DEVELOPING APPROPRIATE ANALYTICAL METHODS FOR COST-EFFECTIVENESS ANALYSES THAT USE CLUSTER RANDOMISED TRIALS
Lead Research Organisation:
London School of Hygiene & Tropical Medicine
Department Name: Public Health and Policy
Abstract
In all societies, people recognise the importance of improving the population?s health, but there are limits on how much we are prepared to spend on health care. Organisations such as NICE, help decide which health care options (e.g. new drugs or doctors) represent the best value for money. NICE make their recommendations, based on studies that consider the relative costs of different options, but also how much they improve people?s health. However, there are concerns about the scientific quality of these studies.
One way of collecting the evidence is through a study called a ?cluster randomised trial?. This study is specially designed to try and give accurate results. However, for the study to provide accurate results it is also essential that it uses appropriate methods of analysis. Studies that have used these ?cluster randomised trials? to assess costs and outcomes have not used appropriate methods to report their results. A key feature of this study is that which drug or service a patient receives depends on which cluster (e.g. a hospital) they attend. However, these studies have not recognised that patients who are in particular cluster (e.g. a hospital) are likely to be more similar (for example in their social background) to each other, than to patients attending other hospitals. A big concern is that by using these methods which are inappropriate, the studies are providing inaccurate results.
This research will develop careful scientific methods for studies that compare costs and health outcomes using these ?cluster randomised trials?. It will find out which of these new methods is the best by carefully considering them in different situations. We will take studies that have already provided results, and redo them using these new methods. We will test the methods across different medical areas (e.g. maternity services, heart disease, sleeping disorders). The research will provide recommendations on which method is best under different circumstances.
The study will provide instructions and training opportunities to teach future researchers how to use these new methods. The investigation will produce scientific publications for international journals. We will also provide information on the methods for policy-makers, and members of the public via our project website. By providing better methods for deciding which drugs and health services should be provided, this research will help to ensure that scarce health care resources are used in the best ways for improving the population s health.
One way of collecting the evidence is through a study called a ?cluster randomised trial?. This study is specially designed to try and give accurate results. However, for the study to provide accurate results it is also essential that it uses appropriate methods of analysis. Studies that have used these ?cluster randomised trials? to assess costs and outcomes have not used appropriate methods to report their results. A key feature of this study is that which drug or service a patient receives depends on which cluster (e.g. a hospital) they attend. However, these studies have not recognised that patients who are in particular cluster (e.g. a hospital) are likely to be more similar (for example in their social background) to each other, than to patients attending other hospitals. A big concern is that by using these methods which are inappropriate, the studies are providing inaccurate results.
This research will develop careful scientific methods for studies that compare costs and health outcomes using these ?cluster randomised trials?. It will find out which of these new methods is the best by carefully considering them in different situations. We will take studies that have already provided results, and redo them using these new methods. We will test the methods across different medical areas (e.g. maternity services, heart disease, sleeping disorders). The research will provide recommendations on which method is best under different circumstances.
The study will provide instructions and training opportunities to teach future researchers how to use these new methods. The investigation will produce scientific publications for international journals. We will also provide information on the methods for policy-makers, and members of the public via our project website. By providing better methods for deciding which drugs and health services should be provided, this research will help to ensure that scarce health care resources are used in the best ways for improving the population s health.
Technical Summary
Health policy-makers use cost-effectiveness analyses (CEA) to decide which interventions should be provided. For CEA to deliver accurate results, the use of appropriate statistical methods is essential. CEA may be undertaken alongside cluster randomized trials (CRTs) where randomization is at the level of the cluster (for example the hospital) rather than the individual. Analytical methods are required which recognise that costs and outcomes within clusters may be correlated. However, most CEA based on CRTs use regression models that assume observations are independent, which may lead to incorrect inferences. This research aims to provide improved methods for CEA that use data from CRTs and guidance on their use.
The objectives are:
(1) To assess, using simulations, the relative performance (bias, precision, mean squared error, coverage) of the following analytical methods: bivariate Bayesian Hierarchical models (BHMs); bivariate hierarchical models (maximum likelihood); bivariate generalised estimating equations (GEEs); and the ?2 stage? non-parametric bootstrap under different scenarios for CEA that use CRTs.
(2) To compare the cost-effectiveness results from applying the alternative methods across seven different case studies and use these findings to develop a general analytical strategy for CEA that use CRTs.
(3) To evaluate the impact of the choice of method on estimates of lifetime cost-effectiveness using decision-analytical models to extrapolate from the CRT results.
(4) To disseminate methods and results for researchers and policy-makers.
The simulations will compare the relative performance of the methods across different scenarios (skewed costs, few clusters, missing data) faced by CEA. We will identify those scenarios where simpler methods will suffice versus those where more complex BHMs are required. The research will reanalyse seven CEAs based on CRTs to see whether inferences change in practice. We will use insights from the simulations and the case studies to develop a statistical analysis strategy.
The research will examine the impact of the choice of statistical method for analysing CRT data on estimates of lifetime cost-effectiveness. It will use three case studies with pre-existing decision models. The analyses will construct overall measures of statistical uncertainty that incorporate uncertainty about the specification of the statistical model with parameter uncertainty.
This research will identify which methods are most appropriate for CEA based on CRTs, and will disseminate these methods widely. We will run conceptual and practical workshops for policy-makers and researchers to make the best methods easily accessible, so that future studies provide a firmer basis for allocating scarce resources.
The objectives are:
(1) To assess, using simulations, the relative performance (bias, precision, mean squared error, coverage) of the following analytical methods: bivariate Bayesian Hierarchical models (BHMs); bivariate hierarchical models (maximum likelihood); bivariate generalised estimating equations (GEEs); and the ?2 stage? non-parametric bootstrap under different scenarios for CEA that use CRTs.
(2) To compare the cost-effectiveness results from applying the alternative methods across seven different case studies and use these findings to develop a general analytical strategy for CEA that use CRTs.
(3) To evaluate the impact of the choice of method on estimates of lifetime cost-effectiveness using decision-analytical models to extrapolate from the CRT results.
(4) To disseminate methods and results for researchers and policy-makers.
The simulations will compare the relative performance of the methods across different scenarios (skewed costs, few clusters, missing data) faced by CEA. We will identify those scenarios where simpler methods will suffice versus those where more complex BHMs are required. The research will reanalyse seven CEAs based on CRTs to see whether inferences change in practice. We will use insights from the simulations and the case studies to develop a statistical analysis strategy.
The research will examine the impact of the choice of statistical method for analysing CRT data on estimates of lifetime cost-effectiveness. It will use three case studies with pre-existing decision models. The analyses will construct overall measures of statistical uncertainty that incorporate uncertainty about the specification of the statistical model with parameter uncertainty.
This research will identify which methods are most appropriate for CEA based on CRTs, and will disseminate these methods widely. We will run conceptual and practical workshops for policy-makers and researchers to make the best methods easily accessible, so that future studies provide a firmer basis for allocating scarce resources.
