MCDA (Deterministic)

Step 1 : Determine the context of the problem

The context of this MCDA problem is to test the feasibility of applying MCDA to evaluate the benefit-risk balance of rimonabant versus placebo.

We present here the benefit-risk balance from the perspective of the prescribing physicians and the regulators. A brief introduction to the decision problem is available here. The full results of the MCDA are available in the Rimonabant Wave 1 Case Study.

Step 2:Determine the objectives of assessment

We used the European Public Assessment Report (EPAR) for rimonabant as the main source for benefit (favourable effects) and risk (unfavourable effects) criteria. The primary benefit of rimonabant was the effect on weight loss and maintenance of weight loss at 12 months. The main adverse event we were concerned about was psychiatric disorders. Other benefit and risk criteria that we took into account in the model are also shown in the value tree in Figure 1 below.

Figure 1 Value tree for MCDA of benefit-risk balance of rimonabant versus placebo (created in HiView 3)

Step 3: Determine the alternatives

The alternatives (options) in this model were rimonabant 20mg and placebo.

Step 4: Effects table

The range of preference score and type of criterion value function would greatly affect the utility score, which may have substantial impact on the final results. As an illustration, we decided that all preference scores for all criteria were based on a fixed scale and on a linear value function. This means that the range of the preference scale was anchored to the "fixed upper" and "fixed lower" values in Table 1(benefit criteria) and Table 2(risk criteria) below. We selected these values arbitrarily in line with the available data and the opinion of the study team's physician. The data were then scored using the specified value functions, to be transformed onto the utility scale.

Table 1 Effects table for favourable effects (benefit criteria)

Table 2 Effects table for unfavourable effects (risk criteria)

Step 5: Weighting

We examined the sensitivity of the model to weights from different stakeholders: physicians, regulators and statisticians (all within the case study team). Medical and regulator members of the case study provided medical/regulatory opinion, whereas, statisticians acted as layman. This was done in an online survey based on the values of "swing" from fixed lower to fixed upper values in Table 1 and Table 2above. The responders were asked to score importance of individual criteria between 0 (not important) and 10 (very important).Table 3 below shows the results of the medical/regulatory opinions from the survey. We summarised the results into average score by criterion, and then wecalculated the weights on each criterion proportional to these average scores, based on a total weight of 100. The weights were then combined with the scores in Step 4 to produce weighted scores.

The results from the layman (statisticians) survey and the comparisons between the two stakeholders are available in the Rimonabant Wave 1 Case Study report.

Table 3 Weighting survey results from physicians and regulators within the case study team

Step 6: Overall value

Figure 2 Overall results

Figure 3 Overall utility scores

Figure 4 A "difference display" of the weighted difference in utility scores between rimonabant and placebo sorted from the most important benefit criterion (top) to the most important risk criterion (bottom)

Step 7: Sensitivity testing

We performed "sensitivity up" analyses on each level of criteria to investigate the amount of changes required in the total weights in the criteria at that level for the benefit-risk balance to reverse to the alternative. Here we show the results of the sensitivity analyses at the highest level i.e. "benefits" and "risks".

In this case, placebo, which was the preferred option, is sensitive to weights assigned to these two criteria. Current weights for benefit and risk were 42.5 and 57.5 (as indicated by the vertical red lines in Figure 5). A small increase in the total weight assigned on benefits of approximately 1.5 to 44 (Panel A), would result in rimonabant becoming the more preferred alternative. Similarly, a small decrease in the total weight assigned on the risks of approximately 1.5 to 56 (Panel B), would also result in rimonabant becoming the more preferred alternative.

Sensitivity analyses on other criteria are available in the Rimonabant Wave 1 Case Study report.

(A)

(B)