A Bayesian anaylsis of data measures the effect of the use of bone marrow cell therapy on left ventricular function
Different from frequentist analysis used in traditional clinical studies which draw confirmed conclusions on a single trial, the Bayesian approach incorporates the results from earlier, related trials as well as taking results from the current trial to update the prior information. Since both prior information and conducted data play a role in the final decision making process, the Bayesian approach delivers the more accurate results conditionally on the observed data, instead of imagined random data and fixed parameters. In this thesis we assessed the role of prior distribution and loss function choice on the Bayes estimate of effect size of data from a randomized controlled clinical trial, Timing in Myocardial Infarction Evaluation (TIME), and make inferences about the effect of bone marrow mononuclear cell delivery on left ventricular function after ST-segment elevation myocardial infarction (STEMI). (Traverse et al., 2012) We composed an informative prior based on the information from formerly conducted clinical trials BOOST and REPAIR-AMI, as well as a non-informative prior and a counterintuitive prior. Three different prior distributions were composed by assigning different weights on these prior components. Updated with the results from the clinical study TIME, the approximate posterior distribution were derived using Bayesian theorem. Moreover, we used a weighted loss function to interpret the derived posterior distribution under more realistic public health scenarios. Specifically, in order to incorporate the prices and penalties into the inference and decision-making process, we built a continuous loss function to evaluate the action of "estimate &thetas; by posterior function". Implementing risk benefit balances that are useful in phase II clinical trials, we weighted findings reflecting the importance of minimizing errors due to overestimations than underestimations. Our results showed that Bayesian estimate is sensitive to the weights allocated on different priors, since the result from BOOST study has bigger mean and smaller variance than REPAIR-AMI. But it was mainly determined by the choice of loss function. Because in linear weight loss function, Bayesian estimate is defined as the percentile of posterior distribution. Thus defining an appropriate loss function for the posterior distribution is dominant in finding Bayesian estimates.
Liu, Xiaopeng, "A Bayesian anaylsis of data measures the effect of the use of bone marrow cell therapy on left ventricular function" (2015). Texas Medical Center Dissertations (via ProQuest). AAI1597539.