N.2.1. Type of evaluation

Although this topic would be a suitable candidate for cost utility modelling, it was determined that a simple cost consequences model adapted from the analysis developed for the 2013 guideline update would be sufficient to provide estimates of incremental costs and outcomes associated with each of the comparators. This type of evaluation was chosen as the previous analysis demonstrated that chemoprevention as an overall strategy is likely to be cost effective at a threshold of £20,000 per QALY. Therefore the objective of this analysis was to compare the relative costs, benefits, and side effects of different chemoprevention in their natural units.

N.2.2. Target population

The population for this analysis is high- and moderate-risk postmenopausal women with no personal history of breast cancer, who have no history or increased risk of thromboembolic disease or endometrial cancer, and who are eligible for chemoprevention with any of the chemopreventive agents included in the analysis. Risk levels are defined according to the following lifetime risks of breast cancer from age 20:

• High risk population: 30% or greater
• Moderate risk population: Greater than 17% but less than 30%

N.2.3. Interventions

The interventions included in the analysis are the following chemopreventive agents, administered once-daily over a five year period:

• Tamoxifen 20mg
• Raloxifene 60mg
• Anastrozole 1mg

N.2.4. Comparator

The comparator in the analysis is no chemoprevention.

N.2.5. Time horizon

Since chemoprevention has the potential to reduce the long-term incidence of breast cancer, a lifetime time horizon was used in this analysis.

N.2.6. Health outcomes

The primary measure of health outcomes in the analysis is cases of breast cancer prevented as a result of chemoprevention. Secondary measures are incremental endometrial cancer cases, thromboembolic events, and fractures, compared to no chemoprevention.

N.2.7. Perspective

The analysis was conducted from the perspective of the NHS and personal and social services (PSS).

N.2.8. Discounting

A discount rate of 3.5% per annum was applied to all costs incurred after the first year.

N.2.9. Model structure

A Markov structure with a cycle length of one year was used to simulate the progression of patients over a lifetime time horizon. The model used three health states: healthy, breast cancer, and dead, as shown in Figure 2.

Figure 2Model structure

For the no chemoprevention arm of the model, age-specific incidence rates of breast cancer and baseline mortality were used to inform progression probabilities. Each year, living patients (either with or without breast cancer) could also experience adverse events: endometrial cancer, venous thromboembolism, or fracture. Patients could experience four categories of fracture: hip fracture, wrist fracture, vertebral fracture, or other fracture. The number of adverse events of each category per year was calculated by multiplying the number of living patients by the annual baseline probability of experiencing each event.

For the treatment arms of the model (tamoxifen, raloxifene, and anastrozole) relative risks were applied to the baseline incidence rates for breast cancer incidence, endometrial cancer, venous thromboembolism, and fracture in order to inform transition probabilities and adverse event probabilities.

N.2.10. Patient cohort characteristics

The assumption was made that patients of the age 50 and above are postmenopausal. The model simulated the progression of four patient age groups of patients: 50-59 years old, 60-69 years old, 70-79 years, and 80 years old plus, in order to capture the heterogeneity of the patient population. The distribution of patients among these age groups was derived from Office for National Statistics mid-2015 population data for females in the UK [accessed 23^rd September 2016], and is shown in Table 17.

N.2.11. Baseline breast cancer risk

Due to the scarcity of data on age-dependent baseline risk of breast cancer, five-year risks of breast cancer were generated using the BOADICEA assessment tool [http://ccge.medschl.cam.ac.uk/boadicea/] for representative high- and moderate-risk patients. Baseline risk data are displayed in Table 18. These values were converted to one year risks in order to inform transition probabilities for the model.

Patient profiles in the BOADICEA tool were selected to produce lifetime breast cancer risks which were consistent with definitions of high- and moderate-risk patients in this guideline: high-risk patients are associated with a 30% or greater risk of breast cancer from age 20 and moderate-risk patients are associated with a risk of between 17% and 30%.

The high-risk BOADICEA profile consisted of a 20 year old female patient who had not been tested for genetic mutations, but whose mother had a confirmed BRCA1 mutation. The moderate-risk profile consisted of a 20 year old female patient who had not been tested for genetic mutations, and whose mother had also not been tested, but had been diagnosed with breast cancer at age 20.

N.2.12. Mortality rate

Age-specific mortality data for females were taken from Office for National Statistics National Life Tables: England and Wales for 2013-15 [accessed 23^rd September 2016]

N.2.13. Baseline adverse event risks

Baseline risks for the incidence of endometrial cancer, thromboembolic events, and fractures were taken from the placebo arm of Cuzick et al. (2015). This study was selected as it is the analysis of tamoxifen with the largest patient numbers identified by the clinical review. Pooling of studies from the clinical literature review to derive baseline data was not possible, as analyses used varying follow-up lengths. Table 19 displays baseline adverse event incidence rates used in the model.

