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

National Collaborating Centre for Cancer (UK). Advanced Breast Cancer: Diagnosis and Treatment. Cardiff (UK): National Collaborating Centre for Cancer (UK); 2009 Feb. (NICE Clinical Guidelines, No. 81.)

Cover of Advanced Breast Cancer

Advanced Breast Cancer: Diagnosis and Treatment.

Show details

Appendix 1A cost-utility analysis of chemotherapy sequences for the treatment of patients with advanced breast cancer


Since metastatic breast cancer is incurable, the quality of patients’ lives during the final stages of life with various forms of active chemotherapy and supportive and palliative care is of great importance. However the economic cost of this treatment and care to the NHS must be considered and balanced.

NICE has previously issued guidance on the use of the taxanes, capecitabine and vinorelbine for use in the treatment of patients with advanced breast cancer in the form of three technology appraisals (TA30 (2001); TA54 (2002); TA62 (2003)). These appraisals are now being updated within the guideline for the treatment of advanced breast cancer. In light of new clinical evidence it is important that the economics of these chemotherapy agents are re-examined. The sequencing of these agents has not been considered in the economic literature to date and the neglect of sequential therapy as a comparator to combination therapies in previous technology appraisals was a concern to both the Appraisal Committee of the recent Gemcitabine STA (TA116) and to the Advanced Breast Cancer Guideline Development Group. As such a de novo economic model has been developed to investigate the cost-utility of chemotherapy sequences for the treatment of patients with advanced breast cancer.

Existing economic evidence

There are a number of good quality economic evaluations investigating the cost-effectiveness of first- and second-line chemotherapy regimes in patients with metastatic breast cancer, most of which were appraised for the original technology appraisals (summarised below). Four new full economic evaluations have been published since the review undertaken for the appraisals (Verma et al. 2005; Cooper et al 2003; Verma & Ilersich, 2003; Li et al. 2001). One partial economic evaluation considering the costs of third-line chemotherapy was published in 1999 but was not included in the previous reviews since third-line therapy was not part of the inclusion criteria. The main limitations of these studies are that none compare more than three types of therapy, nor do they consider more than one line of therapy. This highlights the need for de novo economic modelling to directly answer the review question.

TA30 – Taxanes

In the original appraisal no economic evaluations for the first line treatment1 of breast cancer with a taxane were identified. For second-line treatment2, seven economic evaluations were identified and reviewed. One compared paclitaxel with mitomycin but was submitted in confidence to NICE and therefore was not published in the subsequent HTA report. The other six compared paclitaxel and docetaxel in cost-utility analyses where the range of incremental QALYs gained was £1,990–£2,4313. In addition three analyses compared docetaxel and vinorelbine - one of which was carried out in the UK and yielded a cost-utility ratio for incremental QALYs gained of £14,050. The original guidance did not give any indication as to which taxane was preferred for second-line treatment of breast cancer, despite the evidence showing that docetaxel has a highly favourable cost-effectiveness ratio compared with paclitaxel.

TA54 – Vinorelbine

Evidence at the time of TA54 was scarce. The evidence reviewed for the appraisal showed no clinical benefit of vinorelbine monotherapy over other therapies as first-line treatment. Vinorelbine monotherapy as second-line treatment was slightly less effective than taxane therapy but was much less toxic. For a sub-group of patients (for example those who are elderly) this was considered a useful treatment option and was backed up by economic evidence. None of the RCT data favoured vinorelbine combinations and the case-series data did not provide a robust alternative interpretation. The economics involved in the original appraisal comprised of two literature reviews (one investigated the use of vinorelbine as a single agent and the other investigated vinorelbine in combination with other agents), with no independent modelling. The reviews found no economic evaluations investigating vinorelbine as combination therapy, and identified four economic analyses for vinorelbine monotherapy (Brown et al. 2001; Silberman et al. 1999; Launois et al. 1996; Leung et al. 1999), though one of these was in abstract form and therefore provided little detail. Three of these were fairly well conducted cost-effectiveness or cost-utility analyses, one of which was carried out in a UK setting from an NHS perspective (the remaining three were undertaken in Canada, the USA and France). However they gave conflicting results, “when comparing the cost-effectiveness of vinorelbine, paclitaxel and docetaxel, one economic evaluation reported that vinorelbine was more effective and less costly than taxane therapy, one found vinorelbine to be less effective and less expensive than either of the taxanes and a third evaluation found vinorelbine to be less effective and more expensive than taxane therapy” (Lewis et al. 2002). In addition none of the studies adequately addressed the uncertainty surrounding their results.

TA62 – Capecitabine

The only economic evidence available at the time of the appraisal was one abstract (not reviewed) and the economic model submitted by the manufacturer for both capecitabine monotherapy and in combination with docetaxel. Neither of these models has since been published in a peer-reviewed journal.


This economic evaluation will assess the cost-effectiveness of several sequences of the main chemotherapy regimes (listed below), as well as supportive and palliative care, that are used to treat patients with metastatic breast cancer who have received prior anthracycline therapy.

A secondary objective is to rule out certain strategies (i.e. sequences of therapy) that are likely not to be cost-effective from an NHS perspective.

To facilitate the economic analysis, an indirect treatment comparison will be carried out on RCTs for first-line treatment.


Study population

In contrast to the populations considered in the technology appraisals, the population of interest in this study is patients with metastatic breast cancer who have previously received anthra-cycline treatment which may have been given as adjuvant treatment. Aggressive treatment of early stage breast cancer has led to the presentation of such patients becoming the ‘norm’, and increasingly patients are even presenting with advanced disease that is resistant to or has failed taxane and anthracycline therapy (Jones et al. 2001).

Whilst no explicit distinction is made, it is assumed patients in whom the disease is hormone responsive will receive alternative/additional treatment. The clinical and economic evidence for the management of these patients is explored elsewhere in the guideline.


Six different standard dose chemotherapy regimens (Table A1.1) were compared in the model.

Table A1.1. Standard dosages assumed by the model.

Table A1.1

Standard dosages assumed by the model.

First-line therapy options (T1)

Capecitabine + docetaxel combination therapy (‘T1: CAP + DOC’)

Gemcitabine + docetaxel combination therapy (‘T1: DOC + GEM’)

Paclitaxel monotherapy (‘T1: PAC’)

Docetaxel monotherapy (‘T1: DOC’)

Second-line therapy options (T2)

Capecitabine monotherapy (‘T2: CAP’)

Vinorelbine monotherapy (‘T2: VIN’)

Supportive and palliative care only (‘T2: no chemo’)

Third-line therapy options (T3)

Capecitabine monotherapy (‘T3: CAP’)

Vinorelbine monotherapy (‘T3: VIN’)

Supportive and palliative care only (‘T3: no chemo’)

Structure of the model

A decision tree was constructed in Excel and later rebuilt using TreeAge to represent all the possible consequences resulting from a sequence of treatment, using a model structure adapted from Leung et al. 1999. A total of seventeen different sequences of chemotherapy were considered, as listed in Table A1.2. It was assumed that a chemotherapy agent could not be used twice in the same sequence.

