Evaluating PET–CT in routine surveillance and follow-up after treatment for cervical cancer: a cost-effectiveness analysis

Authors


Abstract

Objective

To undertake a cost-effectiveness analysis that compares positron emission tomography – computed tomography (PET-CT) imaging plus standard practice with standard practice alone in the diagnosis of recurrent or persistent cervical cancer during routine surveillance and follow-up of women who have previously been diagnosed and treated.

Design

Model-based economic evaluation using data from a systematic review, supplemented with data from other sources, and taking a UK National Health Service (NHS) perspective.

Setting

Secondary Care in England.

Population

Women at least 3 months after the completion of treatment, with either recurrent or persistent cervical cancer.

Methods

A state transition (Markov) model was developed using TreeAge Pro 2011. The structure of the model was informed by the reviews of the trials and clinical input. In the model, two diagnostic strategies were examined. A one-way sensitivity analysis, probabilistic sensitivity analysis, and a value of information analysis were also carried out.

Main outcome measures

Cost-effectiveness based on incremental cost per quality-adjusted life year (QALY).

Results

Adding PET–CT to the current treatment strategy of clinical examination and scanning [magnetic resonance imaging (MRI) and/or CT scan] during the routine surveillance and follow-up of women with recurrent or persistent cervical cancer is significantly more costly, with only a minimal increase in effectiveness. The incremental cost-effectiveness ratio (ICER) for the strategy of PET–CT as an adjunct to the standard treatment strategy that included clinical examination, MRI, and/or CT scan, compared with the usual treatment alone, was over £1 million per QALY.

Conclusion

The results of the current analysis suggest that use of PET–CT in the diagnosis of recurrent or persistent cervical cancer is not cost-effective. Current guidelines recommending imaging using PET–CT as a diagnostic or surveillance tool need to be reconsidered in light of these results. This study did not specifically investigate the use of PET–CT in women with symptoms and radiological suspicion of recurrence where exenteration was considered. More research in that specific area is required.

Introduction

Cervical cancer is a common gynaecological malignancy, with an estimated 31 400 new cases diagnosed each year in the European Union.[1] In the UK, approximately 2800 women are diagnosed with cervical cancer per year, accounting for around 2% of all female cancer cases.[2] In England, carcinoma of the cervix is a leading cause of cancer death in women, although it is rare in women below the age of 20.[3] In 2008, there were 1110 deaths from cervical cancer in the UK, giving a European age-standardised death rate of 2.7 per 100 000 person years.[1] First-line treatment for cervical cancer will vary depending on the stage of the disease. Approximately 70–80% of patients can be cured by initial treatment, which may be surgery or chemoradiotherapy, based on the stage and characteristics of the tumour; however, the initial treatment may not effect a cure, and in approximately 15% of patients disease is detected 3 months after treatment, which is defined as persistent cervical cancer (rather than recurrent). Recurrence is more common within the first 24 months after the initial diagnosis, but can happen up to 15 years after the initial diagnosis. Both persistent cervical cancer and recurrent cervical cancer are more likely in patients diagnosed at an advanced stage (stages IIB–IV).[3]

Positron emission tomography (PET) and computed tomography (CT) together provide an imaging technique referred to as PET–CT that combines these two previously independent techniques so that images acquired from both devices can be taken sequentially during the same patient session, and then combined into a single superimposed image.[3]

In recurrent cervical cancer either magnetic resonance imaging (MRI) alone or CT alone are currently considered first when recurrence is suspected, as both provide high-resolution images.[4] PET alone provides an anatomical image resolution of the order of 4–6 mm, which is significantly better than conventional gamma-cameras, but is inferior to the 1–2–mm resolution of CT or MRI. The combined PET–CT imaging adds precision to the anatomical localisation of active disease sites, compared with either technology separately.[5]

Thus PET–CT raises the possibility of identifying recurrent cervical cancer earlier, and typically before the woman experiences symptoms. There is anecdotal evidence to suggest that the use of radiation therapy and cancer staging is changing in response to the influence and availability of PET–CT, so that conventional imaging techniques are being abandoned in favour of PET–CT. The combined device still requires specialist input in its use and interpretation of results, however, and is considerably more expensive than the conventional approaches. Given that healthcare resources are limited, the use of PET–CT must be justified and the additional value in terms of patient outcomes must be clarified if it is to be recommended by decision makers.

As part of a wider National Institute for Health Research (NIHR) Health Technology Assessment project,[6] we carried out an economic evaluation to determine the cost-effectiveness of routinely adding PET–CT imaging to standard practice, and compared this with standard practice alone in the diagnosis of recurrent or persistent cervical cancer. In this article standard practice is defined as comprising clinical examination and either an MRI or CT scan alone, or both MRI and CT. In this article we will refer to a non-PET–CT scanning procedure as a ‘basic scan’. Although highly technical and specialised in their own right, in this instance a basic scan refers to either MRI alone or CT alone, or both, as opposed to PET–CT, as defined above, which is more complex.

