OBJECTIVES: When empirical evidence is not available, decision models often require expert estimates of input parameters. Beta distributions commonly represent uncertainty for proportions of interest. Published methods for eliciting proportions ask experts to characterise the density of the uncertainty distribution in various ways. Estimating treatment costs for melanoma requires an understanding of the proportions of people receiving each intervention. We aimed to elicit proportions receiving each intervention in the care pathway for melanoma.

METHODS: We mapped pathways for the dermatological, surgical and oncological management of melanoma. We designed an expert elicitation exercise to estimate proportions receiving each intervention. Consultants from each speciality piloted the relevant pathway and exercise. We identified a sample of experts from each specialty, considering geographical spread across the UK National Health Service. Experts were clinicians managing melanoma (surgeons, dermatologists and oncologists). Participating experts completed the final exercises. We modelled experts’ beliefs using a scaled beta distributions. We used random-effect meta-analysis to combine estimates in the base case, and linear pooling in a sensitivity analysis.

RESULTS: The experts piloting the elicitation exercise found quartiles for a mean proportion difficult to estimate. We redesigned the exercise to ask for a best estimate, lower and upper plausible limits and a qualitative level of uncertainty for each proportion. Seven dermatologists, five oncologists and four surgeons completed the new exercise (August 2022-February 2023). Experts gave heterogeneous estimates of the proportions of people receiving many of the interventions.

CONCLUSIONS: Our estimates of the proportions of people with melanoma receiving each intervention, and of uncertainty in these proportions, will help to inform costing for decision-analytic models. Although we piloted the exercises, heterogeneity suggests that experts interpreted questions or pathways differently. Using a best estimate and a slider to indicate uncertainty may be a more direct and intuitive way to elicit a beta distribution for a proportion.