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

Brush J, Boyd K, Chappell F, et al. The Value of FDG Positron Emission Tomography/Computerised Tomography (PET/CT) in Pre-Operative Staging of Colorectal Cancer: A Systematic Review and Economic Evaluation. Southampton (UK): NIHR Journals Library; 2011 Sep. (Health Technology Assessment, No. 15.35.)

Cover of The Value of FDG Positron Emission Tomography/Computerised Tomography (PET/CT) in Pre-Operative Staging of Colorectal Cancer: A Systematic Review and Economic Evaluation

The Value of FDG Positron Emission Tomography/Computerised Tomography (PET/CT) in Pre-Operative Staging of Colorectal Cancer: A Systematic Review and Economic Evaluation.

Show details

10Economic evaluation

Aims and objectives

The aim of this chapter is to determine whether or not FDG PET/CT is cost-effective as an add-on test in comparison to routinely used imaging modalities for pre-operative staging in patients with primary, recurrent and metastatic CRC. Probabilistic decision-analytic modelling was undertaken (using Monte Carlo simulation) to address the following questions:

  • Is FDG PET/CT likely to be cost-effective as an add-on test for pre-operative staging in CRC compared with alternative methods of diagnosis and staging, given current evidence and uncertainty?
  • In which patient groups (i.e. primary rectal cancer, primary colon cancer, recurrent rectal cancer, recurrent colon cancer, metastatic disease) is it likely to be cost-effective?
  • Under what circumstances is it likely to be cost-effective?

A value of information analysis was also undertaken to help inform whether or not there is potential worth in undertaking further research.

Methods

The economic evaluation is based upon current evidence, utilising decision modelling techniques to synthesise data from numerous sources.97,98 The evaluation is undertaken from the perspective of the UK NHS, reporting short-term outcomes in terms of the incremental cost per correct diagnosis, and longer-term outcomes in terms of the incremental cost per quality-adjusted life-year (QALY) gained. Reporting QALY outcomes enables the analysis to incorporate the potential patient management implications of accurate and inaccurate diagnoses, particularly the implications for patients' quality of life.

FDG PET/CT and conventional imaging devices have been found to have different diagnostic test accuracies for staging primary, recurrent and metastatic CRC. As such, three separate economic models were designed to address the questions outlined in the aims. Patient management routes also differ between colon and rectal cancer, and so the primary and recurrent models were adapted to incorporate the specifics of rectal and colon cancer separately. The economic evaluation therefore involved the development of five models, based on the three cancer stages of interest. These five evaluations assessed the cost-effectiveness of FDG PET/CT as an add-on imaging device in pre-operative staging for (1) primary rectal cancer, (2) primary colon cancer, (3) recurrent rectal cancer, (4) recurrent colon cancer and (5) metastatic disease.

The following section outlines the various sources of evidence used in the analyses. This is followed by a description of the design, development and data used to populate each model.

Literature

The economic models were designed, developed and populated based on a variety of information sources (in particular published data sources) and literature, and in consultation with clinical experts.

Previous economic evaluations of imaging devices for CRC were used to aid the design of the models, while the preceding systematic reviews were used to derive diagnostic test accuracy evidence for FDG PET/CT and alternative imaging modalities. Economic and non-economic literature was required to inform specific model parameters, such as resource use, implications of diagnosis on patient management and therapeutic impact, quality of life and survival. Costing and resource use information was obtained from both the literature and UK NHS cost information sources such as the British National Formulary,99 Department of Health reference costs100 and the Personal Social Services Research Unit (PSSRU).101

Papers that were considered to be potentially relevant for the health economic evaluation were identified by the systematic reviewers during their screening process and passed on to the health economists as first-line literature to inform the development of the economic models. These initial papers provided an indication of the types of literature that were available, and helped inform the design of the economic evaluations. Having established this first-line literature, a separate non-systematic literature search was undertaken in November 2009 to provide further information on the various parameters for the economic models. The objective was to search for and utilise information from economic evaluations and non-economic papers to develop and populate the economic models. Specifically, the search considered what evidence was available regarding the costs, treatment outcomes, management pathways, overall survival, quality of life and adverse events experienced by CRC patients undergoing pre-operative screening for primary, recurrent or metastatic CRC.

The following electronic databases were searched: MEDLINE, EMBASE, Web of Science, CINAHL Plus, The Cochrane Library [NHS Economic Evaluation Database (NHS EED), Health Technology Assessment (HTA), Cochrane Central Register of Controlled Trials (CENTRAL), Database of Abstracts of Reviews of Effectiveness (DARE)], Health Management Information Consortium and the Cost-effectiveness Analysis Registry. Specific searches were constructed for four main areas (FDG PET/CT imaging for CRC, economics, adverse events or quality of life, and decision analysis) for each of the databases, as detailed in Appendix 3. Inclusion criteria were applied to include relevant publications in any language that provided information on the diagnostic imaging devices FDG PET/CT, contrast-enhanced CT or MRI for detecting CRC with regards to the topic areas of economic evaluation, costing, patient management and therapeutic impact, quality of life and overall survival. Papers that provided details only on diagnostic test efficacy were excluded. Conference proceedings and abstracts were also excluded. The search outputs are detailed in Table 14: a total of 51 papers deemed to be of relevance were identified from the search, plus an additional four quality of life papers identified through handsearching.

TABLE 14. Economic literature search results.

TABLE 14

Economic literature search results.

Information from this literature was used in consultation with the clinical experts involved with the project to design the models, in particular to identify appropriate comparators, management pathways and parameter estimates for each model.

The systematic review undertaken by the research team was intended to yield data on diagnostic test accuracy for the various imaging devices, which would be pooled in meta-analyses to inform the main parameters for the economic models. As discussed, the systematic review found inadequacies and reporting bias in published papers for all stages of CRC disease. Because of the lack of papers it was deemed inappropriate to undertake a meta-analysis in primary CRC. Meta-analyses were undertaken for recurrent and metastatic CRC; however, the pooled estimates for FDG PET/CT were considered to be an inaccurate reflection of diagnostic test accuracy and the CIs were tight around the pooled means, which is restrictive in terms of capturing a wide range of uncertainty. Therefore, the economic analyses considered the papers identified by the systematic review individually along with other literature identified through the economic search and considerable input from the clinical experts in order to decide which data to incorporate in the economic models.

Model structure

Each model was developed using a decision tree design. Decision trees are economic models that illustrate alternative decision options and their possible consequences. The decision trees were used to illustrate the patient pathway from suspected disease through to test outcome to distinguish accurate and inaccurate disease staging. The costs and diagnostic test accuracy of imaging devices were attributed to the appropriate branches in the trees, and then, dependent on the accuracy of the diagnostic test, a longer-term analysis followed to account for the costs, quality of life and survival impact of optimal versus received treatment.

Each model was analysed probabilistically, using Monte Carlo simulation, to determine the expected cost, outcomes (correct diagnoses and QALYs) and cost-effectiveness (cost per correct diagnosis and cost per QALY gained). The Monte Carlo simulations involved 2000 iterations for each model. The stability of the results was tested and found to be within reasonable bounds.

Primary disease

The evaluations considering the cost-effectiveness of FDG PET/CT as an add-on device in primary rectal (and primary colon) cancer relate to the initial pre-operative TNM staging of primary patients. The additional value of incorporating an FDG PET/CT scan to conventional imaging in this disease stage is through the identification of nodal and metastatic disease54,55 (clinical expertise). The only diagnostic test accuracy evidence available for FDG PET/CT in this context relates to the identification of lymph node involvement,54,55 and therefore the primary models were designed to evaluate the cost-effectiveness of FDG PET/CT as an add-on device in nodal staging.

Figure 4 depicts the basic decision tree structure used for the primary rectal (and colon) models. Because of the absence of economic models of FDG PET/CT in primary CRC in the literature (none was identified from the literature search), this model structure was informed primarily through consultation with clinical experts from the research team in order to accurately reflect the clinical pathway. The model was altered to include the disease-specific criteria for rectal and colon cancer separately. The model structure and parameters will be discussed in relation to rectal cancer, followed by a section detailing how and what parameters were altered for the colon model.

FIGURE 4. Staging primary CRC.

FIGURE 4

Staging primary CRC. AJCC, American Joint Committee on Cancer.

The decision tree model begins with patients who have had an initial assessment (involving a clinical examination, colonoscopy or sigmoidoscopy, and a biopsy) that identified them as having primary rectal cancer. The standard procedure for patients suspected of having primary rectal cancer is to use contrast-enhanced CT scans of the chest, abdomen and pelvis and an MRI scan of the pelvis to diagnose and/or stage the extent of the disease. This conventional pathway is represented in the top half of the tree. Alternatively, patients will receive the standard work-up (contrast-enhanced CT and MRI) followed by an additional FDG PET/CT scan, which is depicted in the bottom half of the tree. The primary decision tree model has been designed using actual CRC disease status, splitting the patient population according to the true disease status before having the imaging scans, so that accurate and inaccurate scan diagnosis can be identified. The objective of the scan in this model is to assess whether or not there is any nodal spread and, therefore, after the initial decision node depicting the choice between conventional or add-on FDG PET/CT tests, the tree divides the population according to actual nodal spread disease status using the American Joint Committee on Cancer (AJCC) CRC staging system.8 In the AJCC system, stages 1 and 2 have no nodal involvement, whereas stages 3 and 4 can have some nodal involvement. After dividing patients according to their true nodal spread disease status, the diagnostic tests are undertaken. Patients suspected of having rectal cancer who are in the conventional arm will have a contrast-enhanced CT scan of the chest, abdomen and pelvis and a pelvic MRI scan, which will identify either nodal involvement (test positive) or no nodal involvement (test negative). Having previously specified actual disease status, the top half of this branch represents primary rectal cancer with nodal spread (AJCC stages 3 and 4), and therefore the tree branch splits depending on whether the test was positive (accurately identified nodal involvement) or negative (inaccurately identifying no nodal involvement). These FN outcomes lead to inaccurate understaging, identifying no nodal involvement (AJCC 1 and 2) when the patients do have nodal involvement (AJCC 3 and 4). The bottom branch represents primary rectal cancer with no nodal spread (AJCC 1 and 2). The tree depicts the negative test outcomes that truly were negative (i.e. patients who are staged as AJCC 1 or 2 accurately) and also positive test outcomes that were inaccurate (FPs). These FP outcomes diagnose nodal involvement (AJCC 3 and 4), overstaging the extent of the disease, which is actually no nodal involvement (AJCC 1 and 2). In this way the decision tree separates out accurate and inaccurate diagnoses of nodal involvement.

