AbstractsMedical & Health Science

Using Value of Information Methods to Determine Optimal Designs for Research in Diagnostic Tests and Multi-stage Randomized Clinical Trials

by Chen Maggie

Institution: University of Toronto
Year: 2014
Keywords: Value of Information; Statistical Methods; 0308
Record ID: 2024651
Full text PDF: http://hdl.handle.net/1807/67261


Value of Information (VOI) methods are based on Bayesian decision theory and provide decision makers with a methodological framework to explicitly consider the uncertainty surrounding decision making and to evaluate the need for further research. VOI methods can be used to determine the optimal sample sizes for new research studies that maximize the expected net gain (ENG), i:e. the di erence between the expected value of information provided by the trial and the expected cost of doing it. The purpose of this thesis is to develop new methodologies in VOI research for clinical trials and diagnostic research studies. The rst project is to determine the optimal sample size for multi-stage randomized clinical trials from an industry perspective using VOI methods. A model is proposed for the expected total pro t that includes consideration of per-patient pro t, disease incidence,time horizon, trial duration, market share, and the relationship between the trial results and probability of regulatory approval. The proposed method includes multistage adaptive designs with a speci c solution for two-stage design. With an example, it is demonstrated that signi cant increases in ENG can be achieved by using multi-stage designs and a smaller expected total sample size and less cost will be required. The second project is the development of VOI methods for diagnostic research. Suppose the new research study is proposed to improve the estimation of sensitivity and speci city. The optimal sample sizes for new studies can be determined at the highest ENG considering both independent and correlated sensitivity and speci city. The methods include evaluating a single diagnostic test as well as comparing two diagnostic tests. Normal approximation methods are proposed for di erent scenario. The examples and numerical simulation demonstrate that the normal approximation methods are a good tool for sample size calculation to maximize the ENG for the new research study. VOI methods are relatively new. The methods developed in this thesis will contribute to statistical tools for decision making and help health planners to better evaluate the need for more research studies and determine the optimal sample sizes associated with them. PhD