An economic evaluation is a comparison of two alternative interventions in terms of costs and outcomes. In the UK, the National Institute for Health and Care Excellence (NICE) recommends that the results of economic evaluations should be expressed in terms of cost per quality-adjusted life year (QALY), as far as possible.[7] Model-based economic evaluations have the advantage of incorporating information from a variety of sources, to evaluate the costs and benefits associated with alternative policy decisions, while explicitly accommodating the uncertainty associated with data inputs, and are helpful in presenting this in a clear way to the decision maker.

The economic evaluation is intended to inform current diagnostic policy for suspected recurrent or persistent cervical cancer. We also set out to perform a value-of-information (VOI) analysis. This attempts to estimate the amount of money a decision maker would be willing to pay for additional information prior to making a decision. For the clinical decision presented in this article, VOI estimates how much value a decision maker would place on any additional future evidence related to recurrent cervical cancer, in order for a decision to be made with more certainty than could currently be made with the evidence available. Thus the VOI analysis is intended to highlight future research needs.

Methods

We report on the methods and results of the systematic review and the subjective preference elicitation exercise in an accompanying article.[6] Given the reliance on secondary data and the nature of the data available, the model-based economic evaluation takes the form of a cost–utility analysis and was carried out from a UK National Health Service (NHS) perspective in a secondary care setting.

Cost-effectiveness model

To assess the cost-effectiveness of the various diagnostic procedures along the treatment pathway, we developed a decision analytical model, referred to as a state transition (Markov) model, using TreeAge Pro 2011 (TreeAge Software Inc., Williamstown, MA, USA). The use of a model facilitates the comparison of alternative interventions that have not been directly compared, by explicitly representing the alternative test and treatment pathways a woman can follow for this clinical condition. A Markov model is a commonly used approach in decision analysis to handle the added complexity of presenting and analysing treatment options, which include repeated events and a multiplicity of possible consequences. It is the appropriate modelling approach for this evaluation because the time periods over which women received both the imaging and the interventions were relatively long, and because patients changed health states (e.g. asymptomatic without cancer to symptomatic without cancer) or experienced recurrent events over a long period of time.[8]

In the model we compare two diagnostic strategies: (1) a package that comprised a clinical examination and a basic scan (MRI or CT, or both), which represents the standard practice that women receive during follow-up assessment; and (2) a package that comprised clinical examination, a basic scan (MRI or CT, or both), and the routine addition of a PET–CT scan, regardless of the results of the MRI or CT scans, and regardless of the presence or absence of symptoms.

The starting point in the analysis is when women who have previously been treated for primary cervical cancer, either by surgery or chemoradiotherapy, based on the cancer stage that was defined at diagnosis, enter the model at time zero.[6] In the model it is assumed that women who are initially diagnosed with cervical cancer could receive three different management strategies: the first based on the original stage at diagnosis; the second based on the current development of the malignancy; and the third based on the tumour characteristics and fitness of the patient. At 3 months follow-up, if the results of the history and examination suggest the presence of malignancy-related abnormalities (from symptoms such as pain, vaginal bleeding, weight loss, neuropathy, or swelling of the abdomen or the legs) the woman will receive a biopsy to confirm the presence of persistent or recurrent cervical cancer. Because of these three different management strategies, a cohort of women following a pathway for the detection and treatment of potential recurrent cervical cancer cannot be considered homogenous in the model. The accuracy of detection and probability of treatment success for women in the recurrent stage is affected by their primary diagnosis and the treatment they previously received. To address this issue we used the same model structure for four separate analyses, to account for the four cohorts of women, based on primary treatment.

  1. Women who had already received surgery for early-stage primary cervical cancer.
  2. Women who in addition to surgery, as per cohort 1, had postoperative chemoradiotherapy for early-stage primary cervical cancer because of risk factors, i.e. positive margins, size, deep stromal invasion, etc.
  3. Women who had chemoradiotherapy for early-stage (stages I and IIa) primary cervical cancer, but no surgery.
  4. Women who had chemoradiotherapy for late-stage (stages IIb, III, and IV) primary cervical cancer, but no surgery.

For all four cohorts of women, the clinical pathways and model structure is identical. Figure 1 presents an illustrative Markov model structure and the health-state transitions that are possible within the model. Health states are shown in ovals and rows represent the transitions that can occur between health states. The health state definitions are presented in Table 1. All women who previously received treatment for primary cervical cancer will start in one of the four groups: (1) asymptomatic cancer at 3 months; (2) asymptomatic without cancer; (3) symptomatic without cancer; or (4) symptomatic cancer at 3 months. The transitions are presented in Box 1.