Patients in the ‘conventional arm’ of the model will be staged using the standard diagnostic test work-up described above (contrast-enhanced CT of the chest, abdomen and pelvis and pelvic MRI), represented by ‘Test’ in the top half of the tree in Figure 4. Patients in the ‘intervention arm’ of the model are also given these conventional imaging tests, followed by the addition of a FDG PET/CT scan. This is represented in the bottom half of the tree, which has been abbreviated as its structure is identical to the structure of the top half. The diagnostic test accuracy of FDG PET/CT is influential as an add-on after the conventional test; therefore, in the model, a positive result from any of the scans incorporated in the ‘test’ strategy will be assumed to be a positive result (i.e. negative results from the conventional imaging tests that are refuted by the FDG PET/CT test are treated as positive). Results are treated as negative only when both the conventional and the FDG PET/CT test outcomes are negative.

The accurate and inaccurate nodal staging outcomes at the end of the decision tree branches for the conventional arm of the model (standard contrast-enhanced CT and MRI scans) are compared with those of the intervention arm of the model (standard contrast-enhanced CT and MRI scans plus FDG PET/CT) and assessed in terms of the incremental cost per accurate diagnosis.

These interim decision model outcomes of accurate and inaccurate diagnosis for the four AJCC stages were also used to undertake a longer-term analysis that assessed the impact of accurate and inaccurate staging on patient management, incorporating optimal treatments for each AJCC stage and measures of quality of life and overall survival, so that the conventional and intervention arms could also be compared in terms of the incremental cost per QALY gained.

The decision tree model is populated with parameters representing the prevalence of AJCC disease status and the diagnostic test accuracy of the conventional and intervention imaging devices and their associated costs. Table 15 details these parameters, along with the treatment, overall survival and quality of life parameters used in the longer-term analysis. The model parameters are now discussed including details of the longer-term modelling.

TABLE 15. Primary CRC model parameter estimates.

TABLE 15

Primary CRC model parameter estimates.

Primary colorectal cancer model parameters

Disease prevalence The model incorporated nodal spread disease status using prevalence data from a Scottish network data set145 provided by the clinical experts in the research team. The data comprise detailed clinico-pathological and imaging staging data from an ongoing prospective study involving 2838 Scottish CRC patients (average age 61 years). The data set is a prospective series that identifies all cases of CRC in Scotland by direct clinical and nurse contact, through pathology department returns, managed clinical networks, cancer registration and death registration. This series is considered to represent the generality of CRC in the UK, as any differences in the epidemiology of CRC between Scotland and the rest of the UK will be marginal. The data set provided information on CRC disease status using the AJCC CRC staging system along with 5-year overall survival data for each of the four AJCC stages. This data set is discussed in full in a recent publication detailing the population background characteristics and survival analysis outcomes.145 A previous analysis of a subset of the data set was published in 2006.147

The AJCC stage prevalence data were incorporated into the model under the assumption that AJCC stages 1 and 2 represent patients with no nodal involvement (n = 1518, 53%), and AJCC stages 3 and 4 represent patients with nodal involvement (n = 1320, 47%). The prevalence and number of patients in each AJCC stage in the data set are detailed in Table 15. For the probabilistic analysis, Dirichlet distributions (a multinomial version of the beta distribution) were assigned using the total number of patients and AJCC stage prevalence.

Having merged the AJCC data to distinguish disease in terms of nodal involvement to synchronise with the diagnostic test outcomes, the decision tree then separates the data back into the individual AJCC stages in the final branches, in order to assign treatment strategies for each AJCC stage in the longer-term model. This was done based on an assumption that the extent of disease in the model is linked to the overall stage prevalence [i.e. within the nodal involvement arm (which was calculated by summing AJCC 3 and 4 prevalence), the proportion who are separated back out to stage AJCC 3 was calculated by dividing the AJCC 3 prevalence (31%) by the total nodal involvement prevalence – AJCC 3 + AJCC 4 (47%)]. Following this assumption, patients who are inaccurately staged are done so according to that disease stage prevalence, i.e. patients who have nodal involvement (AJCC 3 and 4) but who are understaged through FN test results are inaccurately staged as either AJCC 1 or 2 based on AJCC 1 and 2 prevalence. AJCC 2 is more prevalent than AJCC 1; as such, this assumption ensures that in the model when FNs inaccurately understage patients as AJCC 2 and 1 (instead of AJCC 3 and 4), a greater proportion of patients will be inaccurately staged as AJCC 2 than inaccurately staged as AJCC 1. It is also more likely that an AJCC 3 patient would be understaged to AJCC 2 than to AJCC 1. The prevalence of AJCC 3 is greater than that of AJCC 4, and therefore a greater proportion of inaccurate overstaging will be attributed to AJCC 3 than to AJCC 4. Similarly, a patient with no nodal involvement (AJCC 1 or 2) will be more likely to be mistaken as an AJCC 3 patient than as an AJCC 4 patient.

Diagnostic test accuracy The systematic review for PET/CT in primary colorectal cancer (see Chapter 5), was intended to yield pooled data on diagnostic test accuracy for the main parameters in the economic model; however, only two papers were identified for FDG PET/CT in primary CRC,53,54 and there were inadequacies and reporting bias in the identified papers for all stages of CRC. Therefore, for the purpose of the economic analyses for primary CRC staging, papers identified by the systematic review were considered individually along with papers previously identified through the economic search; decisions were made to incorporate data that fit with the models.

With regards to the diagnostic test accuracy of contrast-enhanced CT, MRI and FDG PET/CT for staging primary CRC, the systematic review identified two papers53,54 that reported data for FDG PET/CT. The Tsunoda et al.53 data were reported only at a lesion level and were therefore not useful for the model; however, the Tateishi et al.54 paper (which compared FDG PET/CT with contrast-enhanced FDG PET/CT) reported patient-level data on the sensitivity and specificity of FDG PET/CT for staging nodal involvement and provided CIs. No distinction was made between colon and rectal cancer, and because of this and the lack of alternative information, the FDG PET/CT estimates were used in both models. The lower CI was used to calculate a standard error for use in the probabilistic analysis as it represented the widest range of uncertainty. We assumed an independent probability distribution for the sensitivity and specificity estimates, using beta distributions. Diagnostic test accuracy data for contrast-enhanced CT and MRI were taken from Bipat et al.,39 who undertook a meta-analysis in primary CRC and reported diagnostic test accuracy estimates with CIs for these imaging modalities for staging nodal involvement. The lower CI was used to calculate a standard error for use in the probabilistic analysis. FDG PET/CT was not included in this meta-analysis; however, as the authors detail the sensitivity and specificity of contrast-enhanced CT and MRI specifically for nodal involvement, it is reasonable to enter these estimates into the primary models, to compare with the addition of FDG PET/CT using the Tateishi et al. estimates specifically for staging nodal involvement.

The primary rectal model used the diagnostic test accuracy estimates for MRI to represent the ‘conventional’ imaging arm as, overall, it has superior test performance characteristics for lymph node involvement, i.e. both the sensitivity and the specificity of MRI are superior to those of contrast-enhanced CT.39 This approach of using superior test performance to represent joint imaging modalities has been used by others148 and is also reasonable given the evidence identified in the systematic review, which favoured MRI in the identification of nodal involvement. In the intervention arm, the diagnostic test accuracy for FDG PET/CT is added on after the conventional test, assuming that any outcome with a positive test is treated as such. Negative results from the conventional test that are refuted by the FDG PET/CT test are treated as positive. Results are treated as negative only when both the conventional and the FDG PET/CT test result are negative.

Treatments The economic models in our analyses were designed to incorporate the treatment impacts of accurate and inaccurate staging in primary CRC. The systematic review and the non-systematic economics search identified some literature on therapeutic impact and patient management in primary CRC.49,8993,102 This literature found that, although FDG PET/CT can have an impact in terms of more accurate staging of primary CRC, it had only a minor impact on changing patient management, as discussed in the therapeutic impact chapter of this report (see Chapter 8).

Optimal treatment combinations for each AJCC stage were determined through consideration of the literature49,8993,102 and in consultation with clinical experts. It was assumed that, for primary rectal cancer, all AJCC1 patients receive primary surgery and no further treatment. AJCC 2 and 3 primary rectal patients will receive one of three options: surgery alone, long-course CRT prior to surgery or surgery followed by adjuvant chemotherapy. AJCC 4 patients will receive one of five treatment options: primary surgery alone, long-course CRT prior to primary surgery, primary surgery followed by metastatic surgery, primary surgery followed by palliative care or palliative care alone. Primary surgery refers to rectal excision with lymphadenectomy; metastatic surgery refers to surgery at the metastatic site; and palliative care represents an array of palliative treatments that may include chemotherapy. These optimal treatment profiles inform the costing and the utility weights in the model.