Table A1.2. The seventeen strategies considered in the model.

Table A1.2

The seventeen strategies considered in the model.

The model structure is presented in Figure A1.1 and described in the text.

Figure A1.1. Decision tree (for first 28 branches).

Figure A1.1

Decision tree (for first 28 branches).

The model begins by considering patients with metastatic breast cancer (who have received prior anthracycline therapy). The first decision is which first-line treatment to offer the patient. The decision tree shows explicitly all the possible decisions that could be taken (given the confines of our decision problem) and all the possible consequences resulting from this first decision (again we have limited these). Four first-line treatments are considered. Time is not made explicit in a decision tree model, but we assume the patient receives one cycle of the first-line therapy. At this point, there is a possibility that the patient might die of a toxic death. If the patient dies of a toxic death, that is the end of the possible outcomes associated with the treatment. It has been assumed that a toxic death can only occur after the first cycle of therapy.

If that patient survives the risk of toxic death, they will then receive two more cycles of therapy. This brings the total number of cycles of therapy the patient has received at this point to three. The patient then faces another chance event of experiencing toxicity that will lead to the discontinuation of the current first-line treatment (no chance or decision to be taken here, this necessarily follows on from experiencing major toxicity). At this point we face another decision node, the choice of which second-line treatment to take. There is a time-lag of 1 month between discontinuing first-line therapy and starting on second-line therapy. If the patient didn’t experience toxicity, they will continue on first-line therapy. At this point it is assumed that response can be assessed, so the patient faces a probability of responding to therapy, of having stable disease or not.

For the purposes of the model, response is defined as complete or partial tumour response to the first-line therapy. Responders and stable patients go on to receive additional cycles of treatment, receiving in total the median number of cycles as reported in the RCTs investigating that therapy (in the case of all the interventions in the model, this was six cycles). Non-response is defined as patients who are classified as having progressive disease or their tumour was non-assessable. These patients do not receive further treatment. Regardless of whether the patient has responded to first-line treatment or not, progression is an inevitable outcome. However the time to progression will be different. Once the patient is experiencing progressive disease, they face the probability of dying from progressive disease. Indeed death only results from progressive disease or toxicity; the possibility of death from other causes was not considered to be relevant to the model due to the poor prognosis of these patients. This approach is consistent with other published economic evaluations. If the patient survives, they will continue to second-line treatment.

At this decision node, there may be two or three possible second-line therapies. This is because it has been assumed that if capecitabine has been used as first-line treatment, or a part of a combination therapy given as first-line treatment (for example, capecitabine plus docetaxel), then it cannot be considered as a second-line therapy option. This is the scenario depicted in Figure A1.1 above.

The patient then experiences the same chance events as with first-line treatment (chance of toxic death, chance of experiencing toxicity leading to discontinuation, chance of responding to second-line therapy). Once second-line therapy is discontinued or progression has been reached after completing the full course of second-line treatment, the patient continues onto third-line therapy. In Figure A1.1 this decision has only one possible option thus is not depicted with a decision node. Since both capecitabine and vinorelbine have been used by this point, the only treatment option left for this patient is supportive and palliative care (‘no chemotherapy’). There is only one possible outcome from the ‘no chemotherapy’ option, so this branch terminates. If third-line treatment is a chemotherapy regime, the same chance events as with first-line and second-line treatment may occur (the chance of toxic death, chance of experiencing toxicity leading to discontinuation, chance of responding to second-line therapy).

Clinical evidence

First-line treatment – an indirect treatment comparison (ITC)

An RCT or a meta-analysis of RCTs comparing all the interventions of interest to this analysis is not available. Indeed using conventional techniques this would not be possible due to the different comparisons made by each trial. It is common for new therapies to be introduced into clinical practice before formal treatment comparisons with the current standard approach or other new agents have been planned or carried out.

Using just one arm of one RCT to give us information on each intervention would cause a number of methodological problems. Not only would this not make use of all the available evidence, it would also lose the effect of randomization which is what gives the RCT its gold standard.

In the absence of direct comparative evidence, an indirect treatment comparison has been performed to inform the parameters of the economic model and ultimately ensure the recommendations in the guideline are based on all available evidence. Indirect comparisons use evidence from A vs. B and A vs. C trials to draw conclusions about the effect of B relative to C. The main assumption made using this approach to evidence synthesis is that the evidence is consistent. That is, the treatment effect of B relative to C estimated by a real trial comparing B vs. C would be the same as the treatment effect estimated by the A vs. B and A vs. C trials if they had included C and B arms respectively. This assumption is also implicit in cost-effectiveness analysis, since evidence is routinely combined from a variety of sources, thus consistency has to be assumed.

The clinical evidence review for the update of each technology appraisal was performed separately, which informed the search strategy for these topics. As such a full systematic search for all treatments for metastatic breast cancer was not undertaken. The network of RCT evidence is thus made up of trials that were identified for the original appraisals, from the individual update searches for the three technology appraisals and from an unsystematic manual search aiming to identify trials that may have been excluded from the clinical review (due to stricter inclusion criteria). Randomised controlled trials that involved one or more of the interventions of interest were included in the network of evidence. Whilst the economic model assesses three lines of therapy, no RCTs were identified for second- or third-line therapy. Thus, the indirect treatment comparison was only carried out on first-line treatment options.

The indirect comparison was undertaken using two separate statistical models using the statistical computer software, WinBUGS. The first describes the relationship between toxic deaths and discontinuation due to toxicity, whilst the second links the response rate, progression rates and mortality. The networks of RCT evidence for each statistical model are depicted below (Figures A1.2 and A1.3); each line represents one RCT and the shading of certain interventions highlights those that are of interest in our decision problem. Other interventions are included to add to the information we can obtain on the interventions that are of interest, through indirect comparisons. The evidence structure is presented in Table A1.3. If all the trials reported all the data that was needed, all trials would have been included in both the indirect treatment comparisons. Since there were gaps in the data, three of the trials (Sjostrom 1998, Bonneterre 2002 and Monnier 1998) were excluded from the analysis of progression and survival. Whilst the analysis was undertaken from a Bayesian framework, flat priors were used in both statistical models and thus did not impact on the results.

Figure A1.2. RCT network for toxicity model.

Figure A1.2

RCT network for toxicity model. DOC = docetaxel; GEM = gemcitabine; CAP= capecitabine; PAC = paclitaxel; MV= mitomycin plus vinblastine; MtF= methotrexate and 5-fluorouracil; FUN = 5-fluorouracil + vinorelbine

Figure A1.3. RCT network for survival model.

Figure A1.3

RCT network for survival model. DOC = docetaxel; GEM = gemcitabine; CAP= capecitabine; PAC = paclitaxel; MV= mitomycin plus vinblastine;

Table A1.3. Evidence structure.