Box 1. Transitions taking place in the model

Asymptomatic women with cancer, at three months, will move to

  • Asymptomatic recurrence
  • Symptomatic recurrence
  • Post treatment: asymptomatic cancer at three months
  • Death

Asymptomatic women, without cancer, will remain or move to

  • Asymptomatic recurrence
  • Symptomatic recurrence
  • Symptomatic without cancer
  • Death

Symptomatic women, without cancer, will remain or move to

  • Asymptomatic without cancer recurrence
  • Asymptomatic recurrence
  • Symptomatic recurrence
  • Death

Symptomatic women with cancer, at three months, will move to

  • Symptomatic recurrence
  • Post-treatment: symptomatic cancer at three months
  • Death

Asymptomatic women with recurrences will remain or move to

  • Post treatment: asymptomatic
  • Symptomatic recurrence
  • Death

Symptomatic women with recurrence will remain or move to

  • Post treatment: symptomatic
  • Death

Post treatment asymptomatic women with cancer at three months will remain or move to

  • Death

Post treatment symptomatic women with cancer at three months will remain or move to

  • Death

Post-treatment asymptomatic women will remain then move to

  • Death

Post-treatment symptomatic women will remain then move to

  • Death
Table 1. Definition of 11 health states for recurrent cervical cancer pathways
Health stateAsymptomaticSymptomatic
  1. The model does not distinguish between recurrence and persistence. Recurrence: women who had previously been treated for initial cancer with surgery and/or postoperative chemoradiotherapy, or chemoradiotherapy alone, are considered recurrent. Persistence: women who were originally treated with chemoradiotherapy and who are detected at this stage are considered to have persistent disease.

  2. Treatment for primary cancer not included.

At 3 monthsWomen without symptoms of cancer who are likely to have recurrent or persistent cancer which may or may not be detected at 3 months follow-up.Women with symptoms of cancer who have been diagnosed with recurrent or persistent cancer which may or may not be detected at 3 months follow-up (i.e symptoms may or may not be cancer).
Without recurrenceWomen who had previously been treated for initial cervical cancer and are receiving follow-up care but are free of recurrent cervical cancerWomen who experience symptoms which they assume to be related to recurrent or persistent cervical cancer; however, on follow-up and confirmatory testing these women will be cleared of recurrent or persistent cervical cancer
RecurrenceWomen without symptoms of cancer who have cancer that will not have been detected before a potential follow-up appointment; this may include women who may have had cancer not detected at the first 3 months follow-upWomen with symptoms that are related to cancer who received follow-up care and are confirmed of recurrent or persistent cervical cancer
Post-treatment cancer at 3 monthsFollowing diagnosis for cancer at first follow-up having been asymptomatic, women will receive new treatment (treatment type based on initial treatment and location of cancer recurrence or persistence)Following diagnosis for cancer at first follow-up having been symptomatic, women will receive new treatment (treatment type based on initial treatment and location of cancer recurrence or persistence)
Post-treatmentFollowing diagnosis for recurrent cervical cancer after being asymptomatic, women will receive new treatment (treatment type based on initial treatment and location of cancer recurrence or persistence)Following diagnosis for recurrent cervical cancer after being symptomatic, women will receive new treatment (treatment type based on initial treatment and location of cancer recurrence or persistence)
DeathWomen may die from natural causes or may die as a result of recurrent or persistent cervical cancer 
Figure 1.

Markov model structure: health states and patient flow.

A number of assumptions are required in order to develop a workable model structure and enable the analysis to be carried out.

  1. Women are followed-up with examinations and imaging [basic scan(s), with or without the routine addition of PET–CT], regardless of the presence of symptoms, and receive the diagnostic package of examination and imaging every 3 months for 2 years, then every 6 months for 2 years, and then annually for 1 year; the total follow-up period is 5 years.
  2. Women who were symptomatic at 3 months and whose cancer has not been detected cannot become asymptomatic.
  3. Women with symptoms that they suspect are related to cervical cancer are usually given an urgent appointment or their pre-existing follow-up appointment is brought forward.
  4. The sensitivity and specificity of the confirmatory biopsy test was assumed to be 100% accurate.
  5. Women who previously received chemoradiotherapy for primary cervical cancer, and who were not diagnosed with persistence at the 3–months follow-up (i.e. not persistent or persistent cases missed), will be treated similarly to women with recurrent cervical cancer, when detected.
  6. The PET–CT procedure includes both the preparation and scanning of the patient; therefore, the preparation activity is implicit in the PET–CT scan resource use (NHS Reference Cost Team, pers. comm., 12 April 2011).
  7. Women who received treatment for primary cervical cancer and who have not survived at 5 years are assumed to have died from recurrent cervical cancer only.
  8. It is assumed that the utility for recurrent cervical cancer is equivalent to the average of the utility for primary stage–III and -IV cervical cancer.
  9. There is a constant hazard over 5 years for early-stage recurrent cervical cancer (i.e. the risk of recurrence is the same at 4 years as that at 1 year).
  10. It is assumed that women treated for recurrent cervical cancer will have the same quality of life following treatment as they had after treatment for initial cervical cancer.