The proportions of patients receiving each treatment for each stage (detailed in Table 15) were assigned in consultation with the clinical experts on the research team to ensure consistency with the data set used. These were also compared with publications reporting treatment and therapeutic impacts for primary rectal and colon cancer27,49,89,102,109 to ensure that important treatments were included. Uncertainty was incorporated through a series of Dirichlet distributions, one for each AJCC stage, specified using the AJCC stage prevalence data from the Scottish data set and the probabilities of receiving a specific treatment option within each stage.

Assigning these optimal treatment options for each AJCC stage in the longer-term model means that patients in the decision tree who are accurately diagnosed will receive optimal treatment, whereas patients who are inaccurately staged (through FP or FN test outcomes) will receive suboptimal treatment [i.e. patients with no nodal involvement (AJCC 1 or 2 patients) who are inaccurately diagnosed as having nodal involvement (overstaged to either AJCC 3 or 4) will receive unnecessary AJCC 3 or 4 treatments]. In the case of inaccurate staging, the model assumes that patients will receive the treatments for their (mis)diagnosed stage, but within a year their true diagnosis will be correctly identified and optimal treatment will then be given. This assumption was made in consultation with clinical experts and is considered to be valid with 1 year as an appropriate time scale for encompassing most cases of understaging. This way the model accounts for the appropriate treatments and the treatments that are received unnecessarily or that initially fail to be received because of over- or understaging. No transitions between nodal status are allowed during the year.

Survival The longer-term model incorporated overall survival in order to capture any potential impact on mortality over the lifetime of the patients.

The Scottish CRC network data set145 (2838 CRC patients, average age 61 years) detailed the 5-year overall survival of patients for each AJCC stage. These data were used to determine an annual mortality rate under the assumption of an exponential survivor function, used within a Markov simulation to estimate overall life expectancy for each AJCC stage. The Markov model assumed a starting age of 50 years and employed the mortality rate calculated from the data set for each stage for the first 10 years of the extrapolation; beyond 10 years patients were assumed to have survived their cancer and they were assumed to return to the average mortality rate for their age.149 [The starting age of 50 years was used in the model as the data set is based on the Scottish CRC population aged ≥ 50 years (mean age 61 years). The models were also run using an older population (starting age 70 years), with the resultant effect of lowering life expectancy and quality-adjusted life expectancy for patients in each AJCC stage, but with no overall change to the incremental cost-effectiveness outcomes.]

Publications121,122,139 indicate that FDG PET/CT scanning (in comparison with conventional imaging modalities) has no impact on overall survival; however, consultation with clinical experts highlighted that patients with AJCC 3 stage cancer (nodal involvement but no metastases) who fail to receive adjuvant chemotherapy because of inaccurate staging may suffer a reduction in overall survival. This was incorporated into the model for AJCC 3 patients who were inaccurately understaged as AJCC 1 or 2 as a 25% reduction in overall survival.

The 5-year overall survival estimates for each AJCC stage are detailed in Table 15, along with an estimated 25% reduction in overall survival for AJCC 3 patients who were inaccurately diagnosed and failed to receive adjuvant chemotherapy. Beta distributions were applied for the probabilistic analysis.

Quality of life/utility A measure of quality of life (utility) was incorporated into the model, capturing the average quality of life experienced by patients in each AJCC stage, and incorporating disutility experienced by patients who receive unnecessary treatment or who fail to receive optimal treatments because of inaccurate staging. Average quality of life estimates for each of the four CRC stages were derived from data reported in Ramsey et al.142 using the Health Utility Index. It was assumed that the average utility experienced by patients in a particular stage was constant for 5 years post diagnosis. Patients who were still alive 5 years post diagnosis were assigned age-specific utility weights based on UK population norms.143 The quality of life estimates were combined in the survival analysis and discounted at 3.5%150 to derive discounted quality-adjusted life expectancies for each AJCC stage.

During the 5-year post-diagnosis stage, patients who were correctly diagnosed in the model received the average utility for their state, whereas patients incorrectly diagnosed received their true disease stage utility, but with a disutility relating to the inappropriate treatment they received for a specified duration. It was assumed that patients who were inaccurately staged and who failed to receive either long-course CRT pre-surgery or adjuvant chemotherapy post surgery received a disutility for a 6-month duration, whereas patients who were inaccurately diagnosed and who failed to receive metastatic treatments were assumed to receive disutility for 1 year, reflecting the large impact on quality of life for delayed treatment. Patients who received unnecessary long-course CRT, unnecessary adjuvant chemotherapy or unnecessary metastatic surgery received a lower utility during their unnecessary treatment.

Table 15 details the utility and disutility weights used in the model. The probabilistic analysis applied gamma distributions on disutility to represent uncertainty in the parameters.

Costs The costs for the economic model are attributed to the cost of the alternative imaging devices (as a cost per scan) and the cost of the various treatment options for each AJCC stage. NHS reference cost data were used100,151 along with various other data sources for the AJCC stage treatment options.99,101 The various cost items are detailed in Table 16, specifying unit costs and standard errors. Where appropriate, normal distributions were used to represent the uncertainty surrounding cost estimates in the probabilistic analysis.

TABLE 16. Primary CRC model costs.

TABLE 16

Primary CRC model costs.

The cost of the imaging devices was incorporated as a cost per scan, representing staff time and use of the imaging machinery. Cost details regarding contrast-enhanced CT and MRI scans were available in NHS reference costs;100 however, no details were provided for the cost of FDG PET/CT scanning in either the Department of Health100 or the Scottish Information Services Division151 reference costs. Various studies report the cost of an FDG PET/CT scan in the UK as between £750 and £1000 per scan.103,151,152,155 It is also widely reported that FDG PET/CT scans generally have a duration of 20–40 minutes on equipment costing two to three times that of CT scanners, which can perform scans on a patient every 5–10 minutes;25 therefore, assigning a cost of £800 per FDG PET/CT scan seemed appropriate. A standard error for this baseline cost was derived using the upper and lower price range reported for an FDG PET/CT scan.152

The cost of primary rectal surgery (rectal excision with lymphadenectomy) includes the cost of a distal colon procedure, an average hospital inpatient stay of 6 days and CRC surgery consultant follow-up. Long-course CRT treatment consisted of radiotherapy given over 5 weeks (45 Gy in 25 fractions) combined with a 12-week course of chemotherapy – intravenous 5-fluorouracil.154,155 The adjuvant chemotherapy treatment consisted of a 6-month course of intravenous 5-fluorouracil plus oxaliplatin.99,155 The cost of metastatic surgery was represented by the Information Services Division151 cost of surgical specialties in medical oncology, representing the cost of surgery, including theatre time, surgical consultation and follow-up, and an average inpatient stay of 10 days. Resource use and costs for palliative care were taken from a study that assessed the cost to the NHS of palliative care in CRC.124 The costs of palliative care were reported at price year 2000/1, and therefore the hospital and community health services pay and price index101 was used to adjust this to price year 2009.

The average cost per AJCC stage was calculated using the proportion of patients receiving each treatment option within each AJCC stage. In the model, if a patient was staged accurately, he or she would receive his or her optimal treatment option and be assigned the average cost of treatment for that stage. The model also incorporates the extra costs incurred through inaccurate staging. If a patient is inaccurately diagnosed, he or she incurs the cost of the misdiagnosed treatment, followed by the discounted cost of treatment for his or her true stage the following year (i.e. it is assumed that the true disease stage will be identified within a year). Costs were discounted at 3.5%.150

Primary colon model

The basics of the primary model structure were the same for both rectal and colon cancer. The parameters discussed above relate to rectal cancer; however, Tables 15 and 16 also detail the parameters that were used in the colon model. The specific aspects of the model that were altered for the colon model are discussed below.

Conventional imaging and diagnostic test accuracy Magnetic resonance imaging is not used in the assessment of primary colon cancer and therefore the primary colon model incorporated only contrast-enhanced CT as the conventional imaging modality. As previously discussed, the diagnostic test accuracy literature made few distinctions between colon and rectal cancer, and therefore, because of this and the lack of alternative information, the FDG PET/CT and contrast-enhanced CT estimates were used in both models.

The intervention arm adopts the same approach to that described earlier, whereby the diagnostic test accuracy of FDG PET/CT is added on to the contrast-enhanced CT test outcomes, and any outcome with a positive test is treated as such. Negative results from the conventional test that are refuted by the FDG PET/CT test are treated as positive. Results are treated as negative only when both the contrast-enhanced CT and the FDG PET/CT test results are negative.

Treatments The AJCC treatment options for the colon model vary slightly. Primary surgery refers to a colonic resection with lymphadenectomy. The other major treatment change is that pre-operative long-course CRT is not used to treat primary colon cancer, and therefore the treatment options for AJCC stages 2, 3 and 4 were modified for the colon model. In the colon model, AJCC 2 and 3 primary colon patients receive one of two options: surgery alone or surgery followed by adjuvant chemotherapy. Patients with AJCC 4 disease may receive one of four treatment options: primary surgery alone, primary surgery followed by resection of metastases, primary surgery followed by palliative care or palliative care alone. [There is typically a fifth treatment option of primary surgery combined with concomitant resection of metastases; however, after consultation with clinical experts it was decided that this additional treatment option would result only in an unquantifiable and likely marginal effect on overall cost compared with primary surgery alone, hence these are considered together.] The same approach that was used for the rectal cancer model was used to determine the optimal treatment combination within each AJCC stage for colon cancer, i.e. through the literature and in consultation with clinical experts. Table 15 details the within-stage distributions that were assigned to each stage in the colon model.

