NCBI Bookshelf. A service of the National Library of Medicine, National Institutes of Health.

Robertson C, Arcot Ragupathy SK, Boachie C, et al. The Clinical Effectiveness and Cost-Effectiveness of Different Surveillance Mammography Regimens After the Treatment for Primary Breast Cancer: Systematic Reviews, Registry Database Analyses and Economic Evaluation. Southampton (UK): NIHR Journals Library; 2011 Sep. (Health Technology Assessment, No. 15.34.)

Cover of The Clinical Effectiveness and Cost-Effectiveness of Different Surveillance Mammography Regimens After the Treatment for Primary Breast Cancer: Systematic Reviews, Registry Database Analyses and Economic Evaluation

The Clinical Effectiveness and Cost-Effectiveness of Different Surveillance Mammography Regimens After the Treatment for Primary Breast Cancer: Systematic Reviews, Registry Database Analyses and Economic Evaluation.

Show details

7Economic evaluation

Introduction

This chapter has four main sections: a brief outline of the principles of economic evaluation, followed by sections reporting the methods, results and summary of the findings of the economic evaluation.

The objectives of this chapter are to determine whether or not (1) any method of surveillance could potentially be cost-effective for the whole population of women eligible for the service and (2) the method of surveillance should be varied between subgroups of women who are eligible for surveillance following surgery for breast cancer.

Principles of economic evaluation

A brief introduction to economic evaluation

The decision to use resources to provide one method of breast cancer surveillance means that the opportunity to use these resources in other desirable ways (either to provide another method of surveillance or to meet an entirely different health need) is given up. The cost of this decision is the benefits (health gains, etc.) that could have been obtained had the resources been used in another desirable way. This is the economic notion of ‘opportunity cost’. Strictly speaking, the opportunity cost of a decision to use resources in one way is equivalent to the benefits that could have been obtained had the resources been used to provide the next best alternative. Economic evaluation is a method of providing decision-makers with information about the opportunity cost of the decisions that could be made. It does this by comparing alternative courses of action in terms of both their costs and consequences.82

An economic evaluation in this context would involve assessing the relative costs and benefits associated with alternative surveillance regimens for breast cancer. The objective of such an economic evaluation would be to provide information to assist decision-makers in the allocation of available resources so that benefits could be maximised. A cost-effectiveness plane (Figure 24) illustrates how an economic evaluation brings together information on costs and benefits. The vertical axis represents the difference in costs between surveillance regimens (e.g. mammography vs MRI). The horizontal axis represents differences in effectiveness between the two regimens.

FIGURE 24. Relationship between the difference in costs and effects between a new (experimental) method of surveillance and an alternative (control) method.

FIGURE 24

Relationship between the difference in costs and effects between a new (experimental) method of surveillance and an alternative (control) method. NE, north-east; NW, north-west; SE, south-east; SW, south-west.

In the north-west and south-east quadrants of Figure 24 a clear decision about which surveillance regimen should be preferred is provided because one or other regimen is less costly but more effective (i.e. it dominates the other treatment). In the north-west quadrant the experimental regimen is more costly and provides less benefit, therefore the control regimen is more efficient (is dominant). In the south-east quadrant the opposite situation occurs and the experimental regimen is more efficient (is dominant), as it is less costly and provides more benefit. The circle in the centre of the figure represents the possibility that no meaningful differences in costs or benefits exist between the regimens and for practical purposes the two regimens are equally efficient.

In the two remaining areas of the figure, the north-east and south-west quadrants, a judgement is required as to whether the more effective regimen is worth the extra cost. To aid these judgements, information can be provided in terms of an incremental cost-effectiveness ratio (ICER). The higher the ICER of one intervention compared with another, the less likely it is that this intervention will be considered efficient.

Methods

Economic modelling of alternative surveillance regimens

A surveillance programme needs to be not only effective, but also cost-effective. Using Markov modelling methods, the cost-effectiveness of various surveillance programmes is compared. The economic model describes the pathway of care of individuals from the point where they received treatment for breast cancer and will receive some form of ongoing surveillance. This includes their longer-term (ideally their lifetime) costs and consequences, including those that might arise from any subsequent cancers. Surveillance can be considered as an event undertaken at discrete intervals and repeated over time and hence a Markov model was developed. This can be used to describe the logical and temporal sequence of events following the implementation of alternative surveillance regimens. We used the model to provide the estimated costs and outcomes for a selected period for a cohort of women for different surveillance regimens.

The model

Markov models comprise a set of states and at any point in time an individual will be in one of these states and will stay in that state for a defined period of time (the cycle length) before they are allowed to move to another state. The cycle length must be a period relevant to the condition considered (e.g. 6 months, 1 year, 18 months, etc.) At the end of each cycle, individuals can remain in the state in which they started the cycle or move to a different state. The probabilities of moving from one state to another are called transition probabilities. In each state, the model will assign costs and benefits for each individual according to different interventions and/or time spent in each state. In a Markov model, there must be at least one absorbing state, typically death, from which the person will not be able to leave.

Figure 25 shows a simplified version of the model presented for illustrative purposes (Appendix 27 contains a copy of a section of the full model structure). In this figure, the states are presented as ovals, whereas the arrows show the possible directions in which individuals could move at the end of each cycle. The rate at which an individual moves (makes a transition) between states is governed by the transition probabilities. The states considered in the model are thought to reflect possible paths of individuals. The top line in Figure 25 represents the possible path for individuals who start off after ‘successful’ (the belief being that the woman has been successfully treated for cancer but is at risk of developing subsequent disease) treatment free from cancer but who develop breast cancer over time but remain undiagnosed. The bottom section of Figure 25 represents those individuals who start in the model after ‘successful’ treatment free of cancer, but go on to develop IBTR or MCBC over time but are identified and treated for the disease.

FIGURE 25. Depiction of a simplified version of the Markov model.

FIGURE 25

Depiction of a simplified version of the Markov model. Risk profile refers to the mortality risk for a given cancer. In this simplified figure, it is assumed that cancers differ in terms of the risk of death.

If a woman initially has no evidence of IBTR or MCBC then over time she will have the chance of IBTR or MCBC occurring. The natural history of disease and the effectiveness of initial treatment determine the chance of this occurrence. Surveillance will not alter the chance of IBTR or MCBC occurring but may alter the chance of that cancer being detected, the stage at which it is found and hence the treatment and possible final outcome. Within the simplified version of the model shown in Figure 25 only three treatment states are depicted. These treatment states vary according to the risk profile of the breast cancer being treated. Once IBTR or MCBC is identified it is assumed that the cancer is treated and that subsequently individuals may have an altered life expectancy as a result of the recurrence. We also assumed that women who have had a further cancer will be judged as being at ‘moderate’ risk of developing further disease and so will have a more intensive follow-up. The absorbing state in the model is death. Any individual can move into this state from any other state within the model. The chance of moving into this state will be determined by the age of the woman through all-cause mortality and cancer-specific mortality. If a cancer is missed during surveillance then it is assumed that it will remain untreated until it is identified.

The model will compare different regimens but, for each regimen, a cohort of women will pass through the different health states. The costs per woman and speed at which they progress through the states will vary between regimens. The intuitive idea behind the model is to identify the regimen that leads to the most effective and cost-effective surveillance regimen.

Description of a woman's movement through a model regimen

The model includes women who may develop IBTR and or MCBC. The model itself does not differentiate between these situations. The model starts with a surveillance year; based on information from the survey of current practice this frequency could be once every 6, 12, 18, 24 or 36 months. In regimen 1, individuals can either be followed up using surveillance mammography or present to their GP with a symptom, i.e. discover a lump. Women who have a mammogram can either be identified as free of disease or have a positive mammogram. The model structure allows this to be either a true- or false-positive or a true- or false-negative. If the mammogram is a true-negative, individuals will then go back to the surveillance programme. If the mammogram is a false-negative, individuals also return to the surveillance programme, but these individuals would not receive any treatment or care for that cancer although it remains undetected. Furthermore, it is possible that if a cancer is undiagnosed or untreated the severity of the cancer will increase and the prognosis worsen, for example over time the tumour may increase in size. We handle this within the model by increasing the severity of untreated disease over time. It is also possible for individuals who have had a true-negative, over the duration of the cycle, to go on to develop breast cancer in a subsequent cycle. The likelihood of this happening will be dictated by the expected incidence of IBTR and MCBC over the cycle length (for example, if the cycle length were 1 year then it would be based upon an estimate of the annual incidence of IBTR and MCBC).