Table A1.3

Evidence structure.

The text that follows describes of the methods used for the indirect treatment comparisons. The WinBUGS code is not presented here but is available from the author on request (please contact

Toxicity model

A number of assumptions were made in order to get the most out of the data. Firstly, it was assumed that the toxic death rate did not vary by study, so a fixed effects model was used. Secondly, the two measures of toxicity (toxic death and discontinuation due to toxicity) are related by a constant, beta, which was allowed to vary by study (from a random effects model). Thirdly the baseline probability of toxic death (to which all the relative effects are compared, in this case the probability of toxic death for docetaxel) was estimated by a random effects model of the arms of the three trials involving docetaxel.

Survival model

In line with the assumptions made in structuring the economic model, it is assumed that patients are categorised at 9 weeks as responders (r), stable (s), with progressive disease (pd), or non-assessable (na). There is data on the split between these groups from most studies, although one study only reports whether a responder, stable or not, and one study only reports whether a responder or not. It was assumed that the split between categories follow a multinomial distribution:


We model the effect of treatment (i.e. probabilities, represented by p, of tumour response, stabilisation or non-response) using multinomial logistic regression. Let


We assume the following model for the conditional probabilities, q:

logit(qi,1)=ϕs(i)+(θt(i),1-θb(i),1)logit(qi,2)=ϕs(i)+ζs(i)+(θt(i),2-θb(i),2);         ζj~N(mζ,sdζ2)logit(qi,3)=ϕs(i)+γs(i)+(θt(i),3-θb(i),3);         γj~N(mγ,sdγ2)

Key assumptions:

  • fixed treatment effects which differ for different conditional outcomes: responders; stable|non-responder; and prog.disease|{non-responder & non-stable}.
  • the proportion of responders depends on study.
  • the baseline log-odds of the conditional outcomes stable|non-responder; and prog.disease| {non-responder & non-stable} differ from that for responders by study specific terms ζj and γj which come from random effects distributions.

Most studies reported median time to progression for responders and for all. We assume exponential distributions for the time to progression in responders and non-responders with rates λr and λnr respectively. We therefore needed a model for the progression rate for responders, λr, and non-responders, λnr. We put a log-linear model on the progression rate in responders and stable:

log(λr)=αs(i)+(dt(i)-db(i))log(λs)=αs(i)+ηs(i)+(dt(i)-db(i));         ηj~N(mn,sdη2)

Key assumptions:

  • study specific baselines for responders
  • random effects model for log-hazard ratio for stable vs responder
  • fixed treatment effect across studies, which is the same for responders and stable individuals.

The mean progression time in non-responders is a weighted average of mean progression time for stable, non-assessable, and progressive disease patients, giving progression rate in non-responders of:


Key assumptions:

  • since we did not know when progressors progressed, the time to progression for those with progressive disease is assumed to be 4.5 weeks, or 1.125 months. This is the midpoint between zero weeks and nine weeks, at which point tumour response is usually assessed.
  • non-assessable patients have the same progression rate as progressive patients.

Most studies reported median time to mortality for all patients. If we assume a constant term linking progression rates with mortality rates, then we can model mortality in exactly the same way as for progression. However, we do not know the mortality rate (1/3) for those with progressive disease, and so this was estimated from the data.

We assume exponential distributions for the time to mortality in responders and non-responders with rates 3r and 3nr respectively. We therefore need a model for the mortality rate for responders, 3r, and non-responders, 3nr. We put a log-linear model on the mortality rate in responders and stable, which differ from log progression rates by a constant (β), which depends on study, but assumed to come from a random effects distribution:

log(μr)=log(λr)+βs(i);         βj~N(mβ,sdβ2)log(μs)=log(λs)+βs(i)

The mean survival time in non-responders is a weighted average of mean survival time for stable, non-assessable, and progressive disease patients, giving mortality rate in non-responders of:


Key assumptions:

  • random effects model on the log-hazard ratio’s (βj) of mortality relative to progression
  • fixed mean survival time κ for those with progressive disease.
  • non-assessable patients have the same mortality rate as progressive patients.

Assessing model fit

The models described above were the result of a systematic model fitting process. We measured model fit using the posterior mean residual deviance, which we expect to be roughly equal to the number of unconstrained data points for a good fitting model. The posterior mean residual deviance for the toxicity model was 24.4 compared to 28 data points showing adequate model fit. The survival model had a posterior mean deviance of 58.1 compared with 48 data points, indicating some lack-of-fit to the survival and time to progression data. It should be noted, however, that the possible models that we could fit was limited by the data available, and assumptions had to be made. For example, studies only reported median time to progression or median survival time. With a single reported summary measure it is only possible to estimate a single model parameter. This meant we were restricted to Exponential rather than Weibull distributions for progression and survival times, with no possibility of checking this assumption. Additionally, time to progression was only reported for responders or all patients, and survival reported for all patients only without breakdown between patient groups.

Second-line treatment

There is one randomised controlled trial and seven non-randomised studies investigating second-line therapy. No evidence was found to report the effectiveness of the ‘No chemotherapy’ intervention.

The Martin et al. (2007) RCT was used to provide data on vinorelbine monotherapy as second-line treatment by agreement with the GDG since the trial has a mixed patient population (patients received vinorelbine as first-, second- and third-line treatment). Although there were two other observational studies investigating vinorelbine monotherapy (Zelek 2001; Udom 2000) they were both small trials and the Martin et al. (2007) was considered by the GDG to provide the best estimate of vinorelbine monotherapy in the second-line setting.

Five non-randomised studies were identified for capecitabine monotherapy as second-line treatment (Fumoleau et al. 2004; Lee et al. 2004; Pierga et al. 2004; Reichardt et al. 2003; Wist et al. 2004). Whilst all were considered acceptable in terms of being able to provide reasonably robust evidence, not all trials provided data on the same parameters. Pierga et al. 2004 provided data on response duration, duration of stable disease and time to progression for all. As such this trial was used to provide information for the model on capecitabine monotherapy as second-line treatment.

No evidence was found to report the effectiveness of the ‘no chemotherapy’ intervention. It was assumed that ‘no chemotherapy’ would result in no progression-free survival and 5 months survival with progressive disease.

Third-line treatment

No evidence for capecitabine or vinorelbine monotherapy as third-line treatment was identified. It was therefore assumed that the same data for second-line treatment would provide a suitable estimate of third-line treatment, since the patient populations included some patients receiving the study therapy as third-line. In the base-case analysis, no adjustments to the data were made although the effect of reducing the survival estimates by varying degrees will be explored in the sensitivity analysis.

Health benefits


The probabilities of toxic death and of discontinuing treatment due to toxicity shown in Table A1.4 were all estimated via the ITC statistical model. The toxicity data for second and third-line treatment are shown in Table A1.5.

Table A1.4. Probabilities estimated by the indirect treatment comparison.

Table A1.4

Probabilities estimated by the indirect treatment comparison.