Survival and quality of life There was no change to the survival analysis for the colon model; however, the utilities were amended to exclude the CRT-related utilities used in the rectal model. In the case of inaccurate staging, disutilities were still applied but were specifically for failing to receive adjuvant chemotherapy, or for patients who receive unnecessary adjuvant chemotherapy.

Costs The costs were amended in line with the parameter modifications discussed above. In the colon model, conventional imaging involves only contrast-enhanced CT, and therefore it is the only imaging cost incorporated for the conventional arm (the costs of MRI are excluded). With regards to the treatment costs, the cost of primary surgery refers to a colonic resection. Cost data for a proximal colonic procedure, a hospital inpatient stay of 6 days and CRC surgery consultant follow-up were included.100,101 The average cost of treatment per AJCC stage was calculated for the colon model in the same manner as for the rectal model, using the proportion of patients receiving each treatment option within each AJCC stage to calculate the average cost per AJCC stage.

Scenario analysis: contrast-enhanced FDG PET/CT as a lone technology

The previous chapters in this report found suggestions within the literature that in the future, as FDG PET/CT technology improves (i.e. with the development and introduction of contrast-enhanced FDG PET/CT scanners), it may be possible to use these higher-quality devices as an alternative to CT or contrast-enhanced CT in primary CRC rather than using FDG PET/CT as an add-on imaging device.

Although the scope of the current research was focused on FDG PET/CT as an add-on device, we have included a scenario analysis for the primary colorectal models in which contrast-enhanced FDG PET/CT is used as a replacement for conventional contrast-enhanced CT, rather than as an add-on device. The Tateishi et al. paper,54 which provided diagnostic test accuracy evidence for FDG PET/CT, also provided patient-level diagnostic test accuracy estimates for contrast-enhanced FDG PET/CT in nodal staging (with equivalent sensitivity to FDG PET/CT but improved specificity as reported in the diagnostic test accuracy tables for primary CRC in Table 2). These contrast-enhanced diagnostic test accuracy estimates and CIs were used in the scenario analysis to portray the future potential of improved FDG PET/CT imaging. For the primary rectal scenario, the conventional strategy (contrast-enhanced CT followed by MRI) was compared with a contrast-enhanced FDG PET/CT replacement strategy (contrast-enhanced FDG PET/CT followed by MRI); and for the primary colon scenario, conventional contrast-enhanced CT was compared with contrast-enhanced FDG PET/CT alone. All model parameters remain as above, with the exception of the diagnostic test accuracy estimates and the cost of contrast-enhanced FDG PET/CT. The contrast-enhanced FDG PET/CT diagnostic test accuracy estimates and CIs were used,54 and a cost for the contrast-enhanced FDG PET/CT scan was incorporated, assuming an increase of 20% to the FDG PET/CT scan cost to reflect the cost of this more expensive technology.

Recurrent disease

The recurrent model evaluations have been undertaken to assess the cost-effectiveness of FDG PET/CT as an add-on device in detecting recurrent rectal (and recurrent colon) cancer. The value of incorporating an FDG PET/CT scan in addition to conventional imaging in this disease stage is through the ability to confirm or refute local recurrence and potentially identify metastatic recurrence.

Figure 5 depicts the decision tree structure used for the recurrent rectal (and colon) models. This was altered to include the disease-specific criteria for rectal and colon cancer separately. The model structure was informed by the literature58,110 and was based on consultation with clinical experts. The parameters will be discussed in relation to rectal cancer, followed by a section detailing what elements and parameters were altered for the colon model.

FIGURE 5. Staging recurrent CRC.

FIGURE 5

Staging recurrent CRC.

The recurrent decision tree model begins with patients who have previously had surgical treatment for primary rectal cancer and who, in a routine follow-up assessment (involving clinical examination, routine imaging and CEA testing), were found to have rising CEA levels, which identified them as potentially having recurrent rectal cancer. The decision tree then outlines the choice between conventional diagnostic testing and the add-on FDG PET/CT strategy. The standard procedure for patients suspected of recurrent rectal cancer involves contrast-enhanced CT scans of the chest, abdomen and pelvis and an MRI scan of the pelvis to confirm or refute rectal local recurrence and assess whether this is an isolated recurrence or associated with distant metastases. Similar to the structure used in the primary models, this decision tree model has been designed using actual disease status, and therefore the decision tree has split the patient population according to their true status before having the imaging scans, so that accurate and inaccurate diagnoses can be identified. The objective of the scan in this model is to assess whether or not there has been any recurrent disease, and therefore the tree divides into the recurrence (isolated local or local combined with distant metastases) and no recurrence populations. The standard work-up of diagnostic tests is then undertaken. Patients suspected of rectal recurrence in the conventional arm will have a contrast-enhanced CT scan of the chest, abdomen and pelvis and a pelvic MRI scan, which will identify either recurrence (test positive) or no recurrence (test negative). Having previously specified actual recurrence status, the top branch of the tree represents recurrent cancer, and therefore the tree branch splits depending on whether the test was positive (accurately identifying recurrence) or negative (inaccurately identifying no recurrence). Positively identified recurrence is then further separated into curable and non-curable recurrence, which will involve different treatment options in the longer-term model. Negative test outcomes indicate no recurrence and thus no treatment. In the top half of this branch, negative test outcomes represent FNs, which lead to patients being inaccurately diagnosed as having no recurrence. For the longer-term model it is assumed that patients will be accurately re-staged within a year.

The bottom branch in the top half of the tree represents the actual status of no recurrence, so negative test outcomes accurately indicate no recurrence. Positive test outcomes in this branch of the tree are FPs, which inaccurately diagnose recurrence when there is none. This population is further divided into curable and non-curable recurrence in order to determine what treatment patients receive unnecessarily in the longer-term model. In this way the decision tree separates out accurate and inaccurate diagnosis of recurrence.

Patients in the ‘conventional arm’ of the model will be staged using the standard diagnostic test work-up described above (contrast-enhanced CT of the chest, abdomen and pelvis and pelvic MRI), represented by ‘Test’ in Figure 5. Patients in the ‘intervention arm’ of the model will be given these conventional imaging tests, followed by the addition of a FDG PET/CT scan. This is represented in the bottom half of the tree, which has been abbreviated as the structure is identical to that of the top half.

The accurate and inaccurate identification of recurrence at the end of the decision tree branches for the conventional arm of the model (standard contrast-enhanced CT and MRI scans) are compared with the accurate and inaccurate identification of recurrence in the intervention arm of the model (standard contrast-enhanced CT and MRI scans plus the addition of FDG PET/CT) and assessed in terms of the incremental cost per accurate diagnosis. These interim outcomes of accurate and inaccurate diagnosis are then used to undertake a longer-term analysis that assesses the impact of accurate and inaccurate diagnoses of recurrence on patient management, incorporating optimal treatments for curable recurrence, non-curable recurrence and no recurrence, and modelling the impacts on quality of life and overall survival. In this way, the conventional and intervention arms can be compared in terms of the incremental cost per QALY gained.

This decision tree model is populated with parameters representing the prevalence of recurrent CRC and the diagnostic test accuracy of the conventional and intervention imaging devices for staging recurrent rectal (and colon) cancer and their associated costs. Table 17 details these parameters, along with the treatment, overall survival and quality of life parameters used in the longer-term analysis. The model parameters are discussed below, including details of the longer-term modelling.

TABLE 17. Recurrent model parameter estimates.

TABLE 17

Recurrent model parameter estimates.

Recurrent colorectal cancer model parameters

Disease prevalence The literature identified in the economics search and the systematic review was used to provide disease prevalence evidence for the recurrent model. Disease prevalence data on recurrence in CRC were based on estimates provided by Saunders et al.,123 assigning a 30% probability of local recurrence and a 40% probability of metastatic recurrence for patients previously treated for primary CRC. It was assumed that a cohort of patients who were diagnosed as AJCC 1, 2 or 3 for primary CRC would be susceptible to recurrence. Using the Scottish network CRC data set145 to represent this cohort and assigning the probability of recurrence from Saunders et al.,123 it was possible to apply Dirichlet distributions around these baseline estimates to incorporate uncertainty in the probabilistic analysis. Table 17 details these parameters.

This model structure is similar to the structure used in two other economic evaluations that assessed the value of using PET in the identification of recurrent CRC.58,110 These two models also incorporated patient management and quality of life impacts by including a probability of curable and non-curable recurrence in the recurrent population.58,110 Table 17 details the parameters, standard errors and probability distributions applied.

Diagnostic test accuracy As reported in Chapter 6, a meta-analysis was undertaken using relevant papers identified from the systematic review to elicit pooled diagnostic test accuracy estimates of FDG PET/CT for recurrent CRC. As noted above, because of inadequacies and reporting bias in the identified papers, the pooled estimates for FDG PET/CT may not be an accurate reflection of the diagnostic test accuracy of FDG PET/CT. The pooled estimates give tight CIs that do not fully represent the wide uncertainty in the mean estimates. Therefore, papers identified by the systematic review were considered individually along with papers identified through the economic search to find reasonable estimates of diagnostic test accuracy for the economic models with wide uncertainty intervals.