If the mammogram is a true-positive, individuals will be diagnosed with invasive or non-invasive cancer and managed appropriately. If the mammogram is a false-positive it will be assumed that the individual will undergo further invasive tests and on a negative finding of these tests they will return to the surveillance programme at the end of the cycle.

The regimens considered

We outlined the alternative surveillance regimens in Chapter 3 (see Methods for the survey). The intention was to compare each of these within the economic model. We planned to combine surveillance regimens for hypothetical cohorts of the population defined in terms of the nature of primary disease, treatment and demographic characteristics, etc. These cohorts reflect the prior hypothesised risk of IBTR and MCBC in the population of women previously treated surgically for a primary breast cancer.

As described in Methods for the survey, we identified nine different surveillance regimens. We reduced these to three regimens, which we felt broadly represented the most relevant comparators. Furthermore, as reported in Chapter 5, few data on the diagnostic performance of the alternative methods of identifying a breast cancer were available. Consequently, it was not possible to model all of these options. However, some data were available to facilitate the modelling of mammographic surveillance with and without clinical follow-up organised either through secondary care or through the screening service. The presentation of the woman following referral from primary care following the identification of a suspicious lump on self-examination was also modelled. We used this form of diagnosis in two specific ways within the model. First, we used it to define a situation where no formal surveillance is used. It is also used to model the possibility that a woman presents between surveillance points with symptoms suggestive to a GP of breast cancer, for example if surveillance is performed every 36 months then within this 36-month interval the model will allow a woman to present with clinical symptoms that are suggestive of breast cancer and for this cancer to be identified.

Populating the model with parameter estimates

To provide estimates of relative cost-effectiveness, the model requires estimated values for a range of different types of parameters. Such parameter estimates should be derived in a systematic and reproducible manner to avoid bias caused by the distorted and selective use of data.50 The assembly of such data need not necessarily be comprehensive; rather, effort should focus on identifying the most relevant data to the decision problem, which in this case was the comparison of alternative surveillance regimens for women after treatment for primary breast cancer.

We assembled the different types of data required for the economic model from analyses of existing data sets, a series of systematic reviews, and focused searches for specific pieces of data. We report the methods and results of the reviews and analyses of existing data sets in detail in Chapters 46. In brief, the broad types of data required to populate the economic model relate to:

  • the uptake of surveillance and follow-up
  • the prevalence, incidence and risk of progression of the disease, i.e. its epidemiology and natural history
  • the performance of different regimens (e.g. clinical examinations, mammograms, etc.) in terms of the accuracy of the diagnostic tests
  • resource use and unit costs required to estimate the costs of alternative surveillance regimens; the specific parameters and methods used to provide estimates that are relevant to the UK context
  • health-state utilities.

Within the model, we based estimates of uptake upon simplifying assumptions and advice from the members of the project Advisory Group. We derived the data on the natural history of women from the analysis of the large data set reported in Chapter 6. Further data relating to the management and outcomes came from the source data used to inform recent NICE guidelines.24

We derived information on the diagnostic performance of different types of clinical tests, for example the accuracy of mammography, from data reported in Chapter 5.

We derived data on the costs incurred for the different surveillance regimens and their consequences from structured reviews of the published literature, as well as routine data sources such as the NHS Reference Costs.40 The perspective for costs is the NHS.

Data on the utilities associated with differing severities of cancer and the possible differences in quality of life associated with various surveillance regimens were obtained from the published literature, including the review of economic evaluations, as described above, as well as a search of the Cost-Effectiveness Analysis Registry (CEA Registry: www.cearegistry.org/).

We report how we derived each of these sets of data and the values used in the model in more detail in the sections below.

Uptake of surveillance and follow-up

Within the model, we assumed that, if individuals are invited to attend surveillance, they do in fact attend. This may be too high, as approximately 75–80% of the normal population attend for breast screening. The other variable required for the model is the probability that a woman will present to the GP with symptoms that she thinks are suspicious. Based upon advice from the clinical members of the Advisory Group we assumed that 30% of women with prior treatment for breast cancer might present to the GP per annum. We then converted this percentage into a probability of presenting per 6-month cycle by fitting an exponential curve. The probability used within the model was 0.1393, i.e. in the no surveillance arm of the model, and for during the surveillance interval in the surveillance arms of the model, just under 14% of surviving women who have not been diagnosed with a recurrent cancer will present to a GP every 6 months. The following formula assumes that events occur at a constant rate over time: p = 1 − e−rt, where p = probability, e = base of natural logarithm, r = rate and t = time period.

Epidemiology and natural history of breast cancer

Data relating to the natural history of breast cancer required for the model can be split into four components. These are:

  1. recurrence/occurrence rates for women initially treated for breast cancer
  2. estimated survival of women without and with IBTR or MCBC
  3. estimated proportions of the different types of IBTR or MCBC occurring
  4. estimated change in the severity of untreated cancer over time.

IBTR and/or MCBC rates for women initially treated for breast cancer

Part of the analysis reported in Chapter 6 related to the time to event for IBTR and MCBC cancers. Using the estimated 10-year survival and the same methods as described above, an exponential curve was fitted so that the probability of experiencing an event for IBTR or MCBC per cycle (a 6-month period) was estimated. We report the estimated 10-year incidence rate for women initially treated by BCS or mastectomy, as well as the incidence per cycle (per 6-month period) in Table 27. Within the model, we assumed IBTR and MCBC events were independent and that the rates summed within the model to provide a net rate of cancer incidence. As described below this assumption was relaxed in sensitivity analyses, where we modelled the impact on costs and benefits of the incidence of the more serious IBTR events separately.

TABLE 27. Estimated incidence of IBTR and MCBC disease for women following surgery for primary disease.

TABLE 27

Estimated incidence of IBTR and MCBC disease for women following surgery for primary disease.

The data reported in Table 27 are taken to be representative of the rates expected for the whole population of women who received either BCS or mastectomy as part of the treatment of their primary cancer. It should be noted that the rates of MCBC in women who were originally treated by mastectomy are higher than the rates in those who were originally treated using BCS. These estimates are based upon observed estimates and the difference may simply be a reflection of imprecision in estimates, i.e. in reality no difference exists. Whether this is true or not is a matter for debate. A finding from Chapter 6 was that the incidence of subsequent cancer events was predicted in part by the characteristics of the primary cancer. Using the same methods described we have estimated the incidence per cycle for the reference case used in Cox proportional hazard models reported in Chapter 6 (Table 28).

TABLE 28. Alternative values for the subgroup of women taken as the reference case in Cox proportional hazard model.

TABLE 28

Alternative values for the subgroup of women taken as the reference case in Cox proportional hazard model.

We calculated upper and lower estimates of incidence by combining estimates of the hazard rates obtained from the Cox proportional hazard models with the event rates and probabilities reported in Table 29. It was assumed that hazard rates were additive but upper and lower estimates were based only on proportional hazard rates for factors, for example tumour size, age, grade, etc., which were found to be statistically significant at the 5% level in the analyses reported in Chapter 6.

TABLE 29. Hazard rates for high and low estimates and the incidence per cycle estimated from these rates.

TABLE 29

Hazard rates for high and low estimates and the incidence per cycle estimated from these rates.

Estimated survival of women with and without IBTR and/or MCBC

The economic model does not stop with the diagnosis of cancer. It seeks to model the impact on survival caused by delayed identification. To do this estimates of survival of women who develop a further case of cancer are required. Also required are estimates of survival for those women who do not develop further cancer. Estimates of the former depend upon whether or not the cancer was diagnosed and treated and the effectiveness of any treatment.