The distribution of fracture categories was derived from Scholes et al (2014) – an epidemiological study of the prevalence of fractures in adults aged 55 and above. Table 20 shows the relative incidence of fracture categories.

N.2.14. Relative risks for chemoprevention

Relative risks of breast cancer incidence and adverse event incidence were taken from the results of the clinical literature review and are shown in Table 21.

N.2.15. Adherence to chemoprevention

For the base case, the assumption was made that 50% of patients discontinued chemoprevention after one year of treatment, with the remaining 50% continuing treatment for the full 5 years. This estimate was elicited via expert opinion. It was assumed that patients discontinuing after one year had the same risk of breast cancer and adverse events as patients with receiving no chemoprevention.

N.2.16. Duration of treatment effect

The assumption was made that, in the base case, reduction in the incidence of breast cancer resulting from chemoprevention persisted over patients’ lifetime. It was assumed that change in risk level of endometrial cancer and thromboembolic events caused by chemoprevention lasted for the duration of treatment (i.e. the first five years of the model), and the change in risk level of fractures lasted for five years after the end of treatment (i.e. the first ten years of the model), after which incidence rates of adverse events returned to baseline levels.

N.2.17. Costs

Annual costs of tamoxifen, raloxifene, and anastrozole were taken from the September 2016 Electronic Drug Tariff [accessed 23^rd September 2016], and are shown in Table 22.

The assumption was made that all patients receiving chemoprevention would also incur the cost of two GP consultations per year while treatment was ongoing. The cost of a GP consultation was estimated to be £38 (including direct care staff costs and excluding qualification costs), taken from the PSSRU Unit Costs of Health and Social Care 2015. It was also assumed that patients treated with anastrozole incurred the cost of a DEXA scan at the outset of treatment, to screen patients for osteoporosis, due to concerns of reduction in bone mineral density associated with aromatase inhibitors. The cost of a DEXA scan was estimated to be £62, taken from the NHS National Tariff Payment System 2016/17 [accessed 23^rd September, 2016].

Costs associated with a case of breast cancer were taken from the model for the 2013 guideline update, which were originally derived from the NICE costing report published with the full guideline. These costs were adjusted to 2015 prices using consumer price index rates from the Office for National Statistics. Costs associated with a case of breast cancer are shown in Table 23.

Costs of endometrial cancer and thromboembolic events were also taken from the model produced for the 2013 update, adjusted to 2015 prices. The cost of endometrial cancer was originally sourced from Hind et al (2007). The cost of a thromboembolic event was assumed to be that of deep vein thrombosis, derived from NHS Reference Costs 2011/12. Costs of hip fracture, wrist fracture, vertebral fracture, and other fractures were taken from Dolan et al (1998), and adjusted to 2015 prices. Costs of all adverse events used in the model are shown in Table 24.

N.2.18. Sensitivity analysis

Both deterministic and probabilistic sensitivity analyses were used to characterise the uncertainty surrounding the base case results of the model.

For the deterministic sensitivity analysis, cost per breast cancer case prevented was calculated compared to no chemoprevention for each intervention, under each of the following scenarios:

• Relative risks for breast cancer incidence and adverse events replaced by hazard ratios where available from the clinical literature
• Baseline incidence of breast cancer increased by 100% and reduced by 50%
• Relative risks of treatments for breast cancer incidence changed to lower and upper 95% confidence intervals
• Adherence changed to 100% and 25%
• Baseline incidence of adverse events (endometrial cancer, thromboembolic events, and fractures) increased by 100% and reduced by 50%
• Relative risks of adverse events set to lower and upper 95% confidence intervals
• Cost of treatment increased by 100% and reduced by 50%
• Costs of adverse events increased by 100% and reduced by 50%
• Cost of breast cancer increased by 100% and reduced by 50%
• Relative risks of adverse events persist for 20 years after the end of treatment
• Breast cancer treatment is associated with no chemotherapy costs (this scenario was included to reflect the fact that chemoprevention is only effective in preventing ER-positive breast cancers, which are less susceptible to treatment with chemotherapy)
• Relative risks for breast cancer, endometrial cancer and thromboembolic event incidence for raloxifene changed to those from the RUTH trial comparing raloxifene to placebo in postmenopasal women (Barrett-Connor et al, 2006). This sensitivity analysis was conducted as the committee felt that the evidence comparing raloxifene to tamoxifen may provide an unfavourable representation of the effectiveness of raloxifene in preventing breast cancer. The RUTH trial was selected as a representative study comparing raloxifene to placebo in a relevant patient population.Table 25 shows the relative risks used in this sensitivity analysis.