Three papers provided diagnostic test accuracy evidence of FDG PET/CT as an add-on device for diagnosis of recurrent CRC.65,81,87 Schmidt et al.65 compared FDG PET/CT with whole-body MRI but, as reported in the systematic review chapter, there appeared to be reporting bias with this study. In addition, the diagnostic test accuracy for whole-body MRI was inappropriate for the model, which incorporates pelvic MRI rather than whole-body MRI. Ramos et al.81 provide evidence for contrast-enhanced CT in comparison with FDG PET/CT, but the point estimates assigned appear to be biased in favour of FDG PET/CT (reporting a sensitivity of zero for contrast-enhanced CT, but with a CI range up to 0.65). Selzner et al.87 provide diagnostic test accuracy evidence for contrast-enhanced CT in comparison with FDG PET/CT; however, they do not report any CIs or other measures of uncertainty. The point estimates from Selzner et al.87 were deemed to be the best reflection of mean diagnostic test accuracy and were therefore used in the model along with the wide CIs from Ramos et al.81 to ensure a suitably wide range to reflect the considerable uncertainty surrounding the mean diagnostic test accuracy estimates. The pooled meta-analysis diagnostic test accuracy estimates had more restrictive confidence limits and were therefore deemed inappropriate to accurately reflect uncertainty in the economic models. There were no reliable estimates of pelvic MRI diagnostic test accuracy for recurrent CRC reported; therefore, an estimate was taken from the diagnostic test accuracy of MRI used in the Park et al.28 economic evaluation. Diagnostic test accuracy estimates, their standard errors and the distributions used in the probabilistic model are detailed in Table 17.

Treatments The recurrent CRC models were designed to incorporate the treatment impacts of accurate and inaccurate diagnoses of recurrent CRC.

Optimal treatment combinations for curable and non-curable recurrence were determined through the literature and in consultation with clinical experts. The model assumed that 40% of recurrent rectal cancer patients would have received radiotherapy as part of their treatment for primary cancer and therefore would not receive further radiotherapy, while the remaining 60% of those patients who subsequently developed local recurrence but who did not receive radiotherapy for their primary cancer would receive long-course CRT prior to surgery for recurrent disease. Patients with curable recurrence had one of six treatment options: local surgery alone, local surgery followed by adjuvant chemotherapy, long-course CRT prior to local surgery, local surgery followed by metastatic surgery, local surgery and adjuvant chemotherapy followed by metastatic surgery or long-course CRT prior to local surgery followed by metastatic surgery. Patients with non-curable recurrence had one of two treatment options: metastatic surgery followed by palliative care or palliative care alone. It was assumed that all patients with no recurrence would be treated with a wait and watch strategy in which they would be followed up annually.

The proportions of patients receiving each treatment for curable recurrence were assigned based on consultation with clinical experts and publications reporting treatment and therapeutic impacts for recurrent CRC.71,93,102 The Scottish network CRC data set145 was used to derive a cohort of patients (AJCC 1–3) who would be susceptible to colorectal recurrence. A subset of this population was deemed to have curable recurrence, using the probability assigned in the model. A Dirichlet distribution was applied to this subset to capture the uncertainty surrounding the treatment allocation. The proportions of patients receiving each treatment for incurable recurrence were informed by the literature and previous economic models for recurrent CRC.58,110 Beta distributions were applied to these estimates as there were only two options. Table 17 details the treatment options and the proportions.

Survival The longer-term model incorporated overall survival in order to capture any potential impact on mortality over the lifetime of the patients.

The survival analysis was implemented employing an approach similar to that used in the primary model. Five-year overall survival estimates were determined from the literature for patients with no recurrence, recurrence that is curable and non-curable recurrence.119 In addition, it was assumed that patients in the model who had curable recurrence but were inaccurately diagnosed and failed to receive treatment in the first year were assigned a 5-year overall survival midway between curable and non-curable survival estimates. These data were used to determine an annual mortality rate under the assumption of an exponential survivor function, and used within a Markov simulation to estimate overall life expectancy for no recurrence, curable recurrence, non-curable recurrence and curable recurrence when treatment is delayed. The Markov model assumed a starting age of 50 years and employed the mortality rate calculated for each group for the first 10 years of the extrapolation; beyond 10 years patients were assumed to have survived their cancer and were assumed to return to the average mortality rate for their age.149

The cohort population of AJCC 1–3 patients derived from the Scottish network CRC data set145 was used to represent the recurrent model population.

The 5-year overall survival estimates for each of the model groups are detailed in Table 17. Beta distributions were applied for the probabilistic analysis.

Quality of life/utility Utility estimates were incorporated into the model, representing the average quality of life for patients in the no recurrence, curable recurrence and non-curable recurrence groups. Patients who were inaccurately diagnosed as no recurrence (FN test outcomes) and who failed to receive either curable or non-curable treatment in the first year were assigned a disutility for that year to account for the negative impact on their quality of life. Likewise, patients who were inaccurately diagnosed as having recurrent cancer (FP test outcomes) and who received unnecessary curative or non-curative treatments were assigned a lower utility status for that year to account for the negative impact of unnecessary treatment on quality of life.

The utility estimates were incorporated into the survival analysis as described previously in the primary models. It was assumed that the average utility experienced by patients in a particular stage was constant for 5 years post diagnosis. Patients who were still alive 5 years post diagnosis were assigned age-specific utility weights based on UK population norms.143 The quality of life estimates were combined in the survival analysis and discounted at 3.5%150 to derive discounted quality-adjusted life expectancies for each of the model groups (no recurrence, curable recurrence and non-curable recurrence).

Table 17 details the utility and disutility weights used in the model. The probabilistic analysis applied gamma distributions on disutility to represent uncertainty in the parameters.

Costs As in the primary models, the costs for the recurrent models are attributed to the alternative imaging devices (as a cost per scan) and the various treatment options assigned in the model. The various costs used in the recurrent models are detailed in Table 18, specifying unit costs, standard errors and the distributions used in the probabilistic analysis.

TABLE 18. Recurrent model costs.

TABLE 18

Recurrent model costs.

The costs of the imaging devices are the same as those used in the primary models. The treatment option combinations for the recurrent rectal model are different from those in the primary models; however, the costs of the component treatments were assigned in the same way. For example, the cost of recurrent rectal surgery was taken to be the same as the cost of primary surgery (summing the cost of a distal colon procedure, a 6-day hospital inpatient stay and CRC surgery consultant follow-up). The costs of long-course CRT treatment, adjuvant chemotherapy treatment, metastatic surgery and palliative care was also determined by the same means as in the primary models.

The expected costs in the no recurrence, recurrence curable and recurrence non-curable groups were calculated using the proportions of patients receiving each treatment option within each group. In the model, a patient who was diagnosed accurately would receive the optimal treatment option and incur the associated costs of that treatment. A patient who was inaccurately diagnosed would incur the cost of the diagnosed group treatment, followed by the discounted cost of treatment for his or her true diagnosis the following year (i.e. it is assumed that the true diagnosis would be identified within a year). Costs were discounted at 3.5%.150

Recurrent colon model

The structure of the recurrent models was the same for both rectal and colon cancer. The parameters discussed above relate to rectal cancer; however, Tables 17 and 18 also detail the parameters that were used in the colon model. The specific aspects of the model that were altered for the colon model are discussed below.

Disease prevalence Some publications indicate that local recurrence in rectal cancer is more common than local recurrence in colon cancer; however, data for the UK indicate only a very small difference in local recurrence for rectal and colon cancers.155 Therefore, both the rectal and colon recurrent models assumed the same probability of recurrence. Both models incorporated a measure of the uncertainty around this estimate, which was applied in the probabilistic analysis.

Conventional imaging and diagnostic test accuracy The MRI imaging device is not used in the assessment of colon cancer and therefore the recurrent colon model incorporates only contrast-enhanced CT as the conventional imaging modality. The diagnostic test accuracy estimates for contrast-enhanced CT and FDG PET/CT were determined as discussed above for the recurrent rectal model.

Treatments The treatment options for the recurrent colon model vary slightly from those in the recurrent rectal model.

As with the primary colon model, CRT is not included as a treatment option for recurrent colon cancer. Patients with curable colon recurrence had one of four treatment options: local surgery alone, local surgery followed by adjuvant chemotherapy, local surgery followed by metastatic surgery, or local surgery and adjuvant chemotherapy followed by metastatic surgery. Patients with non-curable recurrence and no recurrence had the treatment options detailed above for the recurrent rectal model. The same approach was used to determine the optimal treatment combinations within curable colon recurrence as was used for the recurrent rectal model, i.e. through the literature and in consultation with clinical experts. Table 17 details the within-stage distributions that were assigned to each treatment group in the colon model.

Survival and quality of life There was no change to the survival analysis or to the quality of life parameters for the colon model.

Costs The costs were amended in line with the parameter modifications discussed above. Because MRI is not used for colon cancer, in the recurrent colon model the cost of conventional imaging incorporates only contrast-enhanced CT, and the cost of the intervention arm incorporates only contrast-enhanced CT plus FDG PET/CT. With regards to the treatment costs, the cost of recurrent local surgery refers to a colonic resection. Cost data for a proximal colonic procedure, a 6-day hospital inpatient stay and CRC surgery consultant follow-up were included.100,101 The average cost of treatment for each disease group (no recurrence, recurrence curable and recurrence non-curable) was calculated for the colon model in the same manner as for the rectal model, using the proportions of patients receiving each treatment option within each group.

Metastatic model

The metastatic model was undertaken to assess the cost-effectiveness of FDG PET/CT as an add-on device in detecting metastatic cancer. The added value of incorporating an FDG PET/CT scan in addition to conventional imaging in this disease stage is in its ability to detect unsuspected, metastatic disease and potentially identify unsalvageable extra metastases not detected by conventional imaging devices.

Figure 6 depicts the decision tree structure used for the metastatic model, informed by the literature51,109,111 and based on consultation with clinical experts. The metastatic decision tree begins with patients who have previously had surgical treatment for primary CRC and in a routine follow-up assessment (involving a clinical examination and CEA testing) were found to have rising CEA levels, and were identified as potentially having a metastatic recurrence. The decision node depicts the choice between the conventional or add-on FDG PET/CT arms. Similar to the structure used in the previous models, this decision tree has been designed using actual disease status, and therefore the decision tree has split the patient population according to their true disease status (metastatic recurrence or no metastatic recurrence) prior to applying the diagnostic test accuracy estimates for the tests, so that accurate and inaccurate diagnoses can be identified.