Table A1.5. Probabilities for second- and third-line treatment.

Table A1.5

Probabilities for second- and third-line treatment.

The probabilities of response, stabilisation of disease, disease progression and non-assessability were estimated via the second ITC statistical model, shown in Table A1.6. These data for second- and third-line treatment are shown in Table A1.7.

Table A1.6. Probabilities estimated by the indirect treatment comparison.

Table A1.6

Probabilities estimated by the indirect treatment comparison.

Table A1.7. Probabilities for second- and third-line treatment.

Table A1.7

Probabilities for second- and third-line treatment.

For the economic model, it was assumed that non-assessable patients were the same as patients with progressive disease.


Overall survival (OS) was assumed to be the sum of time to progression (TTPt1) of first-line treatment, TTP from second-line treatment (TTPt2), TTP from third-line treatment (TTPt3) and the period from progression to death (assumed to be 5 months). This assumption implies that chemotherapy impacts on time to progression, and through that overall survival. However the time from (final) progression to death is fixed regardless of prior treatment.

Mean ‘progression-free’ survival times (in months) were estimated from the statistical model on survival and are reported below in Table A1.8. It is assumed that time to progression for patients with progressive disease reported as their best response to treatment (or if the tumour was not assessable) is 1.125 months (4.5 weeks).

Table A1.8. Survival data estimated by the indirect treatment comparison (in months).

Table A1.8

Survival data estimated by the indirect treatment comparison (in months).

Mean values are used for the economic evaluation since they are a more appropriate measure of the average at a population level. Since only median values were reported in the Martin et al. 2007 and Pierga et al 2004 trials, it was assumed that survival and time to progression followed exponential distributions. Median values were then converted to mean values by calculating the baseline hazard and are reported below in Table A1.9.

Table A1.9. Survival data for second- and third-line treatment (in months).

Table A1.9

Survival data for second- and third-line treatment (in months).


where, h=baseline absolute hazard; tmed=mediansurvival time; tmean=mean survival time


Utility weights were linked to the time spent at different points of the pathway (not strictly health states since we did not use a Markov process) to calculate QALYs. No trials reported utility losses due to toxicity or to progressive disease, so the proportion of patients in each arm of an RCT that progressed or discontinued treatment due to toxicity were relevant published utility weights to estimate the overall utility. There are a number of studies that report utility weights in the treatment of advanced breast cancer. The most recent pooling of utilities from different sources (all derived from oncology nurses using the Standard Gamble technique) was published by Cooper et al. (2003) and is shown in Table A1.10. A number of assumptions had to be made about the utility associated with time spent between treatment (we assume utility with progressive disease, 0.45); the time spent on treatment before response could be assessed (we assume utility associated with stable disease, 0.65, to ensure consistency with the indirect treatment comparison since at this stage by definition the disease is not yet progressive); and time before toxicities identified after 3 cycles of treatment (we assume utility associated with progressive disease, 0.45).

Table A1.10. Utility values from Cooper et al. (2003).

Table A1.10

Utility values from Cooper et al. (2003).

Cost estimation

The costs considered in this analysis are only those relevant to the UK NHS, in accordance with the perspective taken by the NICE Reference Case for economic evaluations. Costs were estimated in 2006–07 prices. Where costs have been taken from sources using a different price year, they have been inflated using the Hospital and Community Health Services Pay and Prices Index (PSSRU, 2007).

There are broadly five categories of costs considered in the model:

  • cost of treatment
  • cost of assessment/follow-up
  • cost of treating adverse events
  • cost of supportive and palliative care
  • costs associated with death.

Cost of treatment

The average dose for each regime was presented in Table A1.1. The possibility of reducing the dose (in response to an adverse event) was not allowed for in the model. The drug acquisition cost per cycle were calculated for each chemotherapy regime based on an average dose per patient (standard 1.75m2), the average number of doses per cycle and the average list price per mg, and are shown in Tables A1.11 and A1.12. Whilst it is recognised that discounts are available on some of these drugs, the list price was used in the base case as recommended in the NICE Reference Case. The effect of these drug discounts will be explored in the sensitivity analysis. Where the price is given for both the generic and proprietary drug, the cheapest is used in the base-case.

Table A1.11. Drug acquisition costs (1).

Table A1.11

Drug acquisition costs (1).

Table A1.12. Drug acquisition costs (2).

Table A1.12

Drug acquisition costs (2).

In addition to the drug acquisition costs, the cost of administering the drug was estimated from the NHS National Reference Costs. For therapies administered by i.v. or injection (gemcitabine), the cost used was £293 for outpatient delivery of complex perenteral chemotherapy and subsequent elements. This cost includes hospital overheads, the administration costs of chemotherapy and clinical time, but does not, for example, distinguish between different i.v. infusion times of paclitaxel vs. docetaxel. For drugs administered orally (capecitabine) the administration costs were estimated using the outpatient tariff, £179 per attendance. It has been assumed that one outpatient appointment would be required per cycle of therapy (one every three weeks). In the case of combination therapy it has been assumed that two drugs can be administered at one time, thus requiring the cost of only one administration to be considered. In addition to the drug acquisition and drug administration costs, it has been assumed that a consultation with an oncologist (£179, National Reference Costs 2006–07) would be necessary at the starting cycle.

Cost of assessment/follow-up

The cost of taking one CT scan (2 areas, with contrast) every three cycles of treatment (£96) in addition to a consultant-led attendance was used as a proxy for the cost of assessing response (NHS Reference Costs, 2006–07). This is an attempt to capture the continuous nature of assessing response.

Once the patient has finished chemotherapy and achieves a response there will still be a cost associated with the contact the patient receives from their consultant. (The cost of contact with other health professionals is included in supportive care package 1 below). The cost of one consultation with specialist every 2 months after treatment has finished (£105 per month, NHS Reference costs 2006–07) is used as a proxy for follow-up costs.

Response is not assessed when first-line chemotherapy ends so the cost of an assessment is included before the patient begins the next line of chemotherapy.

Cost of treating adverse events

The cost of treating major toxicities (which necessarily lead to the discontinuation of treatment) was estimated as £1233, a weighted average of two costs from the literature (95% treated in hospital: 5% treated at home); from the cost of treating severe infection or febrile neutropenia in hospital £1,281 (Cooper et al. 2003) and the cost of treating a severe infection or febrile neutropenia at home £328 (Cooper et al. 2003), both reported here already inflated to 2006–07 prices. This cost was used across all treatments, so was not specific to the type of toxicity that leads to discontinuation which we know is likely to vary by therapy.

Cost of supportive and palliative care

Due to the nature of supportive and palliative, three likely ‘packages’ of care4 are described below for patients at different points along the care pathway.

Package 1

The first package of care describes an average level of supportive care a patient receiving chemotherapy might be expected to receive from the time of first cycle of treatment until the onset of progressive disease, at which point the next line of chemotherapy is started. Given the model structure, this package of care is given to a patient until they begin the ‘no chemotherapy’ option. For some strategies this package of care will be given for the whole time spent in the model.

