Fig. 1. Functional structure of the generic multi-attribute heuristic decision method, as described in Manjarrés et al. (1999). In the graphical notation used, the dashed rectangles indicate that the corresponding role includes support knowledge used by the subtask.

Fig. 2. Functional structure of the multi-attribute heuristic decision method in the therapeutic domain.

Fig. 3. Inheritance and composition relations in the metaclass Model.

Fig. 4. Modelling of the State concept and of the Patient concept which inherits it.

Fig. 5. (a) Examples of specific relations (r.h.s.) and the corresponding generic relations (l.h.s.). (b) Other specific relations of the therapy decision domain.

Fig. 6. Hierarchies of objectives (for the inheritance and composition relations) in a generic decision domain.

Fig. 7. Example of an objective-composition hierarchy for a therapy decision model. As an illustration, the objective ‘Conformance to therapy’ is specified in more detail in terms of the corresponding sub-objectives, while other branches of the hierarchy are shown truncated in the interest of simplicity. The ‘negative’ objectives (those whose names are preceded by a minus sign) represent undesirable outcomes of the therapies.

Fig. 8. Basic inference structure for the solution methods of the tasks: Assess effects and cost of applicable therapies and Assess priorities of objectives.

Fig. 9. Illustration of the hierarchies of treatments used by the subtasks Assess effects and cost of applicable therapies and Assess priorities of objectives in the case of an anti-anginal therapy.

Fig. 10. Data-entry screen in a therapy decision support system. Items of a Test to evaluate Attitude towards asthma and its treatment use a 5-point scale.

Fig. 11. Alternative inference structures of the subtask Select optimum alternative. In this graphical notation, the ellipses represent elementary inferences (as is usual in the KADS and CommonKADS notations).

Fig. 12. The generic Select optimum alternative subtask decomposed in accordance with the Lexicographical ordering method. Description of the control and basic inferences.

Fig. 13. The generic Select optimum alternative subtask decomposed in accordance with the Optimise well-being function method. Control and basic inferences description.

Fig. 14. The generic Select optimum alternative subtask decomposed in accordance with the Heuristic satisfaction method. Control and basic inferences description.

Fig. 15. Examples of specific instances of well-being functions, where w_i is the priority of the objective i, At(Al)_i the attribute which measures objective i on the alternative Al and α_i, b_i parameters of the decision model.

Table 1. Examples of the association of effect quantifications (attributes, in the generic decision model) to specific objectives in the case of an anti-anginal treatment

Table 2. Examples of the association of effect quantifications to generic objectives

Table 3. Instances of the class Statistics, associated by means of the MeasuresBasedOnStatistics to the objective UserFriendlyAdministration. The columns Illness and Treatment refer to the most specific classes about which statistics are available (see treatment hierarchy example in Fig. 9). Objective: ‘user-friendly’ administration; Effect quantification: points in interval [0,10]

Table 4. Correspondence between the roles defined in the HM decision method (in italics) and some ontological categories (classes, attributes and relations—see UML descriptions above) of a therapy decision problem. The last column contains the relations constituting the static knowledge implicitly used by the respective subtasks. The attributes are denoted using the constraint language OCL, in use with UML

Table 5. Some examples of specific assumptions about the objectives and attributes in a therapy decision model

Table 6. Instances of the class Statistics, associated by means of the MeasuresBasedOnStatistics to the objective Therapy favourable to conformance. Objective: Therapy favourable to conformance; Effect quantification: Average rate of non-conformance

Table 7. Instances of the class Statistics, associated by means of the MeasuresBasedOnStatistics to the objective Low complexity. Objective: Complexity; Effect quantification: Percentage of patients who interpret the treatment correctly

Table 8. Well-being questionnaire (General well-being: 36−Depression−Anxiety+Positive well-being+Energy; Min SCORE=0; Max SCORE=6)