Cost Effectiveness Analysis in Healthcare Decision-Making: Stochastic Modeling and Statistical Inference.

The field of cost-effectiveness analysis (CEA) deals with the comparison of health interventions based on both costs and effectiveness (ability to improve health). This dissertation makes several methodological contributions to this area. Part I develops direct methods for computing mean costs and effectiveness, and hence conducting CEA, in multi-state disease processes. The common approach in this case is to use discrete-event simulation techniques to simulate the process to get the associated cost and effectiveness outcomes. However, the setting up and implementation of simulation studies can be time and resource intensive. The dissertation develops analytical expressions for the time-to-failure, reward processes, and their (discounted) expectations for time-homogeneous semi-Markov processes with progressive structure. Direct Monte Carlo methods are proposed for time-varying multi-state processes. The advantages of these direct methods over discrete-event simulation are discussed. Sensitivity analysis to parameter estimation is also considered. The results are demonstrated on illustrative applications. Part II deals proposes a richer analysis of cost-effectiveness data from discrete event simulation of disease processes. Such simulations generate extensive amounts of data which are rarely examined in detail. The analysis is typically reduced to computing and comparing simple CEA metrics. This part of the dissertation proposes a comprehensive exploratory analysis of the data through graphical techniques. This includes examining both cross-sectional and temporal views of the time-to-failure, cost, and effectiveness distributions. The concept of a treatment-effect function is discussed, and it leads to generalized versions of two common CEA metrics. The potential for richer analysis is illustrated through various examples. Part III of the dissertation reviews the common CEA metrics based on means of the cost and effectiveness outcomes and discusses comparisons based on first and second-order stochastic dominance as well as utility functions. It also deals with methods for incorporating statistical uncertainty from estimating the unknown parameters in CEA. Large-sample normal approximations and resampling methods are reviewed. New contributions to the CEA literature include stochastic dominance comparisons in the presence of estimation uncertainty, use of rank methods, and analysis with censored data.PhDIndustrial & Operations EngineeringUniversity of Michigan, Horace H. Rackham School of Graduate Studieshttp://deepblue.lib.umich.edu/bitstream/2027.42/89730/1/mcdefauw_1.pd