Time-related elements

Community nurse: home visit 20 minutes, £24.00, 1 per fortnight (PSSRU, 2007)

GP contact: 1 surgery visit £34.00 every month (PSSRU, 2007)

Clinical nurse specialist: 1hr contact time, £74.00, 1 per month (PSSRU, 2007)

Time non-related elements

Social worker: 1hr client-related work but not direct contact time, £34.00 (PSSRU, 2007)

Package 2

The second package of care describes an average level of supportive and palliative care a patient receiving the ‘no chemotherapy’ intervention might be expected to receive until the last two weeks of life. This package of care is also included for the patient that follows the strategies in the model with three lines of chemotherapy, from the time of progression until the two weeks before death. Unlike the care given in package 1, all elements of the care delivered in package 2 are time-related.

Time-related elements

Community nurse: home visit 20 minutes, £24, 1 per week (PSSRU, 2007)

Clinical nurse specialist: 1hr contact time, £74, 1 per week (PSSRU, 2007)

GP contact: 1 home visit, £55, every fortnight (PSSRU, 2007)

Therapist5: 1 hour, £40, every fortnight (PSSRU, 2007)

Package 3

The third package of supportive and palliative care is a cost for the more intensive needs of patients in the final two weeks of life. If this cost was attributable to all patients dying in the model, it would be superfluous to the analysis since we are interested solely in incremental costs and incremental benefits. This package of care is not however given to patients who die in the model from toxic death. Since the toxic death varies (albeit not greatly) between the interventions compared in the model, the cost of package 3 supportive and palliative care does need to be taken into account.

The cost used was a weighted average of the three costs reported in the Marie Curie commissioned report into the cost of dying at home (inflated as previously described to 2006–07 prices):

  • last 14 days - in hospital, £4,706
  • last 14 days - in Marie Curie hospice, £5,867
  • last 14 days - at home (with community support), £2,428.

The weights applied to calculate this average were 40% deaths occurring in hospital, 10% occurring in a hospice and the remaining 50% of deaths occurring at home. The cost of the last two weeks of care was therefore estimated to be £3,418.

Costs associated with death

Apart from package 3 of supportive and palliative care, the other cost associated with death included in the model is the cost of toxic death. No costs related to toxic deaths were reported explicitly for any of the published economic evaluations, despite all papers considering the risk of toxic death. A proxy was used by way of the mean of two costs from the literature; from the cost of 7 days hospitalisation and treatment of severe febrile neutropenia £3,586 (Brown et al. 2001) and the cost of treating a severe infection in hospital £988 (Cooper et al. 2003), both reported here already inflated to 2006–07 prices. In total the cost of toxic death used in the model is £2,287.


Discounting was not conducted, so the results that follow are the undiscounted costs and health outcomes. However we would not expect discounting to have much impact on the results of the model since many of the possible pathways through the model are associated with survival of less than 24 months. In addition the majority of the costs for pathways that do result in a longer survival, come at the beginning rather than spread evenly across the year.

Type of analysis

A cost-utility analysis was performed given that the health outcome preferred by NICE is the QALY and quality of life is of particular importance to patients with metastatic cancer. An incremental cost-effectiveness analysis was conducted after ranking the alternative strategies from the most to the least cost-effective and excluding any dominated strategies (i.e. those strategies achieving lower effectiveness and incurring higher costs when compared to any other, or those which are ruled out if they achieve lower effectiveness and higher costs than a combination of two other strategies).

Sensitivity analysis

Two approaches to testing the robustness of the model results were taken; a series of one-way deterministic sensitivity analyses and a probabilistic sensitivity analysis on just two of the strategies.

‘One-way sensitivity analysis’ describes the process of changing one parameter in the model and analysing the results of the model analysed to see if this parameter influences any of the overall results.

Three sources of uncertainty were investigated using one-way deterministic analysis; the data used on the effectiveness of capecitabine monotherapy, the effectiveness of third-line therapy and possible price discounts.

Effectiveness of capecitabine monotherapy

It was noted that the time to progression associated with capecitabine monotherapy was high. Therefore these estimates were reduced by one third in this scenario.

Effectiveness of third-line treatment

No evidence was available for the effectiveness of third-line therapy, so both capecitabine and vinorelbine monotherapies were assumed to work as well as for second-line therapy. This was justified by the fact that the data used to inform the second-line therapy parameters in the model came from trials with mixed patient populations which included patients who were receiving the study therapy as third-line. The effect of reducing the response and disease stabilisation rates by one third, and separately reducing the time to progression estimates by one third was investigated.

Price discounts

Price discounts are available across England and Wales on paclitaxel and vinorelbine since generic versions are available. However there is not one single agreed price discount available for either agent that is applicable across the whole of England and Wales. Therefore a number of different price discounts for paclitaxel were investigated (50%, 60%, 70%, 80%, 90%).

The major limitation of one-way sensitivity analysis is that we are not just uncertain about one parameter (for example, the utility ascribed to progressive disease) – we are uncertain about many parameters (for example, utility values, cost estimates, response rates) at any one time, and so we need to estimate the joint impact of altering all of these. The method used to do this is known as probabilistic sensitivity analysis (PSA).

Firstly, the stochastic parameters in the model were identified (presented in the first column of Table A1.13). These are parameters which are (arguably) measureable, but are associated with sampling uncertainty. Secondly, these parameters were specified as distributions rather than point estimates (see fourth column of Table A1.13). Where the indirect treatment comparison models were conducted the uncertainty surrounding these parameters were defined directly from random values recorded for each of the 10,000 iterations performed in WinBUGS. In order to maintain the correlation between the posterior estimates for probability of tumour response and time to progression and between the probabilities of toxic death or discontinuation of treatment due to toxicity, data from each of the ITC simulations for these parameters were exported jointly and fitted into the TreeAge model where the probabilistic analysis was carried out. In the other cases where the parameters were not part of the indirect comparison model, a distribution was selected according to a well developed body of methodological literature. The data required to inform these distributions was taken from the same sources as was used for the point estimates.

Table A1.13. Parameters varied in the proababilistic sensitivity analysis.

Table A1.13

Parameters varied in the proababilistic sensitivity analysis.

Parameters not chosen for PSA

  • unit costs of health professionals
  • assumptions for proxy costs for example, for response assessment and resource use inputs for supportive and palliative care packages
  • any structural assumptions for example, number of cycles received before undergoing response assessment, time lag between finishing one line of treatment and starting the next ‘active’ treatment.
  • any methodological assumptions for example, drug acquisition cost.

Thirdly, the analysis was run 10,000 times. For each simulation, different values will be picked from the various distributions for each stochastic parameter in the model.