Data required for the model

We populated the model with the rates of incidence of recurrent cervical cancer derived from the literature and from consultation with clinical experts.[3] We calculated the rates of recurrence using a two-stage process. First, we derived the survival following treatment for primary cervical cancer from the disease-free survival curves in Landoni et al.[9] in addition to the progression-free survival curves presented in Keys et al.[10] and from the overall survival curves following initial treatment presented in Landoni et al. and Vale et al.[9, 11] We used the information from these sources with the standard assumption of an exponential survival function. We calculated the 3–month survival for women who received surgical treatment using the disease-free survival curve presented in the Landoni et al.[9] Similar procedures were used to calculate survival rates following postoperative chemoradiotherapy and following chemoradiotherapy alone. We calculated rates of recurrence based on the initial survival of women, in the branch of women who were symptomatic without cancer, using the conditional probabilities following survival and the formulae presented in Table 2. In the absence of data on the proportions of women in each of the four health states at the start of the model, we used as a proxy the probabilities for women moving from the state ‘symptomatic without cancer’ (Table 2).

Table 2. Percentages of women receiving initial treatment strategies for cervical cancer
ManagementPercentages of women receiving careExplanation
Surgery30–40% of women have surgeryaSurgery typically involves radical hysterectomy or trachelectomy
Of these, 70–80% are curedNo further treatment needed
The remaining 20–30% of women receive adjuvant postoperative chemoradiotherapyThis is because the histological examination of the tumour has shown positive margins, there are positive lymph nodes, or because of tumour size or volume, lymphovascular space invasion, or stromal invasion
Chemoradiotherapy50–60% of women receive chemoradiotherapy 
Of these, 70% of the women are curedNo further treatment needed
The remaining 30% of women are those who have not responded to first-line treatment (chemoradiotherapy) and may have persistent disease.Persistent disease can be detected at 3–months follow-up after the initial course of treatment has finished
Palliative treatment with chemotherapy or radiotherapy (or both)<5% of women receive palliative treatment with chemotherapy or radiotherapy 
ParameterWritten formulaFormula
  1. a

    Source: S. Sundhar, pers. comm., April 2011.

Asymptomatic at 3 months(Probability of becoming recurrent, having been symptomatic without cancer × probability of being asymptomatic recurrence, conditional on recurrence)(F2 × F3)
Asymptomatic without cancer[(1−probability of becoming recurrent having been symptomatic without cancer) × (probability of becoming asymptomatic without cancer, conditional on no recurrence)][(1−F2) × F4]
Symptomatic at 3 months[Probability of becoming recurrent, having been symptomatic without cancer × (1−probability of being asymptomatic recurrence, conditional on recurrence)][F2 × (1−F3)]
Symptomatic without cancer[(1−probability of becoming recurrent, having been symptomatic without recurrent cancer) × (1−probability of becoming asymptomatic without cancer, conditional on no recurrence)][(1−F2) × (1−F4)]

Table 3 presents the rates of recurrence and the accuracy data used in the models. The test accuracy results used in the model were based on values estimated in a subjective preference elicitation exercise,[6] because of a lack of estimates available in the published literature. The results of the subjective preference elicitation exercise had face validity, as judged by feedback from the participants. We converted the elicited predictive values to sensitivities and specificities for use in the models.[6]

Table 3. Rate of recurrence of cervical cancer and accuracy data used in the models
ParameterSurgeryChemoradiotherapyPost-surgery chemoradiotherapySource (reference)
EarlyLateEarlyLateEarlyLate
Asymptomatic at 3 months0.00410.00410.00410.0041Derived using data from the literature and a clinical expert.[3]
Asymptomatic without cancer0.89070.89070.89070.8907
Symptomatic at 3 months0.00620.00620.00620.0062
Symptomatic without cancer0.09900.09900.09900.0990
InterventionPersistent/recurrent cervical cancerSource (reference)
AsymptomaticSymptomatic
SensitivitySpecificitySensitivitySpecificity
Accuracy data used in the model
Clinical follow-up, magnetic resonance imaging (MRI), with or without computed tomography (CT)45.4398.4785.0989.78These data were derived from a preference elicitation, described in detail in Meads et al.[3]
Clinical follow-up, magnetic resonance imaging (MRI), with or without computed tomography (CT) and positron emission tomography/computed tomography (PET/CT)65.2598.5889.7191.88 
Treatment options2–year survival (%)3–year survival (%)5–year survival (%)Source (reference)
Survival following treatment for recurrent cervical cancer
Radiotherapy40.2% (95% CI 31.6–48.6%) for whole groupJain et al.[11]
Chemotherapy64%Pearcey et al.[12]
Chemoradiotherapy44%25%Maneo et al.[13]
Pelvic exenteration63%Beitler et al.[14]
Untreated3.1%Adriano et al.[21]
Model (initial treatment)3–month survival (%)Source (reference)
Three-month survival following treatment for recurrent cervical cancer
Model 1: early stage, treated with surgery0.9307Derived from the survival rates in the literature and the proportions of women receiving treatment for recurrent cervical cancer.[3]
Model 2: early stage, treated with chemoradiotherapy0.9778
Model 3: late stage, treated with chemoradiotherapy0.9779
Model 4: early stage, treated with surgery and postoperative chemoradiotherapy0.9778

The results for survival following treatment for recurrence,[12-15] and for persistence, used in the model are reported elsewhere.[5]

Costs and resources

The costs for resources used were those directly incurred by the NHS: clinical examination; diagnostic imaging (PET–CT, MRI, and CT); confirmatory biopsy; and treatment. Costs were obtained from the NHS reference costs and from the literature.[5, 6] Costs incurred during the primary diagnosis and treatment of cervical cancer were not considered. Other costs included those for long-term and end-of-life care. In the models, we assume that recurrence occurred only once. Unit costs are presented in Table 4. All costs obtained from the literature were adjusted to 2010 prices by the Hospital and Community Health Service (HCHS) combined pay and price inflation index,[16-18] and were discounted at 3.5% per annum (as recommended by NICE).

