A Bayesian framework to handle missing data in cost-effectiveness analysis (CEA) alongside with randomised longitudinal trials

点击翻译按钮获取中文摘要

My PhD aims to develop a semi-parametric Bayesian framework to handle missing data in economic evaluations alongside longitudinal studies and randomised trials. Innovation and research in health technology have achieved remarkable progress over the past decades. While the technological breakthroughs in health care have largely improved patients' clinical outcomes and quality of life, they pose additional economic burden to countries with limited budgets. Thus it has generated wide concerns about how to scientifically assess the value of health technologies and choose between them based on their broader impact on users, payers, policy decision-makers and even the whole society. Economic evaluation, in particular, cost-effectiveness analysis (CEA), has become one of the major tools in health technology assessment. The method usually incorporates all available and appropriate evidence collected from trials through a follow-up period, and formerly compares the costs and health benefits of existing and new medical technologies and public health programs to inform the decision about the coverage or reimbursement of public health services (Briggs et al., 2006). The longitudinal nature of data often encountered in CEA makes it prone to the threat of non-responses which may lead to biased results and the waste of limited resources if missing data could not be appropriately handled.Although standard statistical tools such as multiple imputation (MI) have been well developed to deal with missingness and can produce valid estimates if they can be performed properly, their extensions in CEA have been hindered due to the increasing complexity in the modelling. Examples of this complexity include the necessity of considering a bivariate outcome comprising costs and effectiveness data, as well as specific features such as e.g., skewness, spikes and correlation between observations. More importantly, the underlying decision uncertainty brought by plausible missingness assumptions also needs careful consideration (Gabrio et al., 2018b). These features make Bayesian modelling a preferable alternative in this context. Moreover, Bayesian modelling can bring external evidence such as expert opinions to decision making through prior distributions and this benefit is crucial in economic evaluation alongside longitudinal studies as missing values are more likely to be non-ignorable. Recent research has explored the substantial advantages to perform economic evaluations under a parametric Bayesian framework - the capacity and flexibility to allow the complexity above, assess the robustness of missingness assumptions and avoid potential risks of traditional methods (Gabrio et al., 2018a, Gabrio et al., 2018b). However, models with more relaxing assumptions can be exploited to handle more realistic situations. My PhD project will further this field by developing a semi-parametric model for missingness in economic evaluations with Bayesian nonparametric modelling and formally comparing it with existing models to see if our model has the potential to work as the standard tool in health technology assessment.BRIGGS, A., SCULPHER, M. & CLAXTON, K. 2006. Decision modelling for health economic evaluation, OUP Oxford.GABRIO, A., DANIELS, M. J. & BAIO, G. 2018a. A Bayesian Parametric Approach to Handle Missing Longitudinal Outcome Data in Trial-Based Health Economic Evaluations. arXiv preprint arXiv:1805.07147.GABRIO, A., MASON, A. J. & BAIO, G. 2018b. A full Bayesian model to handle structural ones and missingness in economic evaluations from individual-level data. Statistics in Medicine.