The guideline development group raised issues of heart-rate control, cardioversion, and contra- indications for warfarin. These were not included in the model, but were addressed by introductory statements, produced by group consensus. When using ‘the basic model’, the advice for all patients with three or more risk factors, and for men with left ventricular hypertrophy and any one other risk factor, was to treat with warfarin. This was incorporated in the flow chart. Otherwise, clinicians were referred to the associated tables. The increased risk of cerebral bleed associated with warfarin was taken into account in the model, with the effectiveness of warfarin estimated for all strokes, both haemorrhagic and ischaemic. However, for patients with a baseline risk of stroke less than 50% greater than the risk of cerebral bleed on warfarin, the reduction in the risk of stroke afforded by warfarin could be outweighed by this increased risk (assuming warfarin affords an approximate two-thirds reduction). Using an analogous method to that used in estimating the risk of a non-cerebral bleed, the risk of cerebral bleed on warfarin was estimated: it ranged from 0.15% in patients aged 60 years to 1.6% in patients aged 85 years and over. This suggests that it may be prudent not to treat patients below risks ranging from 0.23% (0.15 ×1.5) at aged 60 to 2.4% (1.6 ×1.5) aged 85 and over. In only ten of the 1512 cells, all for women aged 80 years and over, did the results of the model based on median values for patient utilities recommend treatment for patients below these thresholds. These cells are hence classified as ‘do not treat’ in the tables.
Comment
The use of decision analysis with its explicit population of a model enabled the elements of the decision-making process, and their impli- cations, to be made explicit. It also enabled incorporation of a wider range of available ‘evidence’, particularly patient utilities. While the approach allows for explicit quantification of the uncertainty that underlies an apparently straightforward binary clinical decision this was only explored for two dimensions of the decision – the uncertainty around the effectiveness of warfarin and patient utilities. In both circum- stances the result of modelling such uncertainty
55 was considerable. Interestingly, it was not
possible to model the impact of varying the two dimensions together.
The interaction between the guideline develop- ment group and the evidence (in this case the construction of the model) was very different from the interactions within the other case studies described. Once the clinical problem had been scoped there was little remaining role for the group and they were not called upon to discuss the evidence or the implications of the model. The model produces the decisions and users of the model are required to trust the product. Such packaging of data does not allow explicit
consideration of the uncertainty around the various dimensions of the decision by those involved in making it. It is currently unclear how the model decisions relate to what a patient would decide. Nor is it clear that it is necessary (or feasible) to collect explicit patient utilities within routine care settings.
While a number of these uncertainties are clearly amenable to empirical research (reflecting the relative absence of attempts to use decision analysis in guideline development), the dissociation be- tween the guideline development group and the synthesis of the evidence is a unique feature of this case study.
57
T
he rationale for the development of clinicalpractice guidelines is to present a rigorous exploration of the evidence and delivery issues surrounding treatment options in healthcare, conducted by appropriately constructed groups of health professionals, consumers and specialists. We have presented our experiences based on a case series of 11 guidelines developed over a period of 5 years. The initial focus of this project was to explore the methods of incorporating cost issues within clinical guidelines. However, this exploration has been paralleled by a more general development of the methods of treating evidence within the process of developing guide- line development. Therefore, the process of reviewing evidence in guideline development groups incorporates an increasing sophistication not only in considerations of cost but also in review techniques and group process. At the outset of the project it was unclear how narrowly or broadly the concept of ‘cost’ could be con- sidered. It is now clear that, alongside the effectiveness data and data describing quality of life, cost issues can successfully be repre- sented as part of a broad profile of treatment attributes.
The use of an epidemiological and health-service resource summary early on in the process of developing each guideline has proved a useful device to begin the process of thought in each guideline development group about the import- ance and value of treatments. Following on from this, a profile of costs and consequences provides a representation that is readily comprehensible to guideline readers of any background. By explicitly identifying uncertainties, the presentation of the evidence accurately identifies strengths and weaknesses so that guideline development group members (and subsequently end users of the guideline report) can easily explore alternative values. The profile provides the starting point for a guideline development group to begin the process of valuing treatment alternatives and thereby producing recommendations. Some- times further work may be needed using various forms of modelling to help a group to explore fully the meaning of the information
before them.
As identified in chapter 3, the available evidence on which these presentations are based is not necessarily robust. Patient-oriented outcomes are reported particularly inconsistently in trials and it may be necessary to supplement meta- analytic clinical end-point findings with a narrative summary of quality-of-life findings where available. It may be necessary to access a wide range of sources to describe rare iatrogenic events, resource implications and unit cost data. For example, it was necessary to look at 3 years of coroners’ findings in order to characterise rates of fatal poisoning associated with different antidepressants, since rates are too low to be captured within randomised controlled trials but toxicity is perceived to be a big concern.
There may be evidence that health-associated costs borne by patients and carers (e.g. travel and time to receive care, over-the-counter drugs, disability costs) and indirect costs of lost earnings differ significantly between alternative treatments. This should be considered relevant to a treatment decision at least in as much as it may undesirably influence adherence to treatment. There is the possibility that organisational alternatives may shift costs from the health service to individuals and the appropriateness of this may depend on the disease considered and contextual circumstances. Seldom are there adequate data to address costs borne by patients but where this is a concern these costs can be described as attributes of treatments. While all of the 11 guideline development processes described support the rigorous identi- fication of a range of evidence, they raise a series of generic issues: summarising study outcomes; time-frame issues; approaches to dealing with more complex disease areas; the development of profiles and models; and the role of health economists in guidelines.