Base-case results (from the deterministic analysis)

The base-case results are shown listed by strategy, in Table A1.14. There is a considerable difference between the strategies in terms of survival, quality of life and associated costs. The overall survival from each strategy ranges from just over 23 months (strategy 5: GEM+DOC, CAP, VIN) to just over 8 months (strategy 12: PAC, No Chemo). Strategy 3 yields the highest number of QALYs (1.2) compared to 0.36 for strategy 12. Total costs for each strategy ranged from £13,500 (strategy 12) to over double that for strategy 3, £30,300.

Table A1.14. Base-case results, by strategy.

Table A1.14

Base-case results, by strategy.

Incremental cost-effectiveness analysis (from the deterministic analysis)

Using QALYs as the outcome measure, an incremental cost-effectiveness analysis was performed by first ranking the strategies according to the cost per patient (highest to lowest). This allowed the dominated strategies to be identified and ruled out of the incremental analysis. Any strategies achieving fewer QALYs and incurring higher costs when compared to any other are ruled out by simple dominance and any stategies that achieve fewer QALYs and higher costs than a combination of two other strategies are ruled out via extended dominance. This left five remaining strategies (3, 9, 12, 13 and 14) which are labelled in Figure A1.4.

Figure A1.4. Cost-effectiveness plane.

Figure A1.4

Cost-effectiveness plane.

The incremental cost-effectiveness ratios (ICERs) shown in the last column of Table A1.15 are the ratios of cost and health benefit for each strategy compared to the next best strategy. NICE recommends the use of a threshold of £20,000 per QALY. Using a threshold value of £20,000 per QALY, strategy 14 (docetaxel followed by capecitabine followed by no chemotherapy) was shown to be most cost-effective since it maximises health benefits given the budget constraint. However there may be compelling reasons to accept a slightly higher ICER of up to £30,000 per QALY which would make strategy 13 (docetaxel followed by capecitabine and then vinorelbine) most cost-effective since it maximises QALYs below this threshold. Due to the multitude of strategies in the analysis, the results need careful interpretation. Since there is very little difference between strategies 15 (docetaxel followed by vinorelbine followed by capectaibine) and 13, in terms of QALYs, and given the uncertainty surrounding these point estimates, there may be some ambiguity over which strategy is dominated and thus which should be excluded from the incremental analysis.

Table A1.15. Incremental results.

Table A1.15

Incremental results.

Strategies 9, 12 and 14 would be ruled out since more QALYs can be achieved given the maximum willingness to pay. Similarly strategy 2 would be ruled out since it’s ICER of £62,300 is far above the maximum threshold NICE recommends; the additional 0.1165 QALYs are judged to not be worth the extra £7,258.

Sensitivity analysis

Three sources of uncertainty surrounding the analysis were investigated using one-way sensitivity analysis; the data used on the effectiveness of capecitabine monotherapy, the effectiveness of third-line therapy and possible price discounts.

Effectiveness of capecitabine monotherapy

The time spent without progressive disease having received capecitabine monotherapy was reduced by one third in the sensitivity analysis. Using threshold values of £20,000 and £30,000, strategy 14 or strategy 13 were still most cost-effective, respectively, maximising QALYs given the threshold.

Effectiveness of third-line treatment

Two ‘effectiveness’ parameters for third-line treatment were varied in the sensitivity analysis; the response and disease stabilisation, and the time spent free of progressive disease for responders, stable patients and non-responders. Both parameters were separately tested, reducing them by one third. When the response and stabilisation rates were reduced there was no change to the strategies that were dominated, or to the ranking of strategies.

Price discounts

A number of different price discounts for paclitaxel were investigated (50%, 60%, 70%, 80%, and 90%) and, as expected, changed the base-case results. Paclitaxel replaced docetaxel as the most cost-effective starting therapy, but after this the preferred sequences in terms of cost-effectiveness (at a £20,000 or £30,000 threshold) did not change from the base case.

The results from the analysis using a 90% discount on paclitaxel are shown below in Table A1.16.

Table A1.16. Impact of 90% discount on paclitaxel.

Table A1.16

Impact of 90% discount on paclitaxel.

Overall the one-way sensitivity analyses showed that the results of the base case were reasonably robust to the parameters investigated. The main changes resulted from big potential price discounts, substituting docetaxel for paclitaxel as the preferred starting therapy.

A probabilistic sensitivity analysis was carried out during the consultation process to investigate the sampling uncertainty in the model and the impact this may have on the decision, and in particular to shed light on the uncertainty surrounding strategies 13 and 15.

The probabilistic analysis demonstrated similar expected values (associated with the means from the 10,000 simulations) compared to the point-estimates used in the deterministic analysis.

It also showed that at a threshold of £20,000 per QALY, the strategy with the highest probability of being the most cost-effective option was strategy 14, (docetaxel followed by capecitabine followed by no chemotherapy), at 45%. The probabilities of strategies 13 (docetaxel followed by capecitabine followed by vinorelbine) and 15 (docetaxel followed by vinorelbine followed by capecitabine) being cost-effective are 14% and 12% respectively. At £30,000, strategy 14 has a 34% probability of being the most cost-effective, strategy 13, 28% and strategy 15, 24%.

However the probability of being cost-effective is not the sole criteria on which a treatment decision should be based, indeed the strategy which maximises net monetary benefit is the optimal choice (at least in terms of cost-effectiveness). Figure A1.5 shows the net monetary benefit for the top three strategies (13, 14 and 15) between threshold values of £20,000 and £30,000.

Figure A1.5. Net monetary benefit at different willingness to pay thresholds.

Figure A1.5

Net monetary benefit at different willingness to pay thresholds.

At a willingness to pay of around £26,500, strategy 13 emerges as the optimal strategy, and is almost indistinguishable from strategy 15.

Table A1.17 below shows the strategies which maximise net benefit across different willingness to pay thresholds (£20,000–£30,000 per QALY), and the probability that these each is the most cost-effective option.

Table A1.17. Probability that the strategy that maximises net benefit is the most cost-effective strategy, at different threshold values.

Table A1.17

Probability that the strategy that maximises net benefit is the most cost-effective strategy, at different threshold values.


The base-case results of this analysis provide a clear message for recommendations on this topic, in terms of cost-effectiveness. They show that docetaxel as a single agent therapy dominates the other taxane, paclitaxel, and any combination therapy involving gemcitabine, so all strategies but those starting with first-line docetaxel are ruled out in terms of cost-effectiveness.

Using the threshold of £20,000, the most cost-effective strategy was docetaxel followed by capecitabine and then no further treatment (strategy 14). The GDG may consider there to be circumstances which justify the use of a higher threshold by which to judge cost-effectiveness and thereby accept strategy 13 which allows for a third line of treatment, vinorelbine. This strategy is associated with higher quality-adjusted survival than the two-line treatment strategy (14).