Table 4. Cost data used in the model (all costs presented in UK £, 2010)
DescriptionUnit costSource (reference)
  1. NEC, not elsewhere classified.

Examination & imaging
Clinical examination£28.17Curtis 2010[18]
Positron emission tomography/computed tomography (PET/CT)£744.00National schedule of reference costs 2009–2010[15]
Magnetic resonance imaging (MRI)£366.00National schedule of reference costs 2009–2010[15]
Computed tomography (CT)£162.00National schedule of reference costs 2009–2010[15]
Confirmatory test
Cone biopsy of cervix uteri NEC£968.00National schedule of reference costs 2009–2010[15]
Treatment
Surgical£6723.00National schedule of reference costs 2009–2010[15]
Chemoradiotherapy£14 495.14Brush et al. 2011[16]; Curtis 2010[18]
Palliative
Chemotherapy£356.56Clark et al. 2002[17]; Curtis 2010[18]
Radiotherapy£1167.79Clark et al. 2002[17]; Curtis 2010[18]
Weighted treatment costs
Model 1£13 011.00 
Model 2£1629.85Derived from the
Model 3£993.20Literature and from
Model 4£1629.85Consultation with a clinical expert[3]

The treatment of recurrent cervical cancer varies depending on the site and extent of recurrence, the type of previous treatment received, the time elapsed since primary treatment, and the patient's performance status. Treatment options for recurrent cervical cancer include surgical (radical hysterectomy or pelvic exenteration),[15] chemoradiotherapy,[14] and palliative treatment (which can be radiotherapy or chemotherapy alone).[12, 14] A weighted mean cost of treatment was calculated in the models, based on the relative proportion of women receiving each treatment in a cohort of women with a detection of recurrence or who were undergoing follow-up. All treatment costs used in the models are presented in Table 4.

Outcomes

We used three different outcome measures in the model: (1) recurrent case treated; (2) death from recurrent cervical cancer avoided; and (3) quality-adjusted life years (QALYs).

The systematic review identified no studies reporting quality-of-life data following treatment for recurrent cervical cancer in a form that could be used in the model. It was assumed that women treated for recurrent cervical cancer would have the same quality of life following treatment for the initial cervical cancer. Thus, to estimate the QALYs, utility weightings for the treatment of women who had been diagnosed with recurrent cervical cancer were obtained from Goldie et al.[19] The authors reported utility weightings for women with invasive cancer by FIGO stage. An average utility weighting based on stages III (0.56) and IV (0.48) was calculated, giving a utility weighting for recurrent cervical cancer of 0.52.[20]

Analysis

The recurrent cervical cancer model begins with a hypothetical cohort of women who have previously been treated for primary cervical cancer and who are now receiving follow-up assessment, as outlined above. In the models we estimate the mean costs associated with the diagnostic procedure and we assume that women entering the model are aged 50 years. The model cycles are every 3 months. The follow-up pattern is assumed as every 3 months for 2 years, then twice a year for 3 years.[21] In the model a total time period of 5 years is assumed to represent the length of time women are followed-up after being diagnosed and treated, and the time within which recurrent cervical cancer is likely to occur.[21]

We carried out a cost–utility analysis from the perspective of the NHS in a secondary care setting. The primary outcome is cost per QALY, but a secondary outcome measure of cost per recurrent case treated was also estimated. The results of the cost–utility analysis are presented in terms of the incremental cost-effectiveness ratios (ICERs).

Two deterministic sensitivity analyses were undertaken. First, we carried out sensitivity analysis using the range for 5–year survival for women who are untreated for recurrent cervical cancer (3.0–60%).[13, 22] The wide range represents the uncertainty in the literature. The estimate of 3% is from a study dated from between 1906 and 1926, and whereas survival is now likely to be higher than this for untreated cervical cancer, because treatment does exist now it is unclear what the survival rate would be without treatment. Second, we changed the current imaging and follow-up schedule from 3 months to annual imaging and follow-up. The current follow-up schedule is not based on evidence but is considered conventional practice. It was considered appropriate to explore the impact of a surveillance strategy with annual scanning/follow-up, which may potentially benefit the patient/system by reducing the number of clinical follow-up appointments undertaken without compromising outcomes (S. Sundar, pers. comm.). It was a deliberate approach not to construct a model in which clinical examination and imaging are performed at divergent intervals (e.g. with clinical examination every 3 months, but imaging carried out annually), because there are no data on the sensitivity and specificity of clinical examination at each time point, and therefore incorporating follow-on actions from that clinical examination would be hypothetical and unjustifiable.