For the probabilistic sensitivity analysis, all model input parameters were assigned probability distributions (rather than being expressed as point estimates) to reflect the uncertainty surrounding the available clinical and cost data. 1,000 iterations of the model were run, each drawing random values from parameter distributions.

Probability parameters were assigned beta distributions in order to account for the fact that probability values must lie between 0 and 1. Cost parameters were assigned gamma distributions, to ensure that costs could not be negative. Relative risks were assigned a distribution by raising the exponential constant to the power of a normal distribution of the natural logarithm of the parameter.

Where available, standard errors or 95% confidence intervals were used to inform the shape of distributions. For parameters for which these values were not available, it was assumed that standard error was 10% of the parameter mean.

N.3.1. Deterministic results

Base case cost results for a cohort of 1,000 high risk patients are displayed in Table 26. Results show that tamoxifen and raloxifene incur an additional cost of £97,346 and £237,865 per 1,000 patients, respectively, compared to no treatment. This additional cost is incurred primarily from costs of chemoprevention and monitoring consultations with GPs, although it partially offset by a reduction in the cost of breast cancer treatment. Conversely, anastrozole produces a cost saving of £34,539 per 1,000 patients, mostly achieved through a reduction in the cost of breast cancer treatment.

Table 27 displays health outcomes for a cohort of 1,000 patients. All chemoprevention strategies demonstrate a reduction in breast cancer cases, with anastrozole achieving the highest reduction (36 cases prevented). The incidence of adverse events is relatively uniform across all comparators, although the incidence of thromboembolic events is slightly elevated for all chemopreventive agents (in particular for tamoxifen), and the incidence of fractures is slightly decreased by tamoxifen and raloxifene, and slightly increased by anastrozole.

Table 28 displays incremental cost effectiveness results for each chemopreventive agent compared to no treatment, reported in terms of incremental cost per breast cancer case prevented. Raloxifene is associated with a substantially higher cost per case prevented than other chemopreventive agents, due to a high drug acquisition cost and a relatively low efficacy. Anastrozole is associated with a negative ICER (i.e. it dominates no chemoprevention), as it results in lower costs and fewer breast cancer cases than no treatment.

Results are also reported in terms of the QALY gain required per prevented breast cancer case for each treatment to be cost effective at a threshold of £20,000/QALY. Treatment with tamoxifen and raloxifene requires a gain of 0.23 and 1.27 QALYs per breast cancer case prevented, respectively, while the value for anastrozole is negative, as this treatment dominates no chemoprevention.

In order to verify that chemoprevention with these agents is cost effective at a theshold of £20,000, a simple Markov chain was constructed to estimate the QALY difference between a 50 year old woman with breast cancer and a healthy 50 year old woman over the course of five years. Using an annual cycle length, with the mortality and utility inputs listed in Table 29 and a discount rate of 3.5% per year, an estimate of 1.33 incremental QALYs per breast cancer case prevented was produced. Based on this value, it is likely that, in high risk patients, chemprevention with all three agents is cost effective compared to no treatment.

N.3.2. Sensitivity analysis

Results of one-way sensitivity analysis are displayed in Table 30. Results for anastrozole are displayed as a tornado diagram in Figure 3, to illustrate the magnitude of sensitivity across parameters. These results show that the cost effectiveness of treatment is particularly sensitive to changes in the baseline incidence and relative risks of breast cancer. This is because these parameters have a considerable impact on both the number of breast cancer cases prevented and the magnitude of costs, as breast cancer treatment constitutes a large proportion of total costs. For this reason, results are also sensitive to changes in the cost per case of breast cancer.

Comparatively, cost effectiveness results are insensitive to changes in the baseline incidence, relative risks, and costs of adverse events. This is because the majority of adverse events have a low baseline incidence rate, low cost of treatment, or relative risks for chemoprevention that do not deviate far from 1. This also explains why extending the persistence of relative risks associated with adverse events to 20 years after the end of treatment does not considerably change cost effectiveness.

Removing all chemotherapy costs for breast cancer treatment results a moderate increase in incremental costs per breast cancer case prevented. However, due to chemotherapy costs compising a relatively small proportion of total breast cancer treatment costs, these changes are unlikely to be sufficiently large to affect decisions. This demonstrates that, even in the extreme scenario that no patients with ER-postive breast cancer receive chemoprevention, results are robust.