Due to the multitude of strategies in the analysis, the results need careful interpretation. There is one strategy, strategy 15 (docetaxel followed by vinorelbine then capecitabine) that is narrowly excluded from the incremental analysis on the basis of extended dominance, but only by a tiny difference in total QALYs, 0.0036. Given the uncertainty surrounding these point estimates, it is not clear which strategy is dominated and thus which should be excluded from the incremental analysis. If strategy 13 was dominated, leaving strategy 15 in the incremental analysis, strategy 15 would be associated with a favourable ICER of below £30,000 per QALY. On these grounds the analysis does not provide clear evidence on whether it is always preferable to give capecitabine as second line followed by vinorelbine.

At a threshold of £30,000 per QALY, strategies 3, 9, 12 and 14 can be ruled out in terms of cost-effectiveness since more QALYs can be achieved given the maximum willingness to pay. Similarly strategy 3 would be ruled out since the ICERs of £62,300 is far above the maximum threshold NICE recommends; the additional 0.1165 QALYs are judged to not be worth the extra £7,258.

The sensitivity analysis shows there may however be circumstances in which the base-case results do not hold true. The presence of substantial discounts available nationally for paclitaxel show that if this discount is maintained and is available across England and Wales, the taxane of choice would be paclitaxel rather than docetaxel, since these strategies yielded more favourable ratios of costs and health benefits. In response to doubts over the validity of the utility value for progressive disease, a 10% increase in this value was tested and it was found that the results were not sensitive to this increase.

There are a number of limitations to this analysis. No discounting was undertaken on either the costs or benefits attributed to each strategy. However this is unlikely to have a major bearing on the results since the patients live for a short time and treatment is the biggest contributor to costs which fall at the beginning rather than throughout the year. The sensitivity analyses conducted did not investigate some of the strong structural assumptions made in the model and therefore their impact on the conclusions of the analysis is unknown. The interventions considered in the model were not exhaustive and whilst the most common sequences were included, there may be relevant comparators that have been excluded from the analysis.

Whilst a great deal of effort has been spent on obtaining consistent data on first-line treatment, by undertaking an indirect treatment comparison, many strong assumptions had to be made to combine evidence from different sources to inform the model on the relative effect of the full treatment sequences. Evidence on second-line treatment was poor, and even poorer for third-line treatment. The survival estimates from capecitabine monotherapy seem very high, higher even than first line treatment; although the results seem to be robust to a reduction in these by a third in the sensitivity analysis. No evidence existed for the ‘no chemotherapy’ option, in particular this was not associated with any quality of life increase from the published utility values for progressive disease. Expert opinion from the Guideline Development Group was used to fill in gaps in the data, but this has not been fully explored in the sensitivity analysis and some concerns remain as to the validity of the assumptions.

The costs used were often proxies for costs that were hard to capture and may not fully capture the differences between the different therapies, for instance the differences in i.v. times were not captured by costs (or utilities). It was also assumed that combination therapy was not associated with additional administration times, thus biasing the results in favour of the combination therapies. In addition no vial sharing was assumed, which may not reflect clinical practice.

Despite these acknowledged limitations, this analysis does provide some useful information for which the guideline development group can use in its deliberations over the recommendations to be made on this topic. Single agent taxane (either docetaxel or paclitaxel depending on the price discounts available) is the most cost-effective starting therapy. The combination therapies are much less cost-effective primarily due to the fact repetition of a chemotherapy agent later in the sequence was not allowed in this analytical model. Three lines of chemotherapy were shown to deliver more QALYs than one or two lines. The choice of which order to deliver capecitabine and vinorelbine is not as clear cut, and although the results show capecitabine to be a more cost-effective second line treatment than vinorelbine, the difference between the two strategies (13 and 15) is so small, the Guideline Development Group should interpret this particular result with caution.