Probabilistic sensitivity analysis uses Monte Carlo simulation, which refers to the use of random numbers in a model. Monte Carlo simulation recalculates a model multiple times. It can update any number of parameters between model re-calculations, assigning values that are randomly sampled from probability distributions. The advantage of this type of analysis is that all parameter uncertainties can be incorporated into the analysis. Sampling parameter values from probability distributions, rather than from a simple range defined by the upper and lower bounds, places greater weight on likely combinations of parameter values, and simulation results quantify the impact of major uncertainties on the model, in terms of the confidence that can be placed in the analysis results.

Value-of-information (VOI) methods implicitly assume that current decisions should be made on the basis of cost-effectiveness, but the method helps to explore whether or not there is value in reducing uncertainty and collecting additional information that might lead to the current decision being overturned. We carried out probabilistic sensitivity analysis including a value of information analysis, for which the methods and results are presented in Appendix S1.

Results

The results for women who have been treated for early-stage cancer by surgery (model 1) showed that routinely adding PET–CT to standard imaging practice had an average cost of approximately £18 757, with a corresponding QALY gain of 4.1096. In comparison, the average cost of standard imaging practice alone was approximately half the cost, at £9169, with a corresponding QALY gain of 4.1086. Therefore, adding PET–CT to the standard investigations costs twice as much as standard investigations alone, and the increased QALY gain is just 0.001. The estimated ICER for the addition of PET–CT to standard imaging practice compared with standard practice alone was £9 254 000 per QALY. This indicates that for every additional QALY gained from standard practice, there is an incremental cost of £9 254 000.

For women who have been treated for early-stage cancer by chemoradiotherapy (model 2), the results show that PET–CT together with standard practice had a mean cost of approximately £17 122 with a corresponding QALY of 4.1581. Standard practice had a mean cost of approximately £7695 with a corresponding QALY of 4.1501. The estimated ICER for the PET–CT together with standard practice, compared with standard practice alone, was approximately £1 173 000 per QALY.

The results for models 3 and 4 follow the same pattern, in that there is a high cost for a small gain in QALYs for both models. The base-case deterministic results for all four models are presented in Table 5.

Table 5. Base-case results from the analysis cost per QALY
StrategyMean cost per strategyDifference in costsEffectiveness (QALY)Incremental QALYsIncremental cost-effectiveness ratio (ICER)
  1. a

    The apparent anomaly in the subtraction is a result of rounding effects.

Model 1: women who have been treated for early-stage cancer by surgery
Standard practice£91694.1086
PET/CT together with standard practice£18 757£95884.10960.0010£9 254 000
Model 2: women who have been treated for early-stage cancer by chemoradiotherapy a
Standard practice£76954.1501
PET/CT together with standard practice£17 122£94284.15810.0080£1 173 000
Model 3: women who have been treated for late-stage cancer by chemoradiotherapy
Standard practice£76124.1507
PET/CT together with standard practice£17 031£94194.15950.0088£1 065 000
Model 4: women who have been treated for early-stage cancer by postoperative chemoradiotherapy a
StrategyMean cost per strategy (£)Difference in costsEffectiveness (QALY)Incremental QALYsIncremental cost-effectiveness ratio (ICER)
Standard practice76954.1501
PET/CT together with standard practice17, 12294284.15810.0080£1 173 000

The deterministic results for the cost per recurrent case treated were over £600 000 per case for all four models. PET–CT as an adjunct to standard practice was both more costly and more effective than standard practice alone, with ICERs of approximately £670 000 per recurrent case treated (Table 6). In the deterministic sensitivity analysis, we explored the impact of changing the survival rate according to the extremes of the wide ranging estimates identified for the probability of survival (3% to 60%). The change did affect the results but the effect was predictable. Changes to other variables in the sensitivity analysis had no discernible effect (Table 7).

Table 6. Results from the analysis cost per recurrent case treated
StrategyMean cost per strategyDifference in costsEffectiveness (cases treated)Incremental cases treatedIncremental cost-effectiveness ratio (£) (ICER)
  1. a

    The apparent anomaly in the subtraction is a result of rounding effects.

