How Bayesian decision analysis helps vets choose treatments

Abstract

Simple Summary Decision making in veterinary medicine can be extremely difficult. Often, different choices can have vastly different costs, complications, and outcomes associated with them. Bayesian inference and decision analysis are two tools that, when combined, can help clinicians and pet owners decide on the preferred course of action. In this retrospective case study, we describe a lethargic ferret that is no longer eating. We solicited opinions from three expert veterinarians who were not involved with the case on what the diagnosis could be before and after a series of diagnostic tests. We also asked the original clinical team to estimate how valuable different clinical outcomes were. By combining these data, we were able to assess if the original clinical team was right to take the animal to surgery. We also discuss some of the pitfalls of not using Bayesian inference in diagnosis, some cognitive biases that may have played a role in the case management decisions, and the wider usefulness of decision-analysis methods to help foster shared decision making between client and veterinarian. Abstract Bayesian inference and decision analysis can be used to identify the most probable differential diagnosis and use those probabilities to identify the best choice of diagnostic or treatment among several alternatives. In this retrospective case analysis, we surveyed three experts on the prior probability of several differential diagnoses, given the signalment and history of a ferret presenting for lethargy and anorexia, and the conditional probability of different clinical findings (physical, bloodwork, imaging, etc.), given a diagnosis. Using these data and utility estimates provided by other clinicians, we constructed a decision tree to retrospectively identify the optimal treatment choice between exploratory laparotomy and medical management. We identified medical management as the optimal choice, in contrast to the original clinical team which performed an exploratory laparotomy. We discuss the potential cognitive biases of the original clinical team. We also discuss the strengths, e.g., shared decision making, and limitations of a Bayesian decision analysis in the veterinary clinic. Bayesian decision analysis can be a useful tool for retrospective case analysis and prospective decision making, especially for deciding on invasive interventions or end-of-life care. The dissimilarity of expert-derived probability estimates makes Bayesian decision analysis somewhat challenging to apply, particularly in wide-ranging specialties like zoological medicine.