Use of Bayesian methods to inform health technology assessment decision-making
Lead Research Organisation:
University of Leicester
Department Name: Health Sciences
Abstract
Resources dedicated to health care are limited and under increasing demand from consumers. It is therefore important that the resources available are used in order to produce the greatest benefit for the population. This has lead to the creation of structures in the UK (NICE) and elsewhere to facilitate decision-making regarding the health technologies that are recommended for routine practice. This is achieved through the use of formal statistical modelling of the available evidence on the (potentially competing) technologies of interest. Often, evidence will be limited on the effectiveness and ?value-for-money? of a technology but (difficult) decisions need to be made despite the uncertainty.
This project will explore the potential added value of using an alternative approach (the standard approach is commonly referred to as classical) to health technology assessment (HTA) statistical modelling known as the Bayesian approach. Such an approach provides a framework allowing additional (external) information (such as expert opinion), to be formally and transparently incorporated into the analysis thus increasing the information considered. The Bayesian approach also allows for more realistically complex modelling of the complexities required by HTA (than classical approaches) through computer intensive simulation methods,
The proposed project has three phases. Phase one considers the re-analyses of seven diverse HTAs (including drug treatments, diagnosis and waiting list prioritisation) using Bayesian methods. Advantages of the Bayesian models used include improved modelling of uncertainty and a framework which facilitates the updating of the HTA when new data becomes available. The approach also allows for an assessment of what research studies are required in the future. The results and conclusions from these Bayesian re-analyses will be carefully compared with the original analyses, and the differences in the assumptions made carefully assessed.
Phase two will explore how the analyses in phase one can be presented to decision-makers, and whether, and if so how, decision-makers might adopt Bayesian methods more formally in the actual process of decision-making. This will be achieved by presenting decision makers with both classical and Bayesian analyses in order to assess their understanding of the results obtained. Following this, we will assess the feasibility of using Bayesian methods to incorporate the committee?s judgements about evidence, uncertainty and the assumptions made in HTAs as they make a decision.
Phase three will develop guidelines for the use of Bayesian methods in HTA drawing on the outputs from the previous phases of the project.
This project will explore the potential added value of using an alternative approach (the standard approach is commonly referred to as classical) to health technology assessment (HTA) statistical modelling known as the Bayesian approach. Such an approach provides a framework allowing additional (external) information (such as expert opinion), to be formally and transparently incorporated into the analysis thus increasing the information considered. The Bayesian approach also allows for more realistically complex modelling of the complexities required by HTA (than classical approaches) through computer intensive simulation methods,
The proposed project has three phases. Phase one considers the re-analyses of seven diverse HTAs (including drug treatments, diagnosis and waiting list prioritisation) using Bayesian methods. Advantages of the Bayesian models used include improved modelling of uncertainty and a framework which facilitates the updating of the HTA when new data becomes available. The approach also allows for an assessment of what research studies are required in the future. The results and conclusions from these Bayesian re-analyses will be carefully compared with the original analyses, and the differences in the assumptions made carefully assessed.
Phase two will explore how the analyses in phase one can be presented to decision-makers, and whether, and if so how, decision-makers might adopt Bayesian methods more formally in the actual process of decision-making. This will be achieved by presenting decision makers with both classical and Bayesian analyses in order to assess their understanding of the results obtained. Following this, we will assess the feasibility of using Bayesian methods to incorporate the committee?s judgements about evidence, uncertainty and the assumptions made in HTAs as they make a decision.
Phase three will develop guidelines for the use of Bayesian methods in HTA drawing on the outputs from the previous phases of the project.
Technical Summary
The creation of structures in the UK (NICE) and elsewhere to facilitate evidence-based health policy decision-making has highlighted the role that both systematic reviews and economic evaluations have to play in the health technology assessment (HTA) decision-making process. HTA is concerned with decision-making under uncertainty, and involves the synthesis of evidence from diverse sources. A Bayesian approach to HTA has huge, and multifaceted, potential benefits. It allows the incorporation of external information through the use of prior distributions, and provides the flexibility in statistical modelling required to represent the inherent complexities of HTA.
The proposed project has three phases. Phase one considers the re-analyses of seven diverse HTAs (drug treatments, diagnosis, orphan drugs and waiting list prioritisation) using Bayesian methods. All re-analyses will use a one-stage comprehensive approach in which estimation of all parameters (estimated through realistically complex evidence synthesis models) and (any) corresponding decision model are evaluated simultaneously. Advantages of this approach include the appropriate propagation of all parameter uncertainty through to model outputs and the facilitation of updating. Issues given specific attention include: i) Methods to address bias and uncertainty; ii) Incorporation of external information; iii) Advanced modelling issues including mixed treatment comparisons, extrapolation and sequential diagnostic/treatment options; and iv) Model based assessments of future research needs. While adequately accommodating each of these specific issues individually is complex, in most HTAs, many will be relevant simultaneously. It is the simultaneous management of these complexities that provides the greatest challenge, but also greatest potential gains. The results and conclusions from these Bayesian re-analyses will be carefully compared with the original analyses, and the differences in the assumptions made carefully assessed.
Phase two will explore how the analyses in phase one can be presented to decision makers, and whether, and if so how, decision-makers might adopt Bayesian methods more formally in the actual process of decision making. This will be achieved by presenting members of the NICE Appraisals Committee with both classical and Bayesian analyses in order to elicit and assess their understanding of the results obtained. Following this, we will use Bayesian methods to assess the feasibility of formally (and transparently) incorporating the committee?s judgements about evidence, uncertainty and the assumptions made in HTAs at the appraisal stage.
Phase three will develop guidelines for the use of Bayesian methods in HTA drawing on the outputs from the previous phases of the project.
The proposed project has three phases. Phase one considers the re-analyses of seven diverse HTAs (drug treatments, diagnosis, orphan drugs and waiting list prioritisation) using Bayesian methods. All re-analyses will use a one-stage comprehensive approach in which estimation of all parameters (estimated through realistically complex evidence synthesis models) and (any) corresponding decision model are evaluated simultaneously. Advantages of this approach include the appropriate propagation of all parameter uncertainty through to model outputs and the facilitation of updating. Issues given specific attention include: i) Methods to address bias and uncertainty; ii) Incorporation of external information; iii) Advanced modelling issues including mixed treatment comparisons, extrapolation and sequential diagnostic/treatment options; and iv) Model based assessments of future research needs. While adequately accommodating each of these specific issues individually is complex, in most HTAs, many will be relevant simultaneously. It is the simultaneous management of these complexities that provides the greatest challenge, but also greatest potential gains. The results and conclusions from these Bayesian re-analyses will be carefully compared with the original analyses, and the differences in the assumptions made carefully assessed.
Phase two will explore how the analyses in phase one can be presented to decision makers, and whether, and if so how, decision-makers might adopt Bayesian methods more formally in the actual process of decision making. This will be achieved by presenting members of the NICE Appraisals Committee with both classical and Bayesian analyses in order to elicit and assess their understanding of the results obtained. Following this, we will use Bayesian methods to assess the feasibility of formally (and transparently) incorporating the committee?s judgements about evidence, uncertainty and the assumptions made in HTAs at the appraisal stage.
Phase three will develop guidelines for the use of Bayesian methods in HTA drawing on the outputs from the previous phases of the project.