We conducted a structured review of relevant management guidelines to estimate data on the survival of women who go on to develop further cancer (reported in more detail in Appendix 28). The recent NICE guideline was identified as providing the best available evidence of treatments for early breast cancer relevant to the UK.24 The data used to support the NICE guideline recommendations came from the EBCTCG83 and Adjuvant! Online computer program.84 We prepared estimates of survival following various treatment options using the Adjuvant! Online computer program due to its flexibility. Adjuvant! Online draws on information from mortality statistics in the USA, the SEER database, and meta-analyses and individual clinical trials. Based on well-validated factors, such as age, menopausal status, oestrogen receptor (ER) status, tumour size and grade, nodes status, etc., predictions can be made about survival for alternative adjuvant treatment regimens, such as chemotherapy and hormone therapy. The programme derives survival estimates from the US population, however.

As survival estimates are linked to the effectiveness of treatments it was necessary to also define the therapy given for a cancer. The choice about what therapies would be adopted for which categories of cancer were based upon the recommendations in the NICE guideline24 and clinical advice from members of the study team about typical treatments within the UK. Based upon these data we derived specific therapies for cancers with specific characteristics. Using Adjuvant! Online, we estimated predictions of 10-year cancer-related mortality. Table 30 shows the simplified classification of prognostic factors used by Adjuvant! Online. We grouped these estimates, as described below, for cancers that had similar management costs and survival. This simplification was performed because Adjuvant! Online can provide more data than were readily manageable in the economic model, and the economic model itself was focused on surveillance regimens rather than treatments of women with breast cancer.

TABLE 30. The simplified classification of prognostic factors.

TABLE 30

The simplified classification of prognostic factors.

We grouped the different cancers into five different risk profiles, which had an increasingly worse prognosis. The cancers included in each risk profile had a predicted 10-year mortality that fell into the range defined for the risk profile. We derived the range for each risk profile following consideration of the mortality data derived from Adjuvant! Online and discussions within the Advisory Group.

As each risk profile contained several different cancers (defined in terms of size, ER status, etc.) an average mortality had to be calculated. To calculate this average mortality we required data on the estimated proportion of each type of cancer in each risk profile. These data were derived from the further analysis of the WMCIU Breast Cancer Registry data set used in Chapter 6. Cases were included if they were invasive tumours and diagnosed from 1997 (due to the extent of missing data prior to that date). We considered only tumours that were surgically treated. For the selected cases, descriptive information about the proportions of women with cancers with the following combination of characteristics were derived: ER positive or negative, grade of cancer (grades 1, 2 or 3), tumour size (0.1–2.0 cm, 2.1–5.0 cm, > 5.0 cm) and number of positive lymph nodes (zero, one to three, four or more). Unfortunately, ER status was mostly missing within the data set so could not be provided. Therefore, using published information85 we considered that 70% of cancers would be ER positive with the remainder being ER negative. We report these data in detail in Appendix 28. We assumed that the proportions of the different types of IBTR and MCBC would be the same as those for primary cancer. We made this assumption because there were more cases of primary cancer and hence less likelihood of there being no data to provide estimates for the combination of tumour characteristics described below.

From the WMCIU Breast Cancer Registry data, the proportion of each type of cancer (in terms of the proportion with a particular tumour grade, size, nodal involvement, etc.) was defined in each risk profile. We then multiplied the proportions by the 10-year mortality estimates for the corresponding cancer. We then summed the product of these calculations to give an average 10-year mortality rate for each risk profile.

Using the estimated 10-year mortality derived for each risk profile we fitted an exponential curve so that the probability of dying from cancer per cycle (a 6-month period) for each risk profile (Table 31) could be estimated using a formula similar to the one described above. For example, Table 31 shows that the average mortality rate from cancer at 10 years for risk profile state 1 was 4.86%, and, using the formula reported above, the risk of dying from cancer in any 6-month cycle was estimated to be 0.002%. Further detail of the data underpinning Table 31 is shown in Appendix 30.

TABLE 31. Mortality rates (%) from breast cancer at 10 years and per 6-month cycle for each risk profile (data used within the model).

TABLE 31

Mortality rates (%) from breast cancer at 10 years and per 6-month cycle for each risk profile (data used within the model).

This formula used to estimate the probability of dying from cancer per cycle (a 6-month period) assumes that deaths occur at a constant rate over time. If mortality is positively skewed then for a shorter time horizon of the model this may represent an underestimate of mortality, and it may overestimate mortality for longer time horizons. The cancer mortality data also assume that risk of death from cancer is independent of the women's age; this may underestimate the risks from cancer in younger women (i.e. those under 50 years of age).

These mortality rates are based on data for women who have received treatment for breast cancer. Therefore, they may not be applicable to women whose cancer is untreated because it is undetected. We hypothesised that at each time point a woman with an untreated cancer would face a higher risk of death in the next cycle (6-month period) than an identical woman whose cancer had been treated. This increased risk of death was proxied by comparing the estimated risk of dying at 5 years following a diagnosis of cancer for a woman diagnosed with cancer in the period 1980–4 with the risk for an identical woman from 2000 to 2004. The data used to derive the parameter value used in the model came from information produced by the Information and Statistics Division of NHS Scotland.86 The base-case value was based upon all women aged 15–74 years. In this group of women, expected 5-year mortality for women diagnosed between 1980 and 1984 was 34.9%. In the period 2000–4 the expected 5-year mortality was 14.9%. The ratio of these two numbers gives a value of 2.34. This value was used to inflate the 6-month breast cancer mortality rates reported in Table 31. Table 32 summarises the base-case and high and low values used within the model. Low and high values are based upon the lowest and highest values obtained for any age grouping reported by the Information and Statistics Division of NHS Scotland.

TABLE 32. Summary of breast cancer mortality inflators used within the model to derive breast cancer mortality for those with undiagnosed breast cancer.

TABLE 32

Summary of breast cancer mortality inflators used within the model to derive breast cancer mortality for those with undiagnosed breast cancer.

Within the base-case analysis we assumed that an IBTR that fits within a given risk profile will have the same probability of death per 6-month cycle period as an otherwise identical MCBC. The data reported in Chapter 6 suggests that mortality following IBTR may be substantially higher than the mortality for an otherwise identical MCBC, however. As noted above, we conducted sensitivity analyses to explore the impact of the increased risk of death from IBTR. We derived the increased risk of death per 6-month period by multiplying the HR for mortality from IBTR reported in Chapter 6 with the mortality rates for treated and untreated cancers. The point estimate for the hazard rate for death for IBTR was 1.76 (with an upper value from the 95% CI of 2.13). We used the extremes of the CIs for this hazard rate to define low and high rates within the model.

Data were also required on all-cause mortality. These data were required because women who do not develop cancer still have a chance of dying from other causes. In addition, women who do develop cancer also have the risk of dying from other causes. For both sets of women as they age within the model, mortality will increase. Estimates of all-cause mortality were obtained from the published UK life tables for the years.87 From these data a mortality rate for each 6-monthly cycle was calculated using the previously defined formula. This is reported in detail in Appendix 31.

Estimated proportions of the different types of IBTR and/or MCBC cancers occurring

For the model, information was needed not just on whether a cancer occurs or not, but also on the severity of that cancer. We assumed that at the point where a cancer technically becomes detectable the size of the cancer is below 1 cm in diameter. In terms of the risk profile classification defined above we further assumed that all these people are initially in risk profile category 1 at the point when the cancer becomes detectable. Over time, an undetected cancer will increase in severity and the estimates used to model this are described below.

Estimated change in the risk profile of untreated cancer over time

One variable required for the economic evaluation is the rate at which an undiagnosed cancer may move to a worse risk profile (with a consequent reduction in life expectancy and quality of life and an increase in treatment costs). We sought data on which to base estimates for this variable from a structured review of the literature relating to doubling time of a breast cancer and the factors, for example grade of cancer, which might affect the doubling time of tumours. We sought these data as the individual patient analysis reported in Chapter 6 found that a significant predictor of mortality was tumour size, with larger-sized tumours having a shorter life expectancy than smaller tumours.

It is recognised that there is a considerable degree of uncertainty about the rates of growth of breast cancers.88 However, data were sought about plausible rates of growth and about potential range in the rates of growth that can be explored in a subsequent sensitivity analysis. A summary of the findings of this structured review is reported in Appendix 32. Given the information found in this review of the literature, it is clear that there is little consensus on the doubling times of breast cancer tumours. The data we have used in the economic model are based on the information provided by Peer and colleagues,89 taking the mean doubling time in tumour volume to be 157 days. We tested the consequences of this in a high/low sensitivity analysis where we will vary this rate between the plausible extremes of the data presented in Table 33.