Cost effectiveness results are also relatively sensitive to changes in estimated patient adherence, as a reduction in the number of adherent patients increases drug costs, but infers no benefits in breast cancer cases prevented.

Finally, using relative risks for breast cancer and adverse event rates derived from the RUTH trial results in a considerably lower incremental cost per breast cancer case prevented, indicating that there is significant uncertainty regarding both the effectiveness and the cost effectiveness of raloxifene.

Figure 3Tornado diagram of one way sensitivity analysis results for anastrozole (high-risk patients)

Mean probabilistic sensitivity analysis results are displayed in Table 31. These values show that mean probabilistic results are generally comparable to deterministic results.

Costs and breast cancer cases prevented for each of the 1,000 probabilistic iterations are shown in Figure 4. There is considerable overlap in the results for all chemopreventive agents, but there is a trend towards lower incremental costs and higher number of breast cancer cases for anastrozole. Moreover, anastrozole is associated with a relatively high probability of being cost saving compared to no chemoprevention (59%). Probabilistic results also show that there is an observable negative correlation between number of breast cancer cases prevented and incremental cost. This is because treatment of breast cancer comprises a large proportion of costs in the model, meaning that iterations in which a higher number of breast cancer cases are prevented are also likely to result in higher cost savings.

Figure 4Probabilistic sensitivity analysis results – incremental cost and breast cancer cases prevented per 1,000 high-risk patients

Figure 5 shows probabilistic results displayed as a cost effectiveness acceptability curve, plotting the probability of each comparator being the most cost effective option at a range of willingness to pay thresholds for the prevention of one case of breast cancer. Results show that at all thresholds anastrozole has the highest probability of being the most cost effective treatment. At a threshold of £20,000 per breast cancer case prevented, the probability of anastrozole being the most cost effective treatment is probabilities of anastrozole being the most cost effective treatment is 89%.

Figure 5Probabilistic sensitivity analysis results – cost effectiveness acceptability curve for high-risk patients

N.4.1. Deterministic results

Base case cost and health outcome results for a cohort of 1,000 moderate risk patients are presented in Table 32. Numbers of adverse events were identical to those of the high-risk cohort, so are not tabulated again. Results show that, due to the lower baseline rate of breast cancer incidence, all chemoprevention treatments are associated with a lower number of prevented breast cancer cases, and higher incremental costs than in the high-risk cohort. Anastrozole is also associated with a positive incremental costs compared to no treatment, rather than being cost saving.

However, treatment with anastrozole still results in lower total costs and a higher number of breast cancer cases prevented than tamoxifen and raloxifene. This is reflected in Table 33, which for each chemopreventive agent shows the cost per breast cancer prevented and QALYs required per breast cancer case prevented to be cost effective at a £20,000 threshold. These results indicate that, although chemoprevention is not as cost effective for moderate risk patients, anastrozole is still expected to be cost effective, both compared to no chemoprevention and compared to other chemopreventive agents.

N.4.2. Sensitivity analysis

One-way sensitivity analysis results are displayed in Table 34. Results for anastrozole are displayed as a tornado diagram in Figure 6, to illustrate the magnitude of sensitivity across parameters. As with the high-risk cohort of patients, cost effectiveness results for moderate-risk patients are particularly sensitive to changes in parameters affecting the incidence or cost of breast cancer, and relatively insensitive to changes in parameters affecting incidence or cost of adverse events.

Figure 6Tornado diagram of one-way sensitivity analysis results for anastrozole (moderate-risk patients)

Mean results of the probabilistic sensitivity analysis are shown in Table 35. As with the results of the high-risk patient population, these values show that deterministic and mean PSA results are largely consistent.

Figure 7 shows the results of the 1,000 probabilistic iterations for moderate risk patients plotted on a cost effectiveness plane. These results show a similar overall pattern to those of the high-risk cohort – anastrozole shows a trend towards lower costs and more cases of breast cancer prevented, though these is considerable overlap between results of different treatments. The results differ from those of the high-risk patients in that all treatments are, overall, less cost effective compared to no treatment. For this group of patients, anastrozole has an 18% probability of being cost saving compared to no treatment.

The cost effectiveness acceptability curve for moderate-risk patients (shown in Figure 8) reinforces these findings – at low thresholds no chemoprevention is the most likely option to be cost effective. However, at a relatively low threshold of £2,600 per breast cancer case prevented, anastrozole becomes the treatment with the highest probability of being cost effective.

Figure 7Probabilistic sensitivity analysis results – incremental cost and breast cancer cases prevented per 1,000 moderate-risk patients

Figure 8Probabilistic sensitivity analysis results – cost effectiveness acceptability curve for moderate-risk patients