric decision problems. European Journal of Operational Research, 169 (2006) 286–309.

Description of the problem

Demirer and Shenoy described it as followed:

"Consider a physician who is trying to diagnose whether or not a patient is suffering from Diabetes. Diabetes has two symptoms [signs], glucose in urine, and glucose in blood. Assume we have a Bayes net model for the three variables—Diabetes (D), glucose in blood (B) and glucose in urine (U)—in which the joint distribution for the three variables P(D, B, U) factors into three conditionals, P(D), P(B | D), and P(U | D, B). Furthermore, assume that D has two states, d for Diabetes is present, and ~d for Diabetes is absent, U has two states, u for elevated glucose levels in urine, and ~u for normal glucose level in urine, and B has two states, b for elevated glucose levels in blood, and ~b for normal glucose level in blood. The physician first decides (FT) whether to order a urine test (ut) or a blood test (bt) or neither (nt). After the physician has made this decision and observed the results (if any), she next has to decide whether or not to order a second test (ST). The choices available for the second test decision depend on the decision made at FT. If FT = bt, then the choices for ST are either ut or nt. If FT = ut, then the choices for ST are either bt or nt. Finally, after the physician has observed the results of the second test (if any), she then has to decide whether to treat the patient for Diabetes or not. As described so far, one model of the problem has three chance variables, D, U, B, and three decision variables FT (first test), ST (second test), and TD (treat for Diabetes)."