TABLE 33. Mean breast cancer doubling times by age.

TABLE 33

Mean breast cancer doubling times by age.

To use information on the mean doubling time within the model we had to estimate how long it would take a tumour of the minimum technically identifiable size to increase in size, where an untreated tumour would move from one risk profile to a risk profile with a worse prognosis. We took the minimum diameter of a detectable cancer to be 0.75 cm. This value was taken because the volume of a cancer with this diameter is close to the minimum volume size considered by Adjuvant! Online. The data on time to reach the threshold tumour size were converted into risks of increasing the risk profile by one level for each cycle that a cancer remains untreated. We performed this using the same methods described above to estimate incidence and mortality rates.

Diagnostic performance of tests

As reported in Chapter 5, relatively few data were available on the diagnostic performance of any of the tests. Within the model, we assumed that at the time a woman receives a diagnostic test as part of surveillance she is asymptomatic.

For IBTR, we based data on data reported in Chapter 5, and summarised in Table 20 (Chapter 5), and on discussions with the clinical experts involved in the study. Where relevant published data were available in the absence of pooled data the study judged to be closest to the median of reported results was used to inform the values chosen for the base-case analysis. We used data from other studies to define plausible extremes. Where it was feasible for these tests to be used then they were also used for MCBC, as few additional data were available (Table 34).

TABLE 34. Diagnostic performance of the different tests.

TABLE 34

Diagnostic performance of the different tests.

For surveillance mammography the values used in the base-case analysis were based upon those derived from Drew and colleagues.67 We based low and high estimates of sensitivity upon the ranges for these parameters reported in Table 20 (Chapter 5). These data represent extreme values that will be used in the sensitivity analysis.

As reported in Chapter 5, only one study provided data on the sensitivity and specificity of mammography and clinical follow-up. These data did not seem plausible (e.g. the reported sensitivity was 100%). The values reported in Table 34 are assumptions derived following discussions with clinical experts. The consensus of opinion was that the combination of follow-up and mammography would slightly improve the sensitivity and specificity. In a sensitivity analysis we will explore the impact of changing these values between high and low estimates. We will also seek to identify whether there is a threshold in terms of diagnostic performance, which would make the additional cost of clinical follow-up worthwhile.

Within the model, data are also required for the diagnostic performance of a clinical examination when performed by a GP. Again, few data were available and following discussions we assumed that the rates used within the model would be slightly lower than those reported in the systematic review of diagnostic performance (Chapter 5).

We considered the impact of using a higher cost but more effective diagnostic test. As a proxy for such a test data were based upon the performance of MRI. It should be noted that the values identified, especially at the upper level, are where MRI has been used in a higher risk group of women. Hence, the values are not necessarily illustrative of MRI itself but rather of a hypothetical test. The values for the base-case analysis were based upon those reported by Drew and colleagues67 but it was assumed that the sensitivity was slightly less than perfect (i.e. 95% vs the 100% reported by Drew and colleagues).65 Low values of sensitivity were based upon data from Warner and colleagues90 who conducted a systematic review of prospective studies in which women at very high risk for breast cancer were screened with both MRI and mammography.90 Hence even these data may not be fully representative of women eligible for surveillance mammography. The specificity values were informed by the estimates of one study91 included in the Warner and colleagues review.90 This study had the lowest specificity of any of the studies included in Warner and colleagues' 2008 study.90

Costing data

The costs of surveillance were broken down into the following cost categories:

  • Cost of:

    inviting women for screening

    the surveillance test (e.g. mammogram, MRI, clinical examination)

    health-care professional time (e.g. GP consultation, clinical examination)

    further invasive tests (e.g. core biopsy)

    treatment (e.g. mastectomy, radiotherapy, drug treatment).

Tables 3537 show the cost estimates used in the economic model. All costs are reported in 2008 pounds sterling. Table 35 shows the current cost of the alternative screening strategies. The cost of inviting women to attend screening was obtained from a recent HTA report.92 The cost of the alternative surveillance tests were all derived from routine sources. The cost of a mammogram was based on information from the NHSBSP 2009.93 The NHSBSP estimates the cost of a mammogram in England to be £37.50 per woman invited and £45.50 per woman screened. An alternative costing source was obtained from the Scottish Breast Screening Programme, which estimates the cost of a mammogram to be £77.80.94 The implications of the variation in costs between Scotland and England were explored in a sensitivity analysis. The cost of an MRI was estimated as being twice the cost of that reported in the NHS Reference Costs40 for an outpatient MRI. This is because an MRI on a breast takes twice as long as a normal MRI and involves the use of a contrast. The lower quartile and upper quartile of the NHS Reference Costs40 for this category are used to inform sensitivity analysis. We derived the costs of a clinical examination from routine data sources. The cost of a GP clinical examination was obtained from the Personal Social Services Research Unit (PSSRU)95 and was based on the average cost of a GP consultation. In addition, we also included the cost of a clinical examination conducted in a secondary care setting by either a consultant or non-consultant. These costs were obtained from NHS Reference Costs.40 Information on the range of costs (lower and upper quartile) was also available and these were used as upper and lower estimates in sensitivity analysis.

TABLE 35. Cost of screening regimens.

TABLE 35

Cost of screening regimens.

TABLE 36. Cost of invasive tests and treatments.

TABLE 36

Cost of invasive tests and treatments.

TABLE 37. Surveillance regimen: clinical examination plus mammography for women receiving hormone therapy.

TABLE 37

Surveillance regimen: clinical examination plus mammography for women receiving hormone therapy.

The costs of further invasive tests were obtained from a NICE evidence review group (ERG) report96 and inflated to current prices using the PSSRU inflation index. The cost of a mastectomy was based on the same source, and inflated to 2008 prices. The cost of radiotherapy was based on the cost of complex treatment on a mega-voltage machine,40 assuming that women get on average 20 sessions of radiotherapy. This assumption was based on information from the PRIME trial, which reported that, on average, women receive 20 sessions of radiotherapy.97 Again, lower quartile and upper quartile estimates of the cost of a single session of radiotherapy will be used in a sensitivity analysis.

The costs of drug treatment, for example the cost of hormone treatment, chemotherapy and combined treatment, were obtained from recent NICE guidance. The cost of hormone treatment was based on information reported in the costing template for technology appraisal guidance 112.98 This included the costs of tamoxifen for 5 years and the cost of aromatase inhibitors (anastrozole or letrozole) for 5 years. The cost of chemotherapy was based on the costs reported in NICE technology appraisal guidance 109.99 The cost of chemotherapy is based on the cost of two different regimens (TAC – taxotere, adriamycin and cyclophosphamide; FEC – fluorouracil, epirubicin and cyclophosphamide). This is based on six cycles of treatment.

Each risk profile consists of a series of different types of cancers (defined in terms of ER status, grade, size and number of lymph nodes involved). As described above, an average mortality for each risk profile was estimated by combining information on the expected mortality for each specific cancer within a risk profile with information on the proportion of women in that risk profile that had that specific type of cancer. Adjuvant! Online reports mortality by the type of adjuvant therapy used. The clinical members of the research team determined, based on UK practice, which specific cancer in a profile would receive hormone therapy and/or chemotherapy. Using information on the proportion of cancers in a given risk profile that would be treated with a given adjuvant therapy a proportion of a cost of a course of hormone treatment or radiotherapy was incorporated into the cost assigned to each risk profile.

