Incorporating dependence in risk assessments of future antibiotic resistance: Eliciting and modelling probabilistic dependence

Healthcare systems rely on having a range of effective antibiotics available to treat patients with resistant infections. The threat of multidrug resistance means that assuming rates of resistance to different antibiotics are independent may severely underestimate risks to the current portfolio of drugs. We illustrate the problem and model dependence between certain bug-drug combinations. We present a method for eliciting dependence from experts in a formal and structured manner that promotes transparency and reproducibility of the results and mitigates against common cognitive fallacies.