Cost-effectiveness of using prognostic information to select women with breast cancer for adjuvant systemic therapy
Williams C., Brunskill S., Altman D., Briggs A., Campbell H., Clarke M., Glanville J., Gray A., Harris A., Johnston K., Lodge M.
OBJECTIVES: To investigate the cost-effectiveness of using prognostic information to identify patients with breast cancer who should receive adjuvant therapy. DATA SOURCES: Electronic databases from 1980 through to February 2002. A survey of clinical practice in UK cancer centres and units. Large retrospective dataset containing data on prognostic factors, treatments and outcomes for women with early breast cancer treated in Oxford. REVIEW METHODS: Between six and nine databases were searched by an information expert. Evidence-based methods were used to review and select those studies and the quality of each included paper was assessed using standard assessment tools reported in the literature or piloted and developed for this study. A survey of clinical practice in UK cancer centres and units was carried out to ensure that conclusions drawn from the report could be implemented. These data, along with the information gathered in the systematic reviews, informed the methodological approach adopted for the health economic modelling. An illustrative framework was developed for incorporating patient-level prediction within a health economic decision model. This framework was applied to a large retrospective dataset containing data on prognostic factors, treatments and outcomes for women with early breast cancer treated in Oxford. The data were used to estimate directly a parametric regression-based risk equation, from which a prognostic index was developed, and prognosis-specific estimates of the baseline breast cancer hazard could be observed. Published estimates of treatment effects, health service treatment costs and utilities were used to construct a decision analytic framework around this risk equation, thus enabling simulation of the effectiveness and cost-effectiveness of adjuvant therapy for all possible combinations of prognostic factors included in the model. RESULTS: The lack of good-quality systematic reviews and well-conducted studies of prognostic factors in breast cancer is a striking finding. There are no registers of studies of prognostic factors or of reviews of prognostic studies. Many of the reviews used weak methods, primary studies are similar with poor methodology and reporting of results. In addition, there is much variation in patient populations, assay methods, analysis of results, definitions used and reporting of results. Most studies appear to be retrospective and some use inappropriate methods likely to inflate outcomes such as optimising cut points and failing to test the results in an independent population. Very few reviews used meta-analysis to conduct a pooled analysis and to provide an estimate of the average size of any association. Instead, most reviews relied on vote counting. Although many prognostic models for breast cancer have been published, remarkably few have been re-examined by independent groups in independent settings. The few validation studies have been carried out on ill-defined samples, sometimes of smaller size and short follow-up, and sometimes using different patient outcomes when validating a model. The evidence from the validation studies shows support for the prognostic value of the Nottingham Prognostic Index (NPI). No new prognostic factors have been shown to add substantially to those identified in the 1980s. Improvement of this index depends on finding factors that are as important as, but independent of, lymph node, stage and pathological grade. The NPI remains a useful clinical tool, although additional factors may enhance its use. We accepted that hormone receptor status (ER) for hormonal therapy such as tamoxifen and prediction of response to trastuzumab by HER2 did not require systematic review, as the mechanism of action of these drugs requires intact receptors. There was no clear evidence that other factors were useful predictors of response and survival. The survey confirmed pathological nodal status, tumour grade, tumour size and ER status as the most clinically important factors for consideration when selecting women with early breast cancer for adjuvant systemic therapy in the UK. The protocols revealed that although UK cancer centres appear to be using the same prognostic and predictive factors when selecting women to receive adjuvant therapy, much variation in clinical practice exists. Some centres use protocols based upon the NPI whereas others do not use a single index score. Within NPI and non-NPI users, between-centre variability exists in guidelines for women for whom the benefits are uncertain. Consensus amongst units appears to be greatest when selecting women for adjuvant hormone therapy with the decision based primarily upon ER or progesterone receptor status rather than combinations of a number of factors. Guidelines as to who should receive adjuvant chemotherapy, however, were found to be much less uniform. Searches of the literature revealed only five published papers that had previously examined the cost-effectiveness of using prognostic information for clinical decision-making. These studies were of varying quality and highlight the fact that economic evaluation in this area appears still to be in its infancy. By combining methodologies used in determining prognosis with those used in health economic evaluation, it was possible to illustrate an approach for simulating the effectiveness (survival and quality-adjusted survival) and the cost-effectiveness associated with the decision to treat individual women or groups of women with different prognostic characteristics. The model showed that effectiveness and cost-effectiveness of adjuvant systemic therapy have the potential to vary substantially depending upon prognosis. For some women therapy may prove very effective and cost-effective, whereas for others it may actually prove detrimental (i.e. the reductions in health-related quality of life outweigh any survival benefit). CONCLUSIONS: Outputs from the framework constructed using the methods described here have the potential to be useful for clinicians, attempting to determine whether net benefits can be obtained from administering adjuvant therapy for any presenting woman; and also for policy makers, who must be able to determine the total costs and outcomes associated with different prognosis based treatment protocols as compared with more conventional treat all or treat none policies. A risk table format enabling clinicians to look up a patient's prognostic factors to determine the likely benefits (survival and quality-adjusted survival) from administering therapy may be helpful. For policy makers, it was demonstrated that the model's output could be used to evaluate the cost-effectiveness of different treatment protocols based upon prognostic information. The framework should also be valuable in evaluating the likely impact and cost-effectiveness of new potential prognostic factors and adjuvant therapies.