Table 37 shows the costs of one surveillance regimen for a woman invited to screening and who received a clinical examination and a mammogram. The costs include the costs of screening, the mammogram and clinical examination, conducted by a consultant. On a positive mammogram, the woman would then go on to have further invasive tests to confirm the result (core biopsy). On a true-positive finding, the woman would have a mastectomy followed by radiotherapy, followed by drug treatment (depending on the severity of the IBTR or MCBC). We based the costs of treatment on a number of assumptions:

  • It is assumed that all ER+ women will receive hormone treatment. It is assumed that those women who have an excellent prognosis (survival rate at 10 years of 96% or greater) and are postmenopausal will receive tamoxifen for 5 years. Women who are postmenopausal, with a poorer prognosis, will receive an aromatase inhibitor for 5 years.
  • All women who are premenopausal and are ER+ will receive tamoxifen.
  • All women who have grade 3 tumours will receive chemotherapy.
  • Women who are ER+ and have positive lymph nodes will receive combined treatment (hormone + chemotherapy).
  • Women who are ER− and have 0 nodes will receive no treatment (exception to this is that 15% might get hormone therapy).
  • Women who are ER− and have positive lymph nodes will receive chemotherapy (exception to this is that 15% might receive combined therapy).

Health-state utility values

The primary purpose of the economic model was to inform decision-making in a UK setting, given that treatment for breast cancer affects not only survival, but also quality of life, for example different types and stages of cancer are likely to be associated with differences in quality of life, as would different treatment options. Therefore, we have also sought to assess the impact on quality of life, through the incorporation of health-state utility weights, which have been combined with estimates of survival to estimate QALYs.

Recent guidance suggests that estimates of QALYs should ideally be based on generic health-state valuation methods using UK population tariffs.100 Therefore, we conducted a focused search of the literature and other relevant sources such as the Harvard cost–utility database. We identified a number of studies reporting health-state utilities. In particular, we found a recent systematic review of breast cancer utility weights.101 In their systematic review, 59 studies were identified for review and nine studies included. Of the nine studies included, three were based on UK data.102104 In addition, the utility values used in the paper by Sorensen and colleagues105 were based on a combination of UK and US data.105

It is difficult to determine how comprehensive this review is as, being available as a conference poster, the details provided on the literature searching are brief. The authors searched an appropriate selection of databases but the sensitivity of the search strategies used is unclear due to a lack of information. Missing information included whether MeSH terms were ‘exploded’ to include more specific terms, which Emtree terms were used in EMBASE, and how the terms were combined in the final search. From the information reported, one error was noted: ‘breast neoplasms’ was incorrectly described as a non-MeSH term.

Overall, the authors of the systematic review found considerable variability and inconsistency in the reported utility values. A selection of other studies eliciting health-state utilities was further identified. Overall, there was considerable variation in values and in definitions of health states; however, there is a general trend in the values reported in the literature. As would be expected, utilities decrease with increasing breast cancer severity and utilities are also found to be sensitive to treatment. For example, there is a general trend for those receiving chemotherapy to have lower utility values than those receiving hormone therapy, most likely due to the severity of the side effects of the respective treatments.

For the economic model, we have used the results reported in the systematic review of breast cancer utility weights.101 Using this information, we defined utilities for each of the five risk profiles in the model. For example, risk profile state 1 assumes a utility state with a low value of 0.75 and a high value of 0.85 (based on the distribution of values from the systematic review). We adjusted these utility states to include a decrement for those women who will receive chemotherapy. This decrement is based on the percentage of women in each of the five severity states who would receive chemotherapy. For example, 24% of women in risk profile state 1 would receive chemotherapy. The chemotherapy decrement is based on information on patients' utilities for cancer treatments.106 In their study, using the time trade-off method utilities for chemotherapy were estimated to be 0.74 from an actual health state estimated to be 0.94. All health-state utilities after treatment are assumed to be the same as the utilities defined before treatment without the chemotherapy decrement.

Utility values for risk profile states 3 and 4 are based on the health-state values in Tosteson and colleagues.107 This is based on the value for regional cancer in the age group 50–59 years. The utility value for risk profile state 5 is based on the value provided for distant rather than regional cancer in the age group 50–59 years. Each of these values has also been reduced by the decrement factor for chemotherapy. To achieve the high values reported in Table 38 for risk profile states 4 and 5 an additional 0.05 was added to the low value.

TABLE 38. Health-state utilities.

TABLE 38

Health-state utilities.

The values used in the base-case analysis are the low values reported in Table 38. Individuals in a ‘no-cancer’ state are assumed to have a health-state utility value of 0.80 in the base-case analysis.

Key assumptions of the economic model

This section provides a brief summary of the key assumptions made when developing the economic model.

Structural assumptions

The cycle length is assumed to be 6 months.

It is assumed that, if individuals are invited to attend mammographic screening, they do in fact attend. This assumption may be too high, as approximately 75–80% of the normal population attend for breast screening.

Strategies compared are assumed to be homogeneous in that they do not change over time. More sophisticated strategies where the surveillance intervals and method of follow-up change over time have not been modelled.

Estimates of survival were based upon predictions derived from Adjuvant! Online and were grouped into five broad groupings based upon survival. This is a simplification of the different types of tumour that might occur, as well as how prognosis of untreated disease might change over time.

Parameter value assumptions – natural history assumptions

The incidence of IBTR and MCBC are assumed to be independent.

Incidence is assumed to have occurred at the point when a cancer could technically be identified.

The grade of IBTR does not have to be the same as that of the primary tumour but the grade of IBTR or MCBC does not have to change over time. There is some evidence to suggest that grade does not change. Should it occur as other evidence suggests, then, given the model structure, this would reduce the life expectancy of a woman. However, the impact on cost-effectiveness is unclear as it depends upon the likelihood of a tumour progressing to a higher grade, the speed of progression, the diagnostic performance of the surveillance regimen and the surveillance interval.

Estimates of survival are assumed to be independent of age (age-adjusted all-cause mortality is included as a separate model parameter). This may underestimate risks to younger women whose cancers might be more aggressive.

Treatments for IBTR and MCBC were based upon NICE guidelines and expert opinion.

If a cancer is not detected in a given cycle (6 months) then it is assumed that it can only advance one risk profile level. The likelihood of this occurring was estimated from the literature on doubling times.

Many of the estimates used to model natural history of disease are assumed to be constant over time. Some of these assumptions are not consistent with the observed data. However, the consensus of opinion for other parameters, for example probability of progressing to a risk profile with a worse prognosis, is that they may decline over time. Allowing such probabilities to change over time would not greatly change costs but might be expected to increase QALYs over time for those strategies which have a better diagnostic performance.

Parameter value assumptions – diagnostic performance

Sensitivity and specificity for surveillance mammography and clinical follow-up were based upon expert opinion and assumed that it performed slightly better than surveillance mammography alone.

Sensitivity and specificity for clinical examination was based on clinical opinion. It was assumed that clinical examination has a lower sensitivity and specificity than reported in Chapter 5.

Parameter value assumptions – cost assumptions

Management packages previously prepared for UK guidance and HTAs have been used to estimate care, and hence costs. If treatment patterns have greatly altered in the last few years these data may not be applicable.

It is assumed that all women who have a breast cancer will incur the cost of a mastectomy. Obviously a woman who has had a previous mastectomy cannot have a further mastectomy on the same breast but the cost of a mastectomy has been used as a proxy for the cost of care such a woman would receive.

Parameter value assumptions – utilities assumptions

Data from different populations and elicited using different methods have been assumed to be sufficiently similar to the relevant UK population of women to be useful.

Utilities are assumed to reduce with increasing severity of disease and also on the use of chemotherapy. Other decrements to utility, for example complications of disease or treatment, are not modelled.

Presentation of results

The base-case analysis was run for a cohort of women (starting age in the model 57 years) with surveillance occurring once yearly. The starting age was chosen as this was the mean age of the women contributing to the analysis of data from the WMCIU Breast Cancer Registry, which was reported in Chapter 6. The model was run for different starting ages in further sensitivity analysis. The cycle length of the model is 6 months and cumulative costs and benefits are estimated over a maximum of 100 cycles, which is equivalent to a time horizon of 50 years. This time horizon was taken as a proxy for life expectancy of women treated for primary breast cancer. All costs are reported in 2008 pounds sterling and effectiveness in QALYS. A discount rate of 3.5% for costs and benefits was used following guidelines for NICE.100 Results are presented as incremental cost per QALY gained. The modelling exercise will use a net benefit framework to combine cost and benefit estimates. The results of the analyses will be presented as point estimates of mean incremental costs, effects, incremental cost per QALY. This measure is a ratio of the difference in costs divided by the difference in effectiveness between two alternative strategies. These data can be interpreted as how much society would have to pay for an extra unit of effectiveness. Whether or not a more costly but more effective regimen is considered worthwhile depends upon society's willingness to pay for a QALY and, within England, the threshold adopted by NICE lies somewhere between £20,000 and £30,000.

