Widening the spectrum of health outcomes used in health technology assessment: integrated synthesis and mapping to QALYs
Lead Research Organisation:
University of Bristol
Department Name: Community-Based Medicine
Abstract
NICE has to decide whether the NHS should adopt new medicines. To do this the health gains brought about by treatment, in conditions as diverse as heart disease, cancer, eczema, sleep disorders, deafness, etc., must all be translated into units on a common scale, so that NHS resources are equitably distributed across all conditions. This common scale is measured in units of Quality Adjusted Life Years.
This approach has always had critics because it is difficult for anyone to accept that all the dimensions of human experience can be captured on a single scale. However, while everyone recognises the imperfections of this system, it is difficult to see how fair decisions can be made without a measure of this sort.
But there are difficulties due to the narrow way in which health gains due to treatment are currently assessed. Typically, the effects of treatments are measured in randomised trials, often using patient- or clinician-based assessments. A single measure used in the trials is selected, often just because it is the most commonly reported, and the effect of treatment on this measure is converted into QALYs using data from other studies. Use of a single measure when several are recorded is inefficient and wasteful; the uncertainty in assessments would be reduced if more of the data was used. But the most serious problem is that many diseases are multi-faceted, and lead to a deterioration in life quality in several different ways, for example, a single illness might cause pain, lower cognitive skills, limit mobility, change mood, and so on. If only a single trial outcome is used, which fails to reflect the full range of benefits of treatment, then the QALY gains due to treatment may be seriously under-estimated. Treatments may then appear to be less effective than they really are.
The purpose of this research is to develop new ways of assessing treatment effects, that can combine information on a wider range of trial outcomes, and to devise a method for converting them to QALYs that allows them all to appropriately contribute to the assessment of QALY gains.
These new methods will address a number of problems with current approaches, and will produce methods for health technology evaluations and clinical guidelines development that are more transparent, make better use of available data, and are better validated as a result of checking the consistency of different types of evidence.
This approach has always had critics because it is difficult for anyone to accept that all the dimensions of human experience can be captured on a single scale. However, while everyone recognises the imperfections of this system, it is difficult to see how fair decisions can be made without a measure of this sort.
But there are difficulties due to the narrow way in which health gains due to treatment are currently assessed. Typically, the effects of treatments are measured in randomised trials, often using patient- or clinician-based assessments. A single measure used in the trials is selected, often just because it is the most commonly reported, and the effect of treatment on this measure is converted into QALYs using data from other studies. Use of a single measure when several are recorded is inefficient and wasteful; the uncertainty in assessments would be reduced if more of the data was used. But the most serious problem is that many diseases are multi-faceted, and lead to a deterioration in life quality in several different ways, for example, a single illness might cause pain, lower cognitive skills, limit mobility, change mood, and so on. If only a single trial outcome is used, which fails to reflect the full range of benefits of treatment, then the QALY gains due to treatment may be seriously under-estimated. Treatments may then appear to be less effective than they really are.
The purpose of this research is to develop new ways of assessing treatment effects, that can combine information on a wider range of trial outcomes, and to devise a method for converting them to QALYs that allows them all to appropriately contribute to the assessment of QALY gains.
These new methods will address a number of problems with current approaches, and will produce methods for health technology evaluations and clinical guidelines development that are more transparent, make better use of available data, and are better validated as a result of checking the consistency of different types of evidence.
Technical Summary
Cost-effectiveness analyses used in NICE appraisals and Clinical Guidelines require that treatment differences on outcomes measured in clinical trials are mapped into a common utility scale. The mapping is based on a regression calibration informed by data from a cohort study or trial. This form of analysis is a particularly common feature in neurology, psychiatry, dermatology, and rheumatology, where treatments are evaluated using physician- or patient-reported assessments.
Current methods are based on mapping from a single outcome, usually the one most often reported. This makes inefficient use of data, focuses attention on inappropriate outcomes, and may result in some trials being excluded. Failure to take account of measurement error, a common feature of these outcome scales, leads to under-estimation of QALY gains due to treatment. More seriously, use of a univariate mapping relation may systematically under-estimate the QALY gains due to treatment, and produce biased estimates of relative treatment differences, especially in diseases like Alzheimers where treatments may result in imperfectly correlated changes on distinct dimensions of life quality, that cannot be reflected on any single measure.
We propose an integrated approach that synthesises the multiple outcomes reported in trials and simultaneously effects a multi-variate regression mapping onto a QALY scale. It will be possible to combine outcome information even when different trials report different combinations of outcomes. Our approach will also use methods able to combine different regressions in order to synthesise all the information that could inform the multi-variate mapping. We will use published material on test-retest, inter- and intra-rater reliability to provide additional information on measurement error and correlation parameters.
This is a novel treatment of multi-outcome synthesis that takes better advantage of multiple, correlated outcomes than current approaches, by employing the between-measure relationships that are already assumed in mapping. Our method will also allow multiple sources of data to be used to directly inform the mapping, which will enable a systematic use of available evidence, rather than the somewhat arbitrary selection of a single cohort study that characterises current practice.
We propose to apply these methods to Alzheimer?s disease, depression, psoriasis and psoriatic arthritis. In each case we will contrast the treatment effects mapped onto the QALY scale produced by the proposed multi-outcome approach to those produced by univariate synthesis. We will also show how the method can be adapted to the Markov state structures commonly used in cost-effectiveness models.
Current methods are based on mapping from a single outcome, usually the one most often reported. This makes inefficient use of data, focuses attention on inappropriate outcomes, and may result in some trials being excluded. Failure to take account of measurement error, a common feature of these outcome scales, leads to under-estimation of QALY gains due to treatment. More seriously, use of a univariate mapping relation may systematically under-estimate the QALY gains due to treatment, and produce biased estimates of relative treatment differences, especially in diseases like Alzheimers where treatments may result in imperfectly correlated changes on distinct dimensions of life quality, that cannot be reflected on any single measure.
We propose an integrated approach that synthesises the multiple outcomes reported in trials and simultaneously effects a multi-variate regression mapping onto a QALY scale. It will be possible to combine outcome information even when different trials report different combinations of outcomes. Our approach will also use methods able to combine different regressions in order to synthesise all the information that could inform the multi-variate mapping. We will use published material on test-retest, inter- and intra-rater reliability to provide additional information on measurement error and correlation parameters.
This is a novel treatment of multi-outcome synthesis that takes better advantage of multiple, correlated outcomes than current approaches, by employing the between-measure relationships that are already assumed in mapping. Our method will also allow multiple sources of data to be used to directly inform the mapping, which will enable a systematic use of available evidence, rather than the somewhat arbitrary selection of a single cohort study that characterises current practice.
We propose to apply these methods to Alzheimer?s disease, depression, psoriasis and psoriatic arthritis. In each case we will contrast the treatment effects mapped onto the QALY scale produced by the proposed multi-outcome approach to those produced by univariate synthesis. We will also show how the method can be adapted to the Markov state structures commonly used in cost-effectiveness models.