Cost-effectiveness analysis in health care (II)
Like in the first post I cannot promise I have not already covered the topics I’m about to cover in this post before on the blog. In this post I’ll include and discuss material from two chapters of the book: the chapters on how to measure, value, and analyze health outcomes, and the chapter on how to define, measure, and value costs. In the last part of the post I’ll also talk a little bit about some research related to the coverage which I’ve recently looked at in a different context.
In terms of how to measure health outcomes the first thing to note is that there are lots and lots of different measures (‘thousands’) that are used to measure aspects of health. The symptoms causing problems for an elderly man with an enlarged prostate are not the same symptoms as the ones which are bothering a young child with asthma, and so it can be very difficult to ‘standardize’ across measures (more on this below).
A general distinction in this area is that between non-preference-based measures and preference-based measures. Many researchers working with health data are mostly interested in measuring symptoms, and metrics which do (‘only’) this would be examples of non-preference-based measures. Non-preference based measures can again be subdivided into disease- and symptom-specific measures, and non-disease-specific/generic measures; an example of the latter would be the SF-36, ‘the most widely used and best-known example of a generic or non-disease-specific measure of general health’.
Economists will often want to put a value on symptoms or quality-of-life states, and in order to do this you need to work with preference-based measures – there are a lot of limitations one confronts when dealing with non-preference-based measures. Non-preference based measures tend for example to be very different in design and purpose (because asthma is not the same thing as, say, bulimia), which means that there is often a lack of comparability across measures. It is also difficult to know how to properly trade off various dimensions included when using such metrics (for example pain relief can be the result of a drug which also increases nausea, and it’s not perfectly clear when you use such measures whether such a change is to be considered desirable or not); similar problems occur when taking the time dimension into account, where problems with aggregation over time and how to deal with this pop up. Various problems related to weighting are recurring problems; for example a question can be asked when using such measures which symptoms/dimensions included are more important? Are they all equally important? This goes for both the weighting of various different domains included in the metric, and for how to weigh individual questions within a given domain. Many non-preference-based measures contain an implicit equal-interval assumption, so that a move from (e.g.) level one to level two on the metric (e.g. from ‘no pain at all’ to ‘a little’) is considered the same as a move from (e.g.) level three to level four (e.g. ‘quite a bit’ to ‘very much’), and it’s not actually clear that the people who supply the information that goes into these metrics would consider such an approach to be a correct reflection of how they perceive these things. Conceptually related to the aggregation problem mentioned above is the problem that people may have different attitudes toward short-term and long-term health effects/outcomes, but non-preference-based measures usually give equal weight to a health state regardless of the timing of the health state. The issue of some patients dying is not addressed at all when using these measures, as they do not contain information about mortality; which may be an important variable. For all these reasons the authors argue in the text that:
“In summary, non-preference-based health status measures, whether disease specific or generic, are not suitable as outcome measures in economic evaluation. Instead, economists require a measure that combines quality and quantity of life, and that also incorporates the valuations that individuals place on particular states of health.
The outcome metric that is currently favoured as meeting these requirements and facilitating the widest possible comparison between alternative uses of health resources is the quality-adjusted life year“.
Non-preference-based tools may be useful, but you will usually need to go ‘further’ than those to be able to handle the problems economists will tend to care the most about. Some more observations from the chapter below:
“the most important challenge [when valuing health states] is to find a reliable way of quantifying the quality of life associated with any particular health state. There are two elements to this: describing the health state, which […] could be either a disease-specific description or a generic description intended to cover many different diseases, and placing a valuation on the health state. […] these weights or valuations are related to utility theory and are frequently referred to as utilities or utility values.
Obtaining utility values almost invariably involves some process by which individuals are given descriptions of a number of health states and then directly or indirectly express their preferences for these states. It is relatively simple to measure ordinal preferences by asking respondents to rank-order different health states. However, these give no information on strength of preference and a simple ranking suffers from the equal interval assumption […]; as a result they are not suitable for economic evaluation. Instead, analysts make use of cardinal preference measurement. Three main methods have been used to obtain cardinal measures of health state preferences: the rating scale, the time trade-off, and the standard gamble. […] The large differences typically observed between RS [rating scale] and TTO [time trade-off] or SG [standard gamble] valuations, and the fact that the TTO and SG methods are choice based and therefore have stronger foundations in decision theory, have led most standard texts and guidelines for technology appraisal to recommend choice-based valuation methods [The methods are briefly described here, where the ‘VAS’ corresponds to the rating scale method mentioned – the book covers the methods in much more detail, but I won’t go into those details here].”