Incremental cost per QALYs is a common way for presenting the results of an economic evaluation. They are, however, difficult to interpret when the choice is between several mutually exclusive options. In this circumstance the judgement can be informed by considering the net benefit statistic. The regimen with the greatest net benefit at a given value for society's willingness to pay for a QALY is considered to be most cost-effective. The net benefit statistic itself is defined as:

NBj=(QALYj×λ)costj
[Equation 1]

where NB = net benefit, QALYi = QALYs for intervention i, costi = cost for intervention i, and λ = society's willingness to pay per QALY.

Intervention i would be chosen over intervention j when NBi > NBj.

Sensitivity analysis

We did not conduct probabilistic sensitivity analysis. The reason for this is that parameter values used are statistically imprecise and, as data are so limited, the model estimates may be unreliable. Therefore, the results of the economic evaluation should be interpreted cautiously and, at most, indicate situations where a particular method(s) of surveillance may be worthy of further consideration. Nevertheless, we conducted both one-way and multiway sensitivity analysis to assess how results may change as a consequence of plausible changes in parameter values. We also used deterministic sensitivity analysis to identify threshold values for key parameters. The methods used in the sensitivity analysis are described below.

Probability of developing IBTR or MCBC

We varied the probability of developing IBTR or MCBC in the sensitivity analysis from a low of 0.0030 [the lowest estimated 6-month cycle failure probability was for women who had a mastectomy and cumulatively had a risk of IBTR or MCBC of 0.0036 (Table 27)] to 0.0125 [the highest 6-month probability recurrence rate for IBTR and MCBC combined from the predicted HRs was 0.0125 (Table 29)].

Inflating the risk of death from cancer for people who are unmanaged

We also explored the effect of an increase in the risk of death for unmanaged individuals in sensitivity analyses. This was varied in the sensitivity analysis from the base-case assumption of 2.34 to a high of 2.916 (Table 32 in the base-case model).

We repeated the same analysis in the IBTR model, varying the increase in the risk of death in unmanaged states from 2.33 to 2.916.

Changes to the risk of progressing to a higher risk profile

The consequences of changing the risk of unmanaged women progressing to higher-risk profiles was explored in both the base-case model and the IBTR model. The risk of progressing was altered from the base-case estimate of 0.1555 to 0.2623 in sensitivity analysis; 0.26 relates to a mean doubling time of 80 days and an estimated time of 300 days for a tumour to reach 2 cm3 (Table 33). In addition, in a further sensitivity analysis the risk of progressing was further lowered form the base-case assumption, to a mean doubling time of 942 days. This equates to a risk of 0.0923 per 6-month cycle.

Sensitivity and specificity of the surveillance tests

The diagnostic performance of the surveillance tests (sensitivity and specificity) was varied in the base-case model for both high and low sensitivity and specificity values (reported in Table 34). This was undertaken in multiway sensitivity analysis, varying all the tests simultaneously.

Sensitivity analysis on costs

A range of sensitivity analyses on costs were performed. This included high treatment costs and surveillance costs. For example, the base-case model assumes that all women who receive hormone therapy receive tamoxifen. The consequences of this assumption were tested in sensitivity analyses. This involved re-estimating the model when all treatment costs were set to their highest estimates (highest cost for hormone and chemotherapy). In addition to treatment costs, higher surveillance costs were also incorporated into this sensitivity analysis. This included using the higher mammography cost (£77.80 as opposed to base-case assumption of £45.50) and also doubling the cost of an MRI. These cost estimates can be seen in Tables 35 and 36. The same analyses were conducted for the IBTR model.

The cost of all clinical examinations conducted either at the point of surveillance or in follow-up visits were varied from the base-case assumption that these clinical examinations would be carried out by a consultant grade, to the alternative assumption that these examinations were carried out by a non-consultant (Table 35). This analysis was conducted for both the base-case model and the IBTR model.

Sensitivity analysis on health-state utilities

Health-state utility values were also tested in a range of sensitivity analyses. This included replacing the base-case values for quality of life (assumed to be the low values reported in Table 38) with high estimates of quality of life (Table 38).

In addition, quality of life was further tested in both the base-case model and the IBTR model by varying the quality of life in unmanaged states. The base-case model assumes that women in unmanaged states have the same quality of life as women free of disease. This assumption is tested in sensitivity analysis by giving women in unmanaged states the same quality of life as women after treatment.

Age

The effect of age was tested in the sensitivity analysis with the base-case model and IBTR model, which we ran for a starting age of 40 years and a starting age of 70 years. This analysis was conducted as multiway sensitivity analysis with a range of values varied in the analysis. For the younger age group (starting age 40 years) this included high chemotherapy and high hormone therapy costs (Table 36), high cancer incidence (0.0152, Table 29) and a short doubling time (0.2623, Table 33). In addition, for this age group, a further multiway analysis was conducted, repeating the analysis above, with the addition of the high inflation factor for unmanaged states (2.916, Table 32).

For a starting age of 70 years, sensitivity analysis was conducted for the base-case model and IBTR model using the new starting age of 70 years and the low hazard rate (0.0018, Table 29). All of the other variables were assumed to be as the base-case assumptions.

Results

Base-case results

Results for women who received BCS or mastectomy for their primary cancer

Tables 39 and 40 report the results of the base-case analyses for the average women treated for their primary breast cancer with BCS or with a mastectomy. These data can be used to inform judgements about what would be the single best regimen for the NHS to adopt for all women who had previously been treated with BCS or all women who had previously been treated with mastectomy. For both populations the results are shown for a range of surveillance intervals ranging from 12 months to 36 months. The costs and outcomes for the ‘no surveillance’ option are the same regardless of the surveillance interval. For the other surveillance regimens, both costs and QALYs fall as the surveillance interval increases. However, for each surveillance regimen the reduction in QALYs is more than compensated for by a reduction in cost. This is illustrated by the reduction in the incremental cost per QALY reported for each regimen as the surveillance interval increases. For example, for women who had received BCS the incremental cost per QALY for mammography alone compared with ‘no surveillance’ is £4727 for a 12-month surveillance regimen and £3811 for an 18-month regimen. Similarly, for a 12-month surveillance regimen the incremental cost per QALY for mammography and clinical follow-up alone compared with mammography is £236,826. For an 18-month surveillance interval the incremental cost per QALY falls to £118,455.

TABLE 39. Results of the base-case analysis for women treated for their primary cancer with BCS.

TABLE 39

Results of the base-case analysis for women treated for their primary cancer with BCS.

TABLE 40. Results of the base-case analysis for women treated for their primary cancer with mastectomy.

TABLE 40

Results of the base-case analysis for women treated for their primary cancer with mastectomy.

The results of the two sets of analyses reported in Tables 39 and 40 are very similar. As would be expected, the no surveillance regimen is least costly but also least effective. Whether or not a more costly but more effective regimen is considered worthwhile depends upon society's willingness to pay for a QALY. Within England, the threshold adopted by NICE lies somewhere between £20,000 and £30,000, and, as shown in Tables 39 and 40, only one regimen, mammography alone, is associated with an incremental cost per QALY below £20,000.

Incremental cost per QALYs, as reported in Tables 39 and 40, can be difficult to interpret when the choice is between several different options (e.g. in Table 39 there are 13 different regimens and it is not immediately obvious which option might be considered most cost-effective). Therefore, we used the net benefit statistic to compare regimens. When society's willingness to pay for a QALY is £20,000, the regimen that is associated with the highest net benefit is mammography alone every year (Tables 39 and 40). When the threshold was increased to £30,000 mammography only had the greatest net benefit. Regardless of the surveillance interval, mammography alone had the highest net benefits and the regimen with the lowest net benefit was always no surveillance.

In Table 39 the impact of substituting a more effective but more costly surveillance technology (MRI) for mammography was considered. Regardless of the surveillance interval, this regimen was associated with a net benefit greater than that of the no surveillance regimen but less than that of all of the other regimens.