  1. Albain KS, Nag S, Calderillo-Ruiz G, Jordaan JP, Llombart A, et al. Global phase III study of gemcitabine plus paclitaxel (GT) vs. paclitaxel (T) as frontline therapy for metastatic breast cancer (MBC): first report of overall survival. J Clin Oncol. 2004;22(14S):510.
  2. Bonneterre J, Roch H‚, Monnier A, Guastalla JP, Namer M, et al. Docetaxel vs 5-fluorouracil plus vinorelbine in metastatic breast cancer after anthracycline therapy failure. British Journal of Cancer. 2002;87(11):1210–1215. [PMC free article: PMC2408916] [PubMed: 12439707]
  3. Brown RE, Hutton J. Cost-utility model comparing docetaxel and paclitaxel in advanced breast cancer patients (Structured abstract) Anti Cancer Drugs. 1998;9(10):899–907. [PubMed: 9890701]
  4. Brown RE, Hutton J, Burrell A. Cost effectiveness of treatment options in advanced breast cancer in the UK. [Review] [27 refs] Pharmacoeconomics. 2001;19(11):1091–1102. [PubMed: 11735676]
  5. Chan S. Gemcitabine plus docetaxel versus capecitabine plus docetaxel for patients with anthracycline-pretreated metastatic breast cancer: A review of the results of a European Phase III trial. European Journal of Cancer Supplement. 2005;3(5):17–21.
  6. Cooper NJ, Abrams KR, Sutton AJ, Turner D, Lambert PC. A Bayesian approach to Markov modelling in cost-effectiveness analyses: application to taxane use in advanced breast cancer. Journal of the Royal Statistical Society Series A-Statistics in Society. 2003;166:389–405.
  7. Coyle D, Small N, Ashworth A, Hennessy S, Jenkins-Clarke S, et al. Costs of palliative care in the community, in hospitals and in hospices in the UK. Crit Rev Oncol Hematol. 1999;32:71–85. [PubMed: 10612007]
  8. Dieras V, Marty M, Tubiana N, Corette L, Morvan F, et al. Phase II randomized study of paclitaxel versus mitomycin in advanced breast cancer. Semin Oncol. 1995;22(4 Suppl 8):33–39. [PubMed: 7638640]
  9. Frasci G, Comella R, D’Auito G, Thomas R, Capasso I, et al. Weekly docetaxel plus gemcitabine or vinorelbine in refractory advanced breast cancer patients: a parallel dose-finding study. Annals of Oncology. 2000;11(3):367–371. [PubMed: 10811508]
  10. Jones SE, Erban J, Overmoyer B, Budd GT, Hutchins L, et al. Randomized phase III study of docetaxel compared with paclitaxel in metastatic breast cancer. Journal of Clinical Oncology: Official Journal of the American Society of Clinical Oncology. 2005;23(24):5542–5551. [PubMed: 16110015]
  11. Jones L, Hawkins N, Westwood M, Wright K, Richardson G, et al. Systematic review of the clinical effectiveness and cost-effectiveness of capecitabine (Xeloda) for locally advanced and/or metastatic breast cancer. Health Technology Assessment (Winchester, England) 2001;8(5) [PubMed: 14960257]
  12. Launois R. Economic assessment, a field between clinical research and observational studies. Bulletin du Cancer. 2003;90(1):97–104. [PubMed: 12609811]
  13. Launois R, Reboul Marty J, Henry B, Bonneterre J. A cost-utility analysis of second-line chemotherapy in metastatic breast cancer: docetaxel versus paclitaxel versus vinorelbine. Pharmacoeconomics. 1996;10:504–521. [PubMed: 10169397]
  14. Leung PP, Tannock IF, Oza AM, Puodziunas A, Dranitsaris G. Cost-utility analysis of chemotherapy using paclitaxel, docetaxel, or vinorelbine for patients with anthracycline-resistant breast cancer (Structured abstract) Journal of Clinical Oncology. 1999;17(10):3082–3090. [PubMed: 10506603]
  15. Lewis R, Bagnall AM. The clinical effectiveness and cost-effectiveness of vinorelbine for breast cancer: a systematic review and economic evaluation. Health Technology Assessment (Winchester, England) 2002;6(14) [PubMed: 12583816]
  16. Li N, van Agthoven M, Willemse P, Uyl-de Groot C. A cost--utility analysis comparing second-line chemotherapy schemes in patients with metastatic breast cancer. Anti-Cancer Drugs. 2001;12(6):533–540. [PubMed: 11460000]
  17. Martín M, Ruiz A, Munoz M, Balil A, García-Mata J, et al. Gemcitabine plus vinorelbine versus vinorelbine monotherapy in patients with metastatic breast cancer previously treated with anthracyclines and taxanes: final results of the phase III Spanish Breast Cancer Research Group (GEICAM) trial. Lancet Oncol. 2007;8(3):219–225. [PubMed: 17329192]
  18. Monnier A, Bonneterre J, Roche H, Fargeot P, Namer M, Gustalla JP, et al. Phase III study: taxotere (TXT) versus 5-fluorouracil + Navelbine (FUN) in patients (pts) with metastatic breast cancer (MBC) as 2nd line chemotherapy (CT) (preliminary results) Ann Oncol. 1998;9(4S):12.
  19. Nabholtz J, Thuerlimann B, Bezwoda W, Melnychuk D, Deschenes L, et al. Docetaxel vs mitomycin plus vinblastine in anthracycline-resistant metastatic breast cancer. Oncology. 1997;11(8 Suppl 8):25–30. [PubMed: 9364538]
  20. O’Shaughnessy J, Miles D, Vukelja S, Moiseyenko V, Ayoub JP, et al. Superior survival with capecitabine plus docetaxel combination therapy in anthracycline-pretreated patients with advanced breast cancer: phase III trial results. Journal of Clinical Oncology: Official Journal of the American Society of Clinical Oncology. 2002;20(12):2812–2823. [PubMed: 12065558]
  21. Pierga JY, Fumoleau P, Brewer Y, Zelek L, Martin D, et al. Efficacy and safety of single agent capecitabine in pretreated metastatic breast cancer patients from the French compassionate use program. Breast Cancer Research and Treatment. 2004;88(2):117–129. [PubMed: 15564795]
  22. PSSRU, Unit Costs of Health and Social Care. 2007. http://www​​/uc/uc2007contents.htm.
  23. Reichardt P, von Minckwitz G, Thuss-Patience PC, Jonat W, Kölbl H, et al. Multicenter phase II study of oral capecitabine (Xeloda) in patients with metastatic breast cancer relapsing after treatment with a taxane-containing therapy. Annals of Oncology. 2003;14(8):1227–1233. [PubMed: 12881384]
  24. Silberman G, Gupta S, Berkowitz N, Leyland Jones B. Cost-effectiveness of capecitabine, continuous infusion 5-FU, gemcitabine and vinorelbine in the treatment of metastatic breast cancer [meeting abstract]. Proceedings of the 35th Annual Meeting of the American Society of Clinical Oncology; 1999 May 15–18; Atlanta, GA, USA. Alexandria, VA: ASCO; 1999. p. A1629.
  25. Sjostrom J, Mouridsen H, Pluzanska A. Taxotere versus methotrexate–5-fluorourocil in patients with advanced anthracycline resistant breast cancer: preliminary results of a randomised Phase III study by the Scandinavian Breast Cancer Group [abstract] Proc ASCO. 1998;17:111.
  26. Talbot DC, Moiseyenko V, Van Belle S, O’Reilly SM, Alba Conejo E, et al. Randomised, phase II trial comparing oral capecitabine (Xeloda) with paclitaxel in patients with metastatic/advanced breast cancer pretreated with anthracyclines. British Journal of Cancer. 2002;86(9):1367–1372. [PMC free article: PMC2375384] [PubMed: 11986765]
  27. Verma S, Maraninchi D, O’Shaughnessy J, Jameison C, Jones S, Martin M, et al. Capecitabine plus docetaxel combination therapy: cost-effectiveness in patients with anthracycline-pretreated advanced breast carcinoma. Cancer. 2005;103(12):2455–2465. [PubMed: 15892043]
  28. Verma S, Ilersich AL. Population-based pharmacoeconomic model for adopting capecitabine/docetaxel combination treatment for anthracycline-pretreated metastatic breast cancer. The Oncologist. 2003;8(3):232–240. [PubMed: 12773745]



It is important to note that the term ‘first-line treatment’ is used here to describe treatment given to patients who are not anthracycline-resistant or failing. Since the number of patients in this category is now very small, the term ‘first-line treatment’ in the rest of this report refers to the first therapy received by a patient with advanced disease for which anthracycline therapy is not suitable.


Similarly, ‘second-line treatment’ as referred to here is later referred to as ‘first-line treatment’ in the rest of this appendix.


The accepted threshold for evaluating the cost-effectiveness of any given treatment in the context of the UK is around £20,000–£30,000 per QALY. As such the range of £1,990–£2,431 per QALY shows docetaxel therapy to be very cost-effective compared to paclitaxel therapy.


The packages are artificial constructs designed for use in the model. There is no assumption that each individual will receive precisely this pattern of care, rather this was an attempt to estimate the costs of supportive care in general at different points in the patient pathway.


The type of therapist was not made explicit. The unit cost of all therapists listed in the PSSRU costs was £40 per hour. This was roughly the same for an hour of home visiting time.

Copyright © 2009, National Collaborating Centre for Cancer.

No part of this publication may be reproduced, stored or transmitted in any form or by any means, without the prior written permission of the publisher or, in the case of reprographic reproduction, in accordance with the terms of licenses issued by the Copyright Licensing Agency in the UK. Enquiries concerning reproduction outside the terms stated here should be sent to the publisher at the UK address printed on this page.

The use of registered names, trademarks etc. in this publication does not imply, even in the absence of a specific statement, that such names are exempt from the relevant laws and regulations and therefore for general use.

Bookshelf ID: NBK61867
PubReader format: click here to try


  • PubReader
  • Print View
  • Cite this Page
  • PDF version of this title (3.0M)
  • Disable Glossary Links

Related information

  • PMC
    PubMed Central citations
  • PubMed
    Links to pubmed

Recent Activity

Your browsing activity is empty.

Activity recording is turned off.

Turn recording back on

See more...