PRODUCING QUALITY-ADJUSTED OPINIONS FOR PREFERENCE-SENSITIVE MULTI-CRITERIAL DECISION SUPPORT

Purpose:
   To show how the option performance ratings required in multi-criterial decision support can be adjusted for evidential quality/grade to produce a quality-adjusted opinion that is sensitive to the person's preferences.

Method(s):
   In person-centred care the preferences of the individual over the multiple considerations important to them are explicitly elicited at or near the point of decision and integrated with the Best Estimates Available Now (BEANs) of the performance rates of all the options on all those criteria. When carried out within a Multi-Criteria Decision Analytic framework the result is an opinion made up of the set of option expected values. The BEANs will vary in quality, according to their provenance, and adjustment for this produces a Quality-adjusted opinion.

   We do not prescribe any particular way of arriving at the quality-adjustment rating, itself a preference-sensitive judgement. It must however be on the continuous 0 to 1 ratio scale. We find the following adaption of the GRADE levels helpful. It assesses quality in terms of confidence that the rating is close to the 'true' one and hence whether further research is expected to change it significantly. 1.0 - 0.8 Very confident;  0.7- 0.5 Moderately confident; 0.4 - 0.2 Limited Confidence; 0.1- 0 Very little confidence. (There is no suggestion that GRADE would support this numerical mapping.) A simple interactive example is presented as proof and illustration of method, followed by two empirical cases, drawing on publicly-available demonstration decision aids. 

Result(s):
   The first takes the demonstration MagicApp aspirin decision aid, translates it into MCDA format and uses the provided GRADE assessments as the quality adjustments. (https://www.magicapp.org/app#/guideline/387). The second uses the Norwegian Bipolar Disorder example from mybetterdecisions.org. This aid is already in MCDA format and we use the ratings from the 'Reliable Evidence' criterion to adjust the ratings for the other criteria. In both cases the examples confirm that the effect of quality-adjustment is influenced by the preference weights, as it should be.

Conclusion(s):
   An evidence grading classification can contribute to preference-sensitive decision support by generating a quality-adjusted opinion. If this differs from the original opinion, dissonance resolution should recognise the many sources of a low option grading. These include recent development, which means the option has not been studied as much, or as thoroughly, as older interventions.