Modelling IBTR alone

The analyses reported in Tables 39 and 40 made the assumption that the consequences of an IBTR are the same as those of an otherwise identical contralateral recurrence. However, the analyses reported in Chapter 6 suggest that the mortality associated with IBTR is substantially higher than that associated with an otherwise identical recurrence in the contralateral breast. In this analysis this increased risk of death from IBTR is modelled (Table 41).

TABLE 41. Ipsilateral breast tumour recurrence has a substantial higher risk of death than MCBC.

TABLE 41

Ipsilateral breast tumour recurrence has a substantial higher risk of death than MCBC.

In these analyses the regimen mammography alone has an incremental cost per QALY compared with ‘no surveillance’ of < £4000 and the highest net benefit regardless of the surveillance interval. Mammography alone at 12 months has a marginally higher net benefit than mammography alone at 24 months for both a £20,000 and a £30,000 threshold for society's willingness to pay for a QALY.

Sensitivity analyses

We conducted a range of different sensitivity analyses, as described above in Presentation of results. As the results for the analyses for women who received BCS for their primary cancer are similar to those obtained when we consider women who received a mastectomy for their primary cancer we present sensitivity analyses solely for the scenario where women received BCS for their primary cancer. However, we also report selected analyses for a model that considers IBTR alone.

Sensitivity analysis around the breast-conserving model

Probability of developing cancer

Figures 2629 illustrate the impact on incremental cost per QALYs as the incidence of cancer increases. In each figure, three lines are shown:

FIGURE 26. Incremental cost per QALYs for the different surveillance regimens at a 12-month surveillance interval.

FIGURE 26

Incremental cost per QALYs for the different surveillance regimens at a 12-month surveillance interval.

FIGURE 27. Incremental cost per QALYs for the different surveillance regimens at an 18-month surveillance interval.

FIGURE 27

Incremental cost per QALYs for the different surveillance regimens at an 18-month surveillance interval.

FIGURE 28. Incremental cost per QALYs for the different surveillance regimens at a 24-month surveillance interval.

FIGURE 28

Incremental cost per QALYs for the different surveillance regimens at a 24-month surveillance interval.

FIGURE 29. Incremental cost per QALYs for the different surveillance regimens at a 36-month surveillance interval.

FIGURE 29

Incremental cost per QALYs for the different surveillance regimens at a 36-month surveillance interval.

  1. The incremental cost per QALY of mammography alone compared with no surveillance. This line can be used to inform the question: is it worth adopting the more effective but more costly mammography alone follow-up in place of the less costly and less effective no surveillance regimen?
  2. The incremental cost per QALY of mammography plus clinical follow-up compared with mammography alone. This line can be used to inform the question: is it worth adopting the more effective but more costly mammography plus clinical follow-up in place of the less costly and less effective mammography alone regimen?
  3. The incremental cost per QALY of MRI plus clinical follow-up compared with mammography plus clinical follow-up. This line can be used to inform the question: is it worth adopting the more effective but more costly MRI plus clinical follow-up in place of the less costly and less effective mammography plus clinical follow-up regimen?

The results of the analysis shown in these figures suggest that:

  • At all screening intervals considered some form of active surveillance might be considered cost-effective.
  • Should the incidence of IBTR and MCBC increase towards the upper values of incidence considered, which are typical of those we might expect for higher risk women (e.g. those whose primary cancers were of higher grade, who were younger than 50 years and who had lymph node involvement), a regimen of clinical follow-up and mammography is more likely to be worthwhile. Furthermore, when the surveillance interval is 24 months the incremental cost per QALY compared with mammography alone approaches £30,000. At a surveillance interval of 36 months, it is approximately £25,000.
  • As the screening interval and risk of IBTR and MCBC increases towards 36 months, it becomes more likely that a more costly but more effective surveillance intervention (in this analysis typified by MRI plus clinical follow-up) might be worthwhile.

Inflating the risk of death from untreated cancer

Sensitivity analysis was conducted to test the effect of inflating the risk of death for women who are unmanaged for cancer. In the sensitivity analysis the increased risk of death from cancer in unmanaged states was inflated from the base-case risk of 2.34 to a high of 2.196 (Table 42). The results of this analysis were broadly similar to the base-case analysis presented in Table 39. We conducted this sensitivity analysis for both the base-case model and the IBTR model.

TABLE 42. Increasing the risk of death for unmanaged disease.

TABLE 42

Increasing the risk of death for unmanaged disease.

Changes to the probability of progressing to a more serious risk profile

Table 43 reports the impact of increasing the speed that an untreated cancer progresses to a risk profile with a worse prognosis. As would be expected the higher the probability of progression (which would be analogous to a shorter doubling time of a tumour) the more likely earlier and more intensive follow-up becomes. Nevertheless, in this one-way sensitivity analysis none of the options, other than mammography alone, is associated with incremental costs per QALY approaching a value that society typically might be willing to pay.

TABLE 43. The impact of changing the probability of progressing to a higher risk profile.

TABLE 43

The impact of changing the probability of progressing to a higher risk profile.

Changes to the sensitivity and specificity of the tests

Tables 44 and 45 show multiway sensitivity analysis on the diagnostic performance of the surveillance tests. Again, changes in the sensitivities and specificities alone do not greatly alter the estimated cost-effectiveness of the different regimens.

TABLE 44. High sensitivity and specificity values.

TABLE 44

High sensitivity and specificity values.

TABLE 45. Low sensitivity and specificity values.

TABLE 45

Low sensitivity and specificity values.

Changes to costs of tests and treatments

Sensitivity analysis was also performed on costs. This included a high treatment cost and high surveillance cost sensitivity analysis. This involved re-estimating the model when all treatment costs were set to their highest estimates (highest cost for hormone and chemotherapy). In addition to treatment costs, higher surveillance costs were also incorporated into this sensitivity analysis. This included using the higher mammography cost (£78 as opposed to base-case assumption of £45.50) and also doubling the cost of an MRI. These cost estimates can be seen in Table 46. The same analyses were conducted for the IBTR model (and are reported in the next subsection).

TABLE 46. High-cost sensitivity analysis.

TABLE 46

High-cost sensitivity analysis.

In addition, Table 47 shows sensitivity analysis in which the cost of a clinical examination is priced at the consultant and non-consultant rate and the effect on the incremental cost per QALYs. Although the changes are minor, adopting a lower cost for a clinical examination makes the mammography alone regimen marginally less cost-effective compared with ‘no surveillance’. This is because the lower cost is also incurred for all clinical examinations, including those during follow-up for those with IBTR or MCBC.

TABLE 47. Consultant and non-consultant cost of clinical examinations.

TABLE 47

Consultant and non-consultant cost of clinical examinations.

Changes to utility estimates used

The analysis presented in Table 48 is based on the high estimates of quality of life reported in Table 38. The results suggest that at a willingness-to-pay threshold of £20,000 the regimen with the highest net benefit is likely to be mammography alone at 12-month surveillance intervals, followed by mammography alone at 18-, then 24- and then 36-month intervals. At a willingness-to-pay threshold of £30,000 the ordering is very similar with the exception that mammography plus clinical examination has the fourth highest net benefit at a willingness-to-pay threshold of £30,000.

TABLE 48. Sensitivity analysis using high estimates of quality of life.

TABLE 48

Sensitivity analysis using high estimates of quality of life.

In addition to sensitivity analysis on higher quality-of-life values, we conducted further sensitivity analyses to test the base-case assumption that women in unmanaged states have the same quality of life as women who are disease free. We tested this assumption by giving all women in unmanaged states the same utility as women who had been treated for IBTR or MCBC. These results are presented in Table 49. Decreasing the quality of life of women in unmanaged states has no appreciable effect on the analysis presented above for a threshold value of £20,000 or £30,000.

TABLE 49. Sensitivity analysis on quality of life in unmanaged states: breast-conserving model.

TABLE 49

Sensitivity analysis on quality of life in unmanaged states: breast-conserving model.