Model 1: women who have been treated for early-stage cancer by surgery a
Standard practice£91690.1296
PET/CT together with standard practice£18 757£95880.14360.0141£681 000
Model 2: women who have been treated for early-stage cancer by chemoradiotherapy a
Standard practice£76950.1296
PET/CT together with standard practice£17 122£94280.14360.0141£670 000
Model 3: women who have been treated for late-stage cancer by chemoradiotherapy a
Standard practice£76120.1296
PET/CT together with standard practice£17 031£94190.14360.0141£669 000
Model 4: women who have been treated for early-stage cancer by postoperative chemoradiotherapy a
Standard practice£76950.1296
PET/CT together with standard practice£17 122£94280.14360.0141£670 000
Table 7. Summary of deterministic sensitivity analyses cost utility results
Model number and sensitivity analysesIncremental costsIncremental effectiveness (QALY)ICER (cost per QALY)
Base case (model 1)£95880.0010£9 254 000
1. Changing the 5–year survival rate following untreated cervical cancer (from 3.0 to 60.0%)£9528−0.0072(dominance)
2. Halving the utility value for recurrent cervical cancer from 0.5200 to 0.2600£95880.0052£1 829 000
3. Changing the current follow-up schedule to annual follow-up£49740.0008£6 091 000
Base case (model 2)£94280.0080£1 173 000
1. Changing the 5–year survival rate following untreated cervical cancer (from 3.0 to 60.0%)£9419−0.0015(dominance)
2. Halving the utility value for recurrent cervical cancer from 0.5200 to 0.2600£94280.0122£771 000
3. Changing the current follow-up schedule to annual follow-up£48240.0069£697 000
Base case (model 3)£94190.0088£1 065 000
1. Changing the 5–year survival rate following untreated cervical cancer (from 3.0 to 60.0%)£9413−0.0007(dominance)
2. Halving the utility value for recurrent cervical cancer from 0.5200 to 0.2600£94190.0126£745 000
3. Changing the 3–month survival to 0.9307£94190.0027£3 527 000
4. Changing the current follow-up schedule to annual follow-up£48150.0072£673 000
Base case (model 4)94280.0080£1 173 000
1. Changing the 5–year survival rate following untreated cervical cancer (from 3.0 to 60.0%)9419−0.0015(dominance)
2. Halving utility value for recurrent cervical cancer from 0.5200 to 0.260094280.0122£771 000
3. Changing the current follow-up schedule to annual follow-up48240.0069£697 000

Appendix S1 presents the results of the probabilistic sensitivity analysis for model 1. In addition, Figure S1 presents a Monte Carlo simulation for model 1, and Figure S2 presents the corresponding cost-effectiveness acceptability curves (CEACs) for this model. A CEAC shows the probability that an intervention is cost-effective compared with the alternative, given the observed data, for a range of maximum monetary values that a decision-maker might be willing to pay for a particular unit change in outcome.

Discussion

Principal findings

The results of the deterministic analyses based on the outcome of cost per QALY show that routinely adding PET–CT to the current treatment strategy (clinical examination and basic scan) is significantly more costly, with only a minimal increase in effectiveness. This result holds true for all four models used in the analyses to represent the alternative treatment paths women followed for their treatment of primary cancer. These prior treatment paths were differentiated to ensure the results of the current analysis were not influenced by previous treatment for primary cervical cancer.

The ICER for the strategy of PET–CT as an adjunct to the standard treatment strategy, compared with the usual treatment alone, was over £1 million per QALY, at least, in all four models. For women who had been treated for early-stage cancer by surgery (model 1), this ICER was £9.3 million per QALY. For women who had been treated for early-stage cancer by chemoradiotherapy (model 2), this ICER is £1.2 million per QALY. For women who had been treated for late-stage cancer by chemoradiotherapy (model 3), this ICER is £1.1 million per QALY. For women who had been treated for early-stage cancer by postoperative chemoradiotherapy (model 4) the ICER is £1.2 million per QALY.

We also explored the ICER based on the outcome of cost per additional case of recurrence treated. For all four models the additional cost per additional case of recurrence is in the region of £600 000 per case.

The acceptable ICER threshold used by NICE is £20 000–30 000 per QALY. This means an ICER has to be below this for a technology to be considered cost-effective. The probabilistic sensitivity analysis (Appendix S1) suggests that a strategy with PET–CT as an adjunct to standard practice is not likely to be considered cost-effective for any of the models and data used in this analysis given the current willingness-to-pay thresholds.

The sensitivity analysis showed that in terms of the data used in the models, and based on the current evidence available, there were no doubts (within plausible estimates) about the results of the current analysis. Thus, based on the data currently available and the expert opinion used for the models presented here, PET–CT is not a cost-effective adjunct to standard treatment in the diagnosis of recurrent cervical cancer during the routine surveillance and follow-up of women who have been previously diagnosed and treated.

Strengths and limitations of the study

The strength of this economic evaluation is that the analysis is based on the best available data having established in a systematic review of the evidence that test accuracy data were severely limited. We also found that the other data required for the analysis were scarce. A subjective preference elicitation exercise was carried out using expert opinion before any analysis was undertaken. All assumptions used in the model were agreed by the team and were based on expert advice. No assumption or item of data from the elicitation exercise was changed after the analysis started, apart from in the sensitivity analysis.

There are, however, some major limitations in the analysis that must be considered in the interpretation of the results. The findings reported here are based on a model of routine use of a basic scan and PET–CT at each follow-up point, rather than the use being dictated by the presence of symptoms. But the clinical pathways upon which the model is based were constructed on the advice of the clinical experts in this field and the best available clinical evidence. Serious concerns are also based on the availability of suitable data identified in the clinical literature review. It is apparent that some use of PET–CT and associated tests in clinical practice are based on symptoms, but there is currently no available evidence to support an evaluation to explore the replacement of CT and MRI by PET–CT for this clinical decision.