“Controversies over health state valuation are not confined to the valuation method; there are also several strands of opinion concerning who should provide valuations. In principle, valuations could be provided by patients who have had first-hand experience of the health state in question, or by experts such as clinicians with relevant scientific or clinical expertise, or by members of the public. […] there is good evidence that the valuations made by population samples and patients frequently vary quite substantially [and] the direction of difference is not always consistent. […] current practice has moved towards the use of valuations obtained from the general public […], an approach endorsed by recent guidelines in the UK and USA explicitly recommend that population valuations are used”.
Given the very large number of studies which have been based on non-preference based instruments, it would be desirable for economists working in this field to somehow ‘translate’ the information contained in those studies so that this information can also be used for cost-effectiveness evaluations. As a result of this an increasing number of so-called ‘mapping studies’ have been conducted over the years, the desired goal of which is to translate the non-preference based measures into health state utilities, allowing outcomes and effects derived from the studies to be expressed in terms of QALYs. There’s more than one way to try to get from a non-preference based metric to a preference-based metric and the authors describe three approaches in some detail, though I’ll not discuss those approaches or details here. They make this concluding assessment of mapping studies in the text:
“Mapping studies are continuing to proliferate, and the literature on new mapping algorithms and methods, and comparisons between approaches, is expanding rapidly. In general, mapping methods seem to have reasonable ability to predict group mean utility scores and to differentiate between groups with or without known existing illness. However, they all seem to predict increasingly poorly as health states become more serious. […] all forms of mapping are ‘second best’, and the existence of a range of techniques should not be taken as an argument for relying on mapping instead of obtaining direct preference-based measurements in prospectively designed studies.”
I won’t talk too much about the chapter on how to define, measure and value costs, but I felt that a few observations from the chapter should be included in the coverage:
“When asking patients to complete resource/time questionnaires (or answer interview questions), a particularly important issue is deciding on the optimum recall period. Two types of recall error can be distinguished: simply forgetting an entire episode, or incorrectly recalling when it occurred. […] there is a trade-off between recall bias and complete sampling information. […] the longer the period of recall the greater is the likelihood of recall error, but the shorter the recall period the greater is the problem of missing information.”
“The range of patient-related costs included in economic valuations can vary considerably. Some studies include only the costs incurred by patients in travelling to a hospital or clinic for treatment; others may include a wider range of costs including over-the-counter purchases of medications or equipment. However, in some studies a much broader approach is taken, in which attempts are made to capture both the costs associated with treatments and the consequences of illness in terms of absence from or cessation of work.”
An important note here which I thought I should add is that whereas many people unfamiliar with this field may translate ‘medical costs of illness’ with ‘the money that is paid to the doctor(s)’, direct medical costs will in many cases drastically underestimate the ‘true costs’ of disease. To give an example, Ferber et al. (2006) when looking at the costs of diabetes included two indirect cost components in their analysis – inability to work, and early retirement – and concluded that these two cost components made up approximately half of the total costs of diabetes. I think there are reasons to be skeptical of the specific estimate on account of the way it is made (for example if diabetics are less productive/earn less than the population in general, which seems likely if the disease is severe enough to cause many people to withdraw prematurely from the labour market, the cost estimate may be argued to be an overestimate), but on the other hand there are multiple other potentially important indirect cost components they do not include in the calculation, such as e.g. disease-related lower productivity while at work (for details on this, see e.g. this paper – that cost component may also be substantial in some contexts) and things like spousal employment spill-over effects (it is known from related research – for an example, see this PhD dissertation – that disease may impact on the retirement decisions of the spouse of the individual who is sick, not just the individual itself, but effects here are likely to be highly context-dependent and to vary across countries). Another potentially important variable in an indirect cost context is informal care provision. Here’s what they authors say about that one:
“Informal care is often provided by family members, friends, and volunteers. Devoting time and resources to collecting this information may not be worthwhile for interventions where informal care costs are likely to form a very small part of the total costs. However, in other studies non-health-service costs could represent a substantial part of the total costs. For instance, dementia is a disease where the burden of care is likely to fall upon other care agencies and family members rather than entirely on the health and social care services, in which case considering such costs would be important.
To date [however], most economic evaluations have not considered informal care costs.”
No comments yet.