Exploration of the impact of age at the time surveillance starts

Table 50 reports the results of sensitivity analysis based on a starting age in the model of 40 years old. This table not only reports multiway sensitivity analysis for a starting age of 40 years, but also includes high chemotherapy costs and high hormone therapy costs (Table 36). In addition, a higher incidence rate for cancer is used in this model (0.0152, Table 29) and high probability of moving to the next risk profile. This analysis was conducted for surveillance intervals ranging from 12 to 36 months. In this analysis, the option with the highest net benefit is mammography alone at a surveillance interval of 12 months for a willingness-to-pay threshold of £20,000 per QALY. At a higher willingness-to-pay threshold of £30,000 per QALY, MRI plus clinical examination at a surveillance interval of 12 months has the highest net benefit. The second highest net benefit at a threshold of £20,000 and £30,000 is mammography and clinical examination at 12-month surveillance intervals.

TABLE 50. Sensitivity analysis for a starting age of 40 years: breast-conserving model.

TABLE 50

Sensitivity analysis for a starting age of 40 years: breast-conserving model.

In addition to a starting age of 40 years old, sensitivity analysis was also conducted on a higher starting age. In this model all parameters are assumed to be as the base-case assumptions with two differences: (1) starting age is 70 years old and (2) the incidence of cancer is based on the lowest hazard estimates (Table 29). The results of this analysis are presented in Table 51. At a willingness-to-pay threshold of £20,000 per QALY, the test with the highest net benefit is mammography alone at a 36-month interval. At a willingness-to-pay threshold of £30,000, the regimen with the highest net benefit is mammography alone at intervals of 24 months.

TABLE 51. Sensitivity analysis for a starting age of 70 years: breast-conserving model.

TABLE 51

Sensitivity analysis for a starting age of 70 years: breast-conserving model.

Sensitivity analysis around model considering IBTR only

Probability of developing cancer

Figures 3033 report the impact of increasing the incidence of IBTR only. In these analyses, the impact on costs and QALYs or MCBC is not considered. As described earlier, IBTR has a much worse prognosis than MCBC. As would be expected, as incidence increases the incremental cost per QALYs fall when we compare a more effective but more costly surveillance regimen with the next most costly and effective. For mammographic and clinical follow-up the incremental cost per QALY compared with mammography alone begins to fall below £30,000 once the incidence of disease exceeds 0.00775 every 6 months and the surveillance interval is 24 months or longer. When the surveillance interval reaches 36 months, the adoption of a more effective but more costly regimen (again typified by MRI plus clinical follow-up) may be cost-effective once the incidence per cycle exceeds 0.0068.

FIGURE 30. Incremental cost per QALYs for the different surveillance regimens at a 12-month surveillance interval.

FIGURE 30

Incremental cost per QALYs for the different surveillance regimens at a 12-month surveillance interval.

FIGURE 31. Incremental cost per QALYs for the different surveillance regimens at an 18-month surveillance interval.

FIGURE 31

Incremental cost per QALYs for the different surveillance regimens at an 18-month surveillance interval.

FIGURE 32. Incremental cost per QALYs for the different surveillance regimens at a 24-month surveillance interval.

FIGURE 32

Incremental cost per QALYs for the different surveillance regimens at a 24-month surveillance interval.

FIGURE 33. Incremental cost per QALYs for the different surveillance regimens at a 36-month surveillance interval.

FIGURE 33

Incremental cost per QALYs for the different surveillance regimens at a 36-month surveillance interval.

Inflating the risk of death from untreated cancer

In this sensitivity analysis, the increased risk of death from cancer in unmanaged states was inflated from the base-case risk of 2.34 to a high of 2.196 (Table 52). The results of this analysis were broadly similar to the base-case analysis presented in Table 41.

TABLE 52. Increasing the risk of death for unmanaged disease.

TABLE 52

Increasing the risk of death for unmanaged disease.

Changes to the probability of progressing to a more serious risk profile

Table 53 reports the impact of increasing the speed that an untreated cancer progresses to a risk profile with a worse prognosis. The results of this analysis are similar to those reported above and it is unlikely that changes in this variable alone will result in any regimen other than mammography alone having an incremental cost per QALY that society might be willing to pay.

TABLE 53. The impact of changing the probability of progressing to a higher risk profile.

TABLE 53

The impact of changing the probability of progressing to a higher risk profile.

Changes to costs of tests and treatments

Table 54 shows a high treatment cost and high surveillance cost sensitivity analysis for the IBTR model. Again, the results of this sensitivity analysis are broadly similar to those reported in Table 41.

TABLE 54. Ipsilateral breast tumour recurrence model: high costs.

TABLE 54

Ipsilateral breast tumour recurrence model: high costs.

In addition, Table 55 shows sensitivity analysis in which the cost of a clinical examination is priced at the consultant and non-consultant rate. Although the incremental cost per QALYs changes, none is of sufficient magnitude to change conclusions.

TABLE 55. Consultant and non-consultant cost of clinical examinations.

TABLE 55

Consultant and non-consultant cost of clinical examinations.

Changes to utility estimates used

We tested the base-case assumption that women in unmanaged states have the same quality of life as women who are disease free (Table 56). In this analysis, all women in unmanaged states had the same utility as women who had been treated for IBTR or MCBC. A similar pattern of results is observed in the IBTR model sensitivity analysis as was observed in the breast-conserving model sensitivity analysis.

TABLE 56. Sensitivity analysis on quality of life in unmanaged states: IBTR model.

TABLE 56

Sensitivity analysis on quality of life in unmanaged states: IBTR model.

Exploration of the impact of age at the time surveillance starts

Table 57 reports the results of sensitivity analysis based on a starting age of 40 years old in the IBTR model. This table reports multiway sensitivity analysis for starting age 40 years, and includes the high chemotherapy costs and high hormone therapy costs (Table 36). In addition, the higher incidence rate for cancer is used in this model (0.0152, Table 29) and a short doubling time. This analysis was conducted for surveillance intervals ranging from 12 to 36 months.

TABLE 57. Sensitivity analysis for starting age of 40 years: IBTR.

TABLE 57

Sensitivity analysis for starting age of 40 years: IBTR.

Summary

In the base-case analysis the regimen with the highest net benefit and, therefore, most likely to be considered cost-effective was mammographic surveillance alone provided yearly. This result holds for women who had previously been treated for their primary cancer with either BCS or mastectomy or women who suffer IBTR.

As might be expected in a comparison of surveillance regimens, the results of the model are very sensitive to changes in the incidence of recurrent cancer. When the expected incidence is increased toward the maximum that could possibly be expected for any group of women mammography and clinical surveillance potentially becomes cost-effective when the surveillance interval is 24 months or longer. As the surveillance interval and incidence increase regimens that are more costly but more effective may also have incremental costs per QALY below typical threshold values. This suggests that there may be some scope for research into alternative technologies that could be used for surveillance.

The results of the analysis did not substantially alter for any of the other sensitivity analyses reported. The exception to this is when we changed several parameter values simultaneously. This was undertaken in an attempt to compare surveillance regimens for a hypothetical 40-year-old woman (who can be thought of as having a greater likelihood of developing IBTR or MCBC) and a hypothetical 70-year-old woman (representing a lower likelihood of developing IBTR or MCBC). In the sensitivity analysis conducted for a 40-year-old woman, the following changes were made: the incidence of recurrent cancer was increased and the time it took for an undetected cancer to progress to risk profiles with a worse prognosis was reduced. Furthermore, it was assumed that should IBTR or MCBC be detected then it would be treated more aggressively (and at higher cost). For 40-year-old women facing these risks and costs, mammographic surveillance every 12 months had the highest net benefit, although it was only slightly greater than mammography and clinical follow-up every 12 months. These results suggest that a more intensive follow-up of women judged to be at high risk may be cost-effective. Conversely, for women at lower risk it may be more cost-effective for surveillance to be performed less often (every 2 or 3 years) with mammography alone or another similarly less intensive and costly test or tests.

© 2011, Crown Copyright.

Included under terms of UK Non-commercial Government License.

Bookshelf ID: NBK100087

Views

  • PubReader
  • Print View
  • Cite this Page
  • PDF version of this title (4.4M)

Other titles in this collection

Recent Activity

Your browsing activity is empty.

Activity recording is turned off.

Turn recording back on

See more...