FIGURE 6. Staging metastatic recurrence.

FIGURE 6

Staging metastatic recurrence.

The conventional procedure for patients suspected of metastatic recurrence is to undertake a contrast-enhanced CT scan of the chest, abdomen and pelvis to confirm or refute metastatic recurrence and potentially identify additional sites of metastases. This scan will identify either metastases (test positive) or no metastases (test negative). In the conventional arm, having specified actual disease status, the top half of this branch represents metastatic recurrence, and therefore the tree branch splits depending on whether the test was positive (accurately identifying metastatic recurrence) or negative (inaccurately identifying no metastatic recurrence). Positive identification of metastatic recurrence is further separated in this model to distinguish between metastases at one site or extra metastases at numerous sites, as the extent of the metastatic recurrence will affect the treatment options in the longer-term model. The negative test outcomes in the top branch of the decision tree indicate a misdiagnosis of no metastatic recurrence (FN). For the longer-term model it is assumed that patients will be accurately re-staged within a year.

The bottom half of the conventional tree branch represents the status of no metastatic recurrence, so negative test outcomes accurately indicate no metastases. Positive test outcomes in the bottom half of the tree are FPs, which inaccurately diagnose metastatic recurrence when there is no recurrence. This population is then further divided to distinguish between inaccurate diagnosis of metastases at one site and inaccurate diagnosis of extra metastases at numerous sites. In this way the decision tree separates out accurate and inaccurate diagnoses of metastatic recurrence.

Patients in the ‘conventional arm’ of the model will be staged using the standard diagnostic test (contrast-enhanced CT of the chest, abdomen and pelvis), represented by ‘Test’ in the top half of Figure 6. Patients in the ‘intervention arm’ of the model will also be given the contrast-enhanced CT scan, followed by the addition of an FDG PET/CT scan. This is represented in the bottom half of the tree, but these branches have been abbreviated as the structure is identical to that of the top half.

The accurate and inaccurate identification of metastases at the end of the decision tree branches for the conventional arm of the model is compared with the accurate and inaccurate identification of metastases in the intervention arm and assessed in terms of the incremental cost per accurate diagnosis. These interim outcomes of accurate and inaccurate diagnosis are then used to undertake a longer-term analysis that assesses the impact of accurate and inaccurate diagnosis of metastases on patient management, incorporating optimal treatments for metastases at one site, extra metastases and no metastatic recurrence, and modelling the impacts on quality of life and overall survival. In this way the conventional and intervention arms can be compared in terms of the incremental cost per QALY gained.

This decision tree model is populated with parameters representing the prevalence of metastatic recurrence, the diagnostic test accuracy of contrast-enhanced CT and FDG PET/CT and the probability of having extra metastases (at more than one site). Table 19 details these parameters, along with the treatment, overall survival and quality of life parameters used in the longer-term analysis. The model parameters, including details of the longer-term modelling, are discussed below.

TABLE 19. Metastatic model parameter estimates.

TABLE 19

Metastatic model parameter estimates.

Metastatic parameters

Disease prevalence The literature identified in the economics search and the systematic review was used to provide disease prevalence evidence for the metastatic model. Estimates provided by Saunders et al.123 were used for the prevalence of metastatic recurrence for patients previously treated for primary CRC. It was assumed that a cohort of patients who were diagnosed as AJCC 1, 2 or 3 for primary CRC would be susceptible to metastatic recurrence. Using the Scottish network CRC data set145 to represent this cohort, and assigning the probability of recurrence from Saunders et al.,123 it was possible to apply a Dirichlet distribution to represent the uncertainty around the prevalence of metastatic recurrence in the probabilistic analysis. Table 19 details these parameters.

The model structure for this evaluation is similar to that used by previous economic evaluations assessing the cost-effectiveness of using add-on FDG PET/CT in the identification of metastatic disease.29,51 Previous models have attempted to incorporate patient management and quality of life impacts by distinguishing between resectable and unresectable metastases51 or by distinguishing between hepatic and extra metastases.29 Our evaluation distinguished between metastases at one site, and at multiple sites (extra metastases). Assigning a probability for each in the overall metastatic recurrence population. In this way the model could distinguish between metastatic (at one site) and extra-metastatic (at more than one site) disease, even though the diagnostic test accuracy estimate referred only to the identification of metastases.

Diagnostic test accuracy As reported in Chapter 8, a meta-analysis was undertaken using relevant papers identified from the systematic review to elicit pooled diagnostic test accuracy estimates of FDG PET/CT for metastatic CRC. Because of inadequacies and reporting bias in the identified papers, these pooled estimates for FDG PET/CT may not be an accurate reflection of the mean diagnostic test accuracy. The CIs for the pooled estimates were also tight around the pooled mean, restricting the level of uncertainty represented. Therefore, the meta-analysis of diagnostic test accuracy data was deemed to be inappropriate for use in the economic model, and papers identified by the systematic review were considered individually, along with papers previously identified through the economic search, to find reasonable estimates of diagnostic test accuracy for the economic models.

Four papers provided diagnostic test accuracy evidence of FDG PET/CT in comparison with contrast-enhanced CT for diagnosing metastatic recurrence.79,80,83,87 These diagnostic test accuracy papers were all deemed to be of poor quality and suffering from the reporting bias discussed in the systematic review (see Chapter 8). After considering the available evidence, the economic model incorporated the diagnostic test accuracy evidence for contrast-enhanced CT and FDG PET/CT from the Chua et al. paper,79 with an adjustment to the (low) point estimate for the specificity of contrast-enhanced CT and incorporating a wide range for the uncertainty based on the CI data from Selzner et al.87 Diagnostic test accuracy estimates, their standard errors and the distributions used in the probabilistic model are detailed in Table 19.

Treatments The metastatic model was designed to incorporate the treatment impacts of accurate and inaccurate diagnosis of metastatic recurrence.

Treatment combinations for metastatic recurrence at one site, extra metastases and no metastatic recurrence were determined from the literature. Although extreme, the model assumes that all patients with metastases at a single site will receive pre-operative chemotherapy and metastatic surgery. Similarly, taking an extreme position, patients with extra metastases are assumed to be non-curable and will receive one of two treatment options: pre-operative chemotherapy followed by metastatic surgery and palliative care, or chemotherapy and palliative care. It was assumed that all patients identified as having no metastatic recurrence would be treated with a wait and watch strategy in which they would be followed up annually.

The proportions of patients receiving each of the two treatment options for extra metastases were determined from previous economic evaluations58,110 and uncertainty was represented by a beta distribution. Table 19 details these treatment options.

Survival The longer-term model incorporated overall survival to capture any potential impact on mortality over the lifetime of the patients.

The survival analysis was implemented employing an approach similar to that used in the primary and recurrent models. Five-year overall survival estimates were determined from the literature for patients with no metastatic recurrence and patients with metastases at one site.119,120 Patients with extra metastases were assigned different 5-year overall survival estimates dependent on the type of treatment that they received, i.e. patients with extra metastases who received metastatic surgery with palliative intent had a greater 5-year survival estimate than patients with extra metastases who received palliative care alone.119,120 These data were used to determine an annual mortality rate under the assumption of an exponential survivor function, and used within a Markov simulation to estimate overall life expectancy for each group. The Markov simulation assumed a starting age of 50 years and employed the mortality rate calculated for each group for the first 10 years of the extrapolation; beyond 10 years patients were assumed to have survived their cancer and to return to the average mortality rate for their age.149

The cohort population of AJCC 1–3 patients derived from the Scottish network CRC data set145 was used to represent the metastatic model population.

The 5-year overall survival estimates for each of the model groups are detailed in Table 19. Beta distributions were applied for the probabilistic analysis.

Quality of life/utility Utility estimates were incorporated into the model, representing the average quality of life for patients in the no metastatic recurrence, metastases at one site and extra metastases (at numerous sites) groups. Patients who were inaccurately diagnosed as no metastatic recurrence (FNs) and who therefore failed to receive treatment for metastases at either one or more than one site in the first year were assigned a disutility for that year to account for the negative impact on their quality of life. Likewise, patients who were inaccurately diagnosed as having metastases (FPs) and who received unnecessary metastatic surgery or treatments for extra metastases were assigned a lower utility status for that year to account for the negative impact of unnecessary treatment on their quality of life.

The utility estimates were incorporated into the survival analysis as previously described for the recurrent model. It was assumed that the average utility experienced by patients in a particular stage was constant for 5 years post diagnosis. Patients who were still alive 5 years post diagnosis were assigned age-specific utility weights based on UK population norms.143 The quality of life estimates were combined in the survival analysis and discounted at 3.5%150 to derive discounted quality-adjusted life expectancies for each of the groups (no metastatic recurrence, metastases and extra metastases).

Table 19 details the utility and disutility weights used in the model. The probabilistic analysis applied gamma distributions on disutility to represent uncertainty in the parameters.

Costs As in the recurrent model, the costs for the metastatic model are attributed to the alternative imaging devices (as a cost per scan) and the various treatment options assigned in the model. The various costs used in the metastatic model are detailed in Table 20, specifying unit costs, standard errors and the distributions used in the probabilistic analysis.

TABLE 20. Metastatic model costs.

TABLE 20

Metastatic model costs.

The costs of the imaging devices are the same as those used in the previous models. The treatment option combinations for the metastatic model are different to those in the primary and recurrent models; however, the costs of the component treatments were assigned in the same way. For example, the costs of metastatic surgery, palliative care and pre-operative chemotherapy were determined by the same means used in the primary and recurrent models.