In addition to the absence of accuracy data for PET–CT, data on the effectiveness of appropriate treatments were also lacking. Thus data for the proportions of patients receiving treatment for recurrent cervical cancer were again provided by clinicians based on their best clinical knowledge. Utility data for women diagnosed with recurrent cervical cancer were, with the approval of the clinicians on the team, calculated based on the average utility values for women who had been diagnosed for primary cervical cancer at stages III and IV. Also, utility values for women treated are based on the utility values from Lang et al. [23] for primary cervical cancer but not recurrent cervical cancer. It is also worth clarifying that the data in the literature on survival did not report survival according to stage for women who have been treated for recurrent cervical cancer.

Where confidence intervals were not reported in the literature, in order to conduct the probabilistic sensitivity analysis arbitrary ± ranges were used. The limited availability of data also means that any correlations existing between the sensitivity and specificity data for the range of diagnostic tests have been ignored. Cost data for tests were available in very few published studies, and only unit costs for relevant resource use were available.

Comparison with existing studies

No studies were identified that had considered the relative cost-effectiveness of available technologies for recurrent cervical cancer, and therefore appropriate comparisons with other existing studies are not possible.

Meaning of the study

Based on the current model and given the limitations in the availability of data, the results of the current analysis suggest that the routine addition of PET–CT in the diagnosis of recurrent cervical cancer is not cost-effective. The results are not even close to the current willingness-to-pay thresholds that are accepted in the UK by decision-making bodies such as NICE.

The results reflect enormous uncertainty at many levels, and so a better expression of our current understanding is that the cost-effectiveness of PET–CT combined with usual tests and treatment for detecting recurrent cervical cancer is not proven.

Existing guidelines recommend a PET–CT scan in patients in whom recurrence is suspected on the basis of MRI or CT prior to exenterative surgery.[21] Guidelines also recommend a PET–CT scan at 9 months of follow-up for women who have had chemoradiotherapy.[21] Although we did not explicitly evaluate the selected use of PET–CT in addition to imaging with CT/MRI, where standard imaging was suspicious of or equivocal for recurrence, our analysis demonstrates that on the basis of the best-available evidence, PET–CT scanning in women for surveillance is not cost-effective.

Unanswered questions and future research

The VOI analysis showed that the expected value of perfect information (EVPI) is zero. This is because at all willingness-to-pay levels plotted, the probability that PET–CT is cost-effective never went above zero. This means that according to all current and available evidence there is no missing piece of information that would change the current decision; however, the EVPI reflects parameter uncertainty in the model, which in this case was based on the preference elicitation exercise, and so has limitations in itself. Therefore, the EVPI of zero does not necessarily imply that it would not worth be measuring the accuracy of PET–CT directly (Appendix S1).

A diagnosis of recurrent cervical cancer must be an extremely distressing situation for women and their families. Current evidence suggests that there are huge gaps in our knowledge about their quality of life and survival given such a diagnosis. Adding an additional PET–CT test to the toolkit for the woman might add something in terms of reassurance and have a utility, regardless of the accuracy of the test; however, given that the incremental accuracy of such a test is currently not clear, in addition to the lack of evidence about survival and quality of life, a case for its implementation in current practice cannot yet be supported.

This study did not specifically investigate the use of PET–CT in women with symptoms and radiological suspicion of recurrence where exenteration was considered, and more research in that specific area is required.

What is already known on this topic

  • Both CT and MRI are high-resolution anatomical imaging techniques that are commonly used to detect potential tumours. MRI and CT are currently considered first when recurrence is suspected.
  • The addition of PET–CT to current imaging practice may increase the accuracy of detection of recurrent cervical cancer in symptomatic and asymptomatic women.
  • Existing guidelines recommend the use of PET–CT in women regardless of symptom status at 9 months post-completion of treatment for cervical cancer, and in symptomatic women with positive MRI or CT results, in whom salvage therapy (pelvic exenteration or radiotherapy) is being considered.

Disclosure of interests

All authors declare that: (1) no authors have support from any company for this work; (2) no authors have any relationships with any company that might have an interest in this work in the previous 3 years; (3) their spouses, partners, or children have no financial relationships that may be relevant to this work; and (4) no authors have any non-financial interests that may be relevant to this work.

Contribution to authorship

T.E.R. contributed to the design of the whole project and obtained funding with K.K. and S.S. Both T.E.R. and P.B. designed the model-based cost-effectiveness analysis. T.E.R. prepared the article as lead writer. Under the supervision of P.B., P.A. constructed the model with advice from T.E.R., C.M., and S.S. Under the supervision of T.E.R. and P.B., P.A. identified and collected the required data on costs and effectiveness for the model, and carried out the analysis. C.D., S.M., M.K., A.Z., P.C., P.G., P.M.–H., E.B. all commented on drafts of the article. T.R. is the guarantor.

Details of ethics approval

No ethical approval was required for this model-based economic evaluation.

Funding

This review was funded by the National Institute for Health Research Health Technology Assessment Programme (9/29/02).

Ancillary