Emulators for complex decision models in healthcare: feasibility of calibration, explication of analyses, and development of intitution with an application in prostate cancer

Brown University, Providence RI

Investigators

Abstract

? DESCRIPTION (provided by applicant): Patients, physicians, and policy makers face complex decisions about health. These decisions are complex because of the existence of numerous alternatives, aggregation of benefits and harms, differing patient preference, and uncertainty in the evidence base. Formal approaches, such as decision analytic models, facilitate the decision-making process in the presence of these complexities. However, these models are not widely used in practice as they are difficult to develop, computationally expensive, and opaque. An emulator, or metamodel, is a statistical approximation of a model and has been used in a range of disciplines to address complex problems. Although it has not been widely used in health, it may be a useful tool to facilitate complex decisions about health. The overall project objective is to develop a framework for facilitating the development, analysis and dissemination of computationally expensive decision models by means of emulators, and to apply the framework in a prostate cancer screening example. Specifically, the project aims to: (1) develop a framework for using emulators to facilitate the development and analysis of detailed models; (2) provide a proof of concept application of the framework using a comprehensive model for prostate cancer screening; (3) explore whether the use of emulators can facilitate tailoring models to new contexts, and building intuition about their behavior. The proposed work is significant because it may help mitigate the challenges (particularly relating to the computational burden) of implementing and disseminating decision models in practice. We will make an easy- to-use and easy-to-tailor implementation of a well-developed and validated prostate cancer screening model. We anticipate the results of this work may encourage and enable future research and applications in this field.