The expected costs of treatment for the groups were calculated using the proportions of patients receiving each treatment option within each group. In the model, if a patient was diagnosed accurately, he or she would receive the optimal treatment option and incur the associated costs of that treatment. If a patient was inaccurately diagnosed, he or she would incur the cost of the treatment for the (mis)diagnosed group, followed by the discounted cost of treatment for his or her true diagnosis the following year (i.e. it is assumed that the true diagnosis would be identified within a year if the patient were still alive). Costs were discounted at 3.5%.150

Results

Primary rectal cancer model

Table 21 details the expected costs of the imaging involved in the conventional and the intervention test strategies, the expected probability of a correct diagnosis under each strategy and the cost-effectiveness in terms of cost per correct diagnosis for primary rectal cancer. The addition of FDG PET/CT was dominated by the conventional strategy, i.e. FDG PET/CT was both more expensive and less effective.

TABLE 21. Primary rectal cancer – cost per correct diagnosis.

TABLE 21

Primary rectal cancer – cost per correct diagnosis.

Table 22 details the expected costs of the imaging and treatment associated with the conventional and the intervention test strategies, the expected outcomes in terms of QALYs under each strategy and the cost-effectiveness in terms of cost per QALY gain for primary rectal cancer. On this basis, the addition of FDG PET/CT to the conventional strategy involved an additional cost of approximately £432,000 per QALY gained and would not be considered cost-effective under the usual definition [£20,000 per QALY < incremental cost-effectiveness ratio (ICER) < £30,000 per QALY].150

TABLE 22. Primary rectal cancer – cost per QALY gain.

TABLE 22

Primary rectal cancer – cost per QALY gain.

Figure 7 illustrates the uncertainty surrounding the expected incremental costs and incremental QALYs for primary rectal cancer. The figure shows that there was considerable uncertainty about the extent, but not the existence, of the additional expected costs (shown in the vertical plane) and the existence and extent of the additional expected QALYs (shown in the horizontal plane).

FIGURE 7. The cost-effectiveness plane for FDG PET/CT in primary rectal cancer.

FIGURE 7

The cost-effectiveness plane for FDG PET/CT in primary rectal cancer.

The cost-effectiveness acceptability curve (CEAC) (Figure 8) illustrates the uncertainty in the cost-effectiveness estimate for primary rectal cancer. The CEAC shows the probability that FDG PET/CT was cost-effective as an add-on imaging device in comparison to CT and MRI at different values for the maximum acceptable cost-effectiveness ratio (λ). Figure 8 shows that, at a monetary threshold of < £100,000, the probability that the addition of FDG PET/CT was cost-effective is < 20%. Within the usual range of values for the maximum acceptable cost-effectiveness ratio (λ), the CEAC illustrates that the conventional CT and MRI devices have a (approximately) 100% probability of being cost-effective and the FDG PET/CT intervention has a (approximately) 0% probability of being cost-effective.

FIGURE 8. The CEAC for primary rectal cancer.

FIGURE 8

The CEAC for primary rectal cancer.

The expected value of perfect information (EVPI) analysis shows that, at a willingness-to-pay threshold of £30,000 per QALY, the EVPI per decision is < £2. To determine the overall population value of EVPI we assumed an annual incidence of 13,315156 cases and a time frame of 2 years (i.e. FDG PET/CT in its current form will be considered as an add-on for imaging for 2 years). This time frame was determined in part by the continual development and upgrading of FDG PET/CT, such that the estimates for diagnostic test accuracy are likely to change outside this time frame. Figure 9 details the results from the EVPI analysis at a population level. At a willingness-to-pay threshold of £30,000 per QALY, the EVPI for the population is approximately £34,000; thus it would not be worthwhile seeking additional information for FDG PET/CT for primary rectal cancer.

FIGURE 9. The EVPI for primary rectal cancer – population level.

FIGURE 9

The EVPI for primary rectal cancer – population level.

Primary colon cancer model

Table 23 details the expected costs of the imaging involved in the conventional and the intervention test strategies, the expected probability of a correct diagnosis under each strategy and the cost-effectiveness in terms of cost per correct diagnosis for primary colon cancer. On this basis, the addition of FDG PET/CT was dominated by the conventional strategy, i.e. FDG PET/CT was both more expensive and less effective.

TABLE 23. Primary colon cancer – cost per correct diagnosis.

TABLE 23

Primary colon cancer – cost per correct diagnosis.

Table 24 details the expected costs of the imaging and treatment associated with the conventional and the intervention test strategies, the expected outcomes in terms of QALYs under each strategy and the cost-effectiveness in terms of cost per QALY gain for primary colon cancer. On this basis, the addition of FDG PET/CT to the conventional strategy involved an additional cost of approximately £171,000 per QALY gained and would not be considered cost-effective under the usual definition (£20,000 per QALY < ICER < £30,000 per QALY).150

TABLE 24. Primary colon cancer – cost per QALY gain.

TABLE 24

Primary colon cancer – cost per QALY gain.

Figure 10 illustrates the uncertainty surrounding the expected incremental cost and incremental QALY results for primary colon cancer. The figure shows that there was considerable uncertainty about the extent, but not the existence, of the additional expected costs (shown in the vertical plane) and the existence and extent of the additional expected QALYs (shown in the horizontal plane).

FIGURE 10. The cost-effectiveness plane for FDG PET/CT in primary colon cancer.

FIGURE 10

The cost-effectiveness plane for FDG PET/CT in primary colon cancer.

The CEAC (Figure 11) illustrates the uncertainty in the cost-effectiveness estimate for primary colon cancer. The figure shows that, at a monetary threshold of £100,000 per QALY, the probability that the addition of FDG PET/CT will be cost-effective is approximately 30%. At a threshold of £30,000 per QALY, the CEAC illustrates that the probability that FDG PET/CT will be cost-effective is approximately 1%. At this threshold the probability that the conventional CT strategy will be cost-effective is approximately 99%.

FIGURE 11. The CEAC for primary colon cancer.

FIGURE 11

The CEAC for primary colon cancer.

The EVPI results show that, at a willingness-to-pay threshold of £30,000 per QALY, the EVPI per decision is < £2. To determine the overall population value of the EVPI we assumed an annual incidence of 21,574156 cases and a time frame of 2 years (i.e. FDG PET/CT in its current form will be considered for imaging for 2 years). As noted above, this time frame was determined in part by the continual development and upgrading of FDG PET/CT, such that the estimates for diagnostic test accuracy are likely to change outside of this time frame. Figure 12 details the results from the EVPI analysis at a population level. At a willingness-to-pay threshold of £30,000 per QALY, the EVPI for the population is approximately £70,000; thus it would not be worthwhile seeking additional information for FDG PET/CT for primary colon cancer.

FIGURE 12. The EVPI for primary colon cancer.

FIGURE 12

The EVPI for primary colon cancer.

Scenario analysis: contrast-enhanced FDG PET/CT as a lone technology

Primary rectal cancer scenario

The results from the primary rectal cancer scenario, which replaced contrast-enhanced CT with contrast-enhanced FDG PET/CT in addition to an MRI scan, are detailed in Table 25. The results show an improvement in cost-effectiveness compared with the baseline add-on FDG PET/CT results (detailed in Table 21); however, with an ICER of £107,652 this potential future strategy of contrast-enhanced FDG PET/CT as a replacement for contrast-enhanced CT in primary rectal cancer would not be considered to be cost-effective under the usual definition of willingness to pay (£20,000 per QALY < ICER < £30,000 per QALY).150

TABLE 25. Primary rectal cancer scenario – cost per QALY gain.

TABLE 25

Primary rectal cancer scenario – cost per QALY gain.

Figure 13 illustrates the uncertainty surrounding the expected incremental cost and incremental QALY results for the primary rectal cancer scenario. The figure shows that there is considerable uncertainty about the extent and existence of the additional expected costs (shown in the vertical plane) and the existence and extent of the additional expected QALYs (shown in the horizontal plane).

FIGURE 13. The cost-effectiveness plane for contrast-enhanced FDG PET/CT in the primary rectal cancer scenario.

FIGURE 13

The cost-effectiveness plane for contrast-enhanced FDG PET/CT in the primary rectal cancer scenario.

The CEAC (Figure 14) illustrates the uncertainty in the cost-effectiveness estimate for the primary rectal cancer scenario. The figure shows that, at a monetary threshold of £30,000 per QALY, the probability of contrast-enhanced FDG PET/CT and MRI being cost-effective in comparison with contrast-enhanced CT and MRI was < 20%.

FIGURE 14. The CEAC for the primary rectal cancer scenario.

FIGURE 14

The CEAC for the primary rectal cancer scenario.

Figure 15 details the results from the EVPI analysis at a population level. The results indicate an EVPI per decision of £68, which translated to a population EVPI of £1.7M. Therefore, we concluded that it was potentially worthwhile to undertake further research to explore whether or not contrast-enhanced FDG PET/CT can be used as a replacement for contrast-enhanced CT in primary rectal cancer.

FIGURE 15. The EVPI for the primary rectal cancer scenario.

FIGURE 15

The EVPI for the primary rectal cancer scenario.

Primary colon cancer scenario

With regards to the primary colon cancer scenario, which compared conventional contrast-enhanced CT with contrast-enhanced FDG PET/CT alone, the results indicated that there was potential for this to be highly cost-effective. Table 26 shows that the ICER is £12,832, which is considerably below the usual definition of willingness to pay (£20,000 per QALY < ICER < £30,000 per QALY).150

TABLE 26. Primary colon cancer scenario – cost per QALY gain.

TABLE 26

Primary colon cancer scenario – cost per QALY gain.

