An internet-based method to elicit experts' beliefs for Bayesian priors: a case study in intracranial stent evaluation.

Abstract

Bayesian methods provide an interesting approach to assessing an implantable medical device (IMD) that has evolved through successive versions because they allow for explicit incorporation of prior knowledge into the analysis. However, the literature is sparse on the feasibility and reliability of elicitation in cases where expert beliefs are used to form priors. To develop an Internet-based method for eliciting experts' beliefs about the success rate of an intracranial stenting procedure and to assess their impact on the estimated benefit of the latest version. The elicitation questionnaire was administered to a group of nineteen experts. Elicited experts' beliefs were used to inform the prior distributions of a Bayesian hierarchical meta-analysis model, allowing for the estimation of the success rate of each version. Experts believed that the success rate of the latest version was slightly higher than that of the previous one (median: 80.8 percent versus 75.9 percent). When using noninformative priors in the model, the latest version was found to have a lower success rate (median: 83.1 percent versus 86.0 percent), while no difference between the two versions was detected with informative priors (median: 85.3 percent versus 85.6 percent). We proposed a practical method to elicit experts' beliefs on the success rates of successive IMD versions and to explicitly combine all available evidence in the evaluation of the latest one. Our results suggest that the experts were overoptimistic about this last version. Nevertheless, the proposed method should be simplified and assessed in larger, representative samples.