Inverse decision theory with applications to screening and diagnosis of cervical intraepithelial neoplasia

Abstract

Objective: Medical decision makers would like to use decision theory to determine optimal treatment strategies for patients, but this requires priors, likelihoods, and losses. It can be very difficult to specify a loss or utility function in a medical setting, especially when considering both patient health outcomes and economic costs. These issues led to the development of Inverse Decision Theory (IDT), which involves determining the set of losses under which a given decision rule is optimal.

Methods: We apply IDT to the current standard of care for the diagnosis and treatment of precancerous lesions to the cervix, using a Bayesian approach to estimate the probabilities associated with diagnostic tests and make inferences about the region of optimality. There are two ways in which Inverse Decision Theory can be useful: (i) if the decision rule of interest is optimal, then we obtain information about the losses for the optimal treatment strategy, and (ii) if the decision rule of interest is not optimal, then we characterize the losses under which it would be optimal, and assess whether or not it contains reasonable values of the losses.

Results: This paper introduces important clinical results: in particular, we find that the current standard of care for cervical precancer is probably not optimal, and a new decision rule which requires a confirmatory biopsy for all patients with a positive Pap smear test result is better.

Conclusion: We have developed a very general and flexible approach for evaluating treatment strategies that could prove useful in a variety of medical applications.