Figure 16 illustrates the uncertainty surrounding the expected incremental cost and incremental QALY results for the primary colon cancer scenario. The figure shows that there was considerable uncertainty about the extent and existence of the additional expected costs (shown in the vertical plane) and the existence and extent of the additional expected QALYs (shown in the horizontal plane).

FIGURE 16. The cost-effectiveness plane for contrast-enhanced FDG PET/CT in the primary colon cancer scenario.

FIGURE 16

The cost-effectiveness plane for contrast-enhanced FDG PET/CT in the primary colon cancer scenario.

The CEAC (Figure 17) illustrates the uncertainty in the cost-effectiveness estimate for the primary rectal cancer scenario. The figure shows that, at a monetary threshold of £30,000 per QALY, there was a 60% probability of contrast-enhanced FDG PET/CT being cost-effective in comparison with contrast-enhanced CT.

FIGURE 17. The CEAC for the primary colon cancer scenario.

FIGURE 17

The CEAC for the primary colon cancer scenario.

Figure 18 details the results from the EVPI analysis at a population level. The results indicated an EVPI per decision of £290, which translated to a population EVPI of £12.3M. Therefore, we concluded that it is potentially worthwhile undertaking further research to explore whether or not contrast-enhanced FDG PET/CT can be used as a replacement for contrast-enhanced CT in primary colon cancer.

FIGURE 18. The EVPI for the primary colon cancer scenario.

FIGURE 18

The EVPI for the primary colon cancer scenario.

Recurrent rectal cancer model

Table 27 details the expected costs of the imaging involved in the conventional and the intervention test strategies, the expected probability of a correct diagnosis under each strategy and the cost-effectiveness in terms of cost per correct diagnosis for recurrent rectal cancer. On this basis, the addition of FDG PET/CT involved an additional cost of approximately £12,000 per correct diagnosis and would be considered cost-effective compared with the conventional strategy under the usual definition (£20,000 per QALY < ICER < £30,000 per QALY).150

TABLE 27. Recurrent rectal cancer – cost per correct diagnosis.

TABLE 27

Recurrent rectal cancer – cost per correct diagnosis.

Table 28 details the expected costs of the imaging and treatment associated with the conventional and the intervention test strategies, the expected outcomes in terms of QALYs under each strategy and the cost-effectiveness in terms of cost per QALY gain for recurrent rectal cancer. On this basis, the addition of FDG PET/CT to the conventional strategy involved an additional cost of £21,409 per QALY gained and was likely to be considered cost-effective under the usual definition (£20,000 per QALY < ICER < £30,000 per QALY).150

TABLE 28. Recurrent rectal cancer – cost per QALY gain.

TABLE 28

Recurrent rectal cancer – cost per QALY gain.

Figure 19 illustrates the uncertainty surrounding the expected incremental cost and incremental QALY result for recurrent rectal cancer. The figure shows that there was considerable uncertainty about the existence and extent of the additional expected costs (shown in the vertical plane) and the existence and extent of the additional expected QALYs (shown in the horizontal plane).

FIGURE 19. The cost-effectiveness plane for FDG PET/CT in recurrent rectal cancer.

FIGURE 19

The cost-effectiveness plane for FDG PET/CT in recurrent rectal cancer.

The CEAC (Figure 20) illustrates the uncertainty in the cost-effectiveness estimate for recurrent rectal cancer. The figure shows that, at a monetary threshold of < £20,000 per QALY, there was a greater probability that the conventional CT and MRI strategy was the most cost-effective, but at a monetary threshold of > £20,000 per QALY the add-on FDG PET/CT strategy had a greater probability of being the most cost-effective. At the £30,000 per QALY threshold recommended by the National Institute for Health and Clinical Excellence (NICE) the CEAC indicated an approximately 70% probability that FDG PET/CT would have been cost-effective in comparison with the conventional strategies.

FIGURE 20. The CEAC for recurrent rectal cancer.

FIGURE 20

The CEAC for recurrent rectal cancer.

The EVPI results show that it is potentially worthwhile to collect more information about the use of FDG PET/CT for recurrent rectal cancer; at a willingness-to-pay threshold of £30,000 per QALY, the EVPI per decision is £316. To determine the overall population value of EVPI we assumed an annual incidence of 9054 cases (derived from the annual incidence of rectal cancer,156 assuming 70% recurrence123 and a death rate of 2.8% prior to recurrence diagnosis) and a time frame of 2 years (i.e. FDG PET/CT in its current form would be considered for imaging for 2 years). As noted above, this time frame was determined in part by the continual development and upgrading of FDG PET/CT, such that the estimates for diagnostic test accuracy would be likely to change outside this time frame. Figure 21 details the results from the EVPI analysis at a population level. At a willingness-to-pay threshold of £30,000 per QALY, the EVPI for the population was approximately £5.6M; thus it is potentially worthwhile seeking additional information for FDG PET/CT for recurrent rectal cancer.

FIGURE 21. The EVPI for recurrent rectal cancer.

FIGURE 21

The EVPI for recurrent rectal cancer.

Recurrent colon cancer model

Table 29 details the expected costs of the imaging involved in the conventional and the intervention test strategies, the expected probability of a correct diagnosis under each strategy and the cost-effectiveness in terms of cost per correct diagnosis for recurrent colon cancer. On this basis, the addition of FDG PET/CT involved an additional cost of approximately £3000 per correct diagnosis and would be considered cost-effective compared with the conventional strategy under the usual definition (£20,000 per QALY < ICER < £30,000 per QALY).

TABLE 29. Recurrent colon cancer – cost per correct diagnosis.

TABLE 29

Recurrent colon cancer – cost per correct diagnosis.

Table 30 details the expected costs of the imaging and treatment associated with the conventional and the intervention test strategies, the expected outcomes in terms of QALYs under each strategy and the cost-effectiveness in terms of cost per QALY gain for recurrent colon cancer. On this basis, the addition of FDG PET/CT to the conventional strategy involved an additional cost of approximately £6000 per QALY gained and would be considered cost-effective under the usual definition (£20,000 per QALY < ICER < £30,000 per QALY).150

TABLE 30. Recurrent colon cancer – cost per QALY gain.

TABLE 30

Recurrent colon cancer – cost per QALY gain.

Figure 22 illustrates the uncertainty surrounding the expected incremental cost and incremental QALY results for recurrent colon cancer. The figure shows that there was considerable uncertainty about the extent, but not the existence, of the additional expected costs (shown in the vertical plane) and the existence and extent of the additional expected QALYs (shown in the horizontal plane).

FIGURE 22. The cost-effectiveness plane for FDG PET/CT in recurrent colon cancer.

FIGURE 22

The cost-effectiveness plane for FDG PET/CT in recurrent colon cancer.

The CEAC (Figure 23) illustrates the uncertainty in the cost-effectiveness estimate for recurrent colon cancer. The figure shows that, at a monetary threshold > £6000 per QALY, the FDG PET/CT strategy had the greatest probability of being cost-effective. At the £30,000 per QALY threshold recommended by NICE, the CEAC indicated an approximately 85% probability that FDG PET/CT would be cost-effective in comparison with the conventional strategies.

FIGURE 23. The CEAC for recurrent colon cancer.

FIGURE 23

The CEAC for recurrent colon cancer.

The EVPI results show that it would be potentially worthwhile collecting more information about the use of FDG PET/CT for recurrent rectal cancer; at a willingness-to-pay threshold of £30,000 per QALY, the EVPI per decision is £178. To determine the overall population value of the EVPI we assumed an annual incidence of 14,670 cases (derived from the annual incidence of colon cancer,156 assuming 70% recurrence123 and a death rate of 2.8% prior to recurrence diagnosis) and a time frame of 2 years (i.e. FDG PET/CT in its current form would be considered for imaging for 2 years). As noted above, this time frame was determined in part by the continual development and upgrading of FDG PET/CT, such that the estimates for diagnostic test accuracy are likely to change outside this time frame. Figure 24 details the results from the EVPI analysis at a population level. At a willingness-to-pay threshold of £30,000 per QALY, the EVPI for the population was approximately £5.1M; thus, it would be potentially worthwhile seeking additional information for FDG PET/CT for recurrent colon cancer.

FIGURE 24. The EVPI for recurrent colon cancer.

FIGURE 24

The EVPI for recurrent colon cancer.

Metastatic cancer model

Table 31 details the expected costs of the imaging involved in the conventional and the intervention test strategies, the expected probability of a correct diagnosis under each strategy and the cost-effectiveness in terms of cost per correct diagnosis for metastatic CRC. On this basis, the addition of FDG PET/CT involved an additional cost of approximately £19,000 per correct diagnosis and would be considered cost-effective compared with the conventional strategy under the usual definition (£20,000 per QALY < ICER < £30,000 per QALY).

TABLE 31. Metastatic cancer – cost per correct diagnosis.

TABLE 31

Metastatic cancer – cost per correct diagnosis.

Table 32 details the expected costs of the imaging and treatment associated with the conventional and the intervention test strategies, the expected outcomes in terms of QALYs under each strategy and the cost-effectiveness in terms of cost per QALY gain for metastatic CRC. On this basis, the addition of FDG PET/CT to the conventional strategy involved an additional cost of approximately £21,000 per QALY gained and would be considered cost-effective under the usual definition (£20,000 per QALY < ICER < £30,000 per QALY).150

TABLE 32. Metastatic cancer – cost per QALY gain.

TABLE 32

Metastatic cancer – cost per QALY gain.

Figure 25 illustrates the uncertainty surrounding the expected incremental cost and incremental QALY results for metastatic CRC. The figure shows that there was considerable uncertainty about the existence and extent of the additional expected costs (shown in the vertical plane) and the existence and extent of the additional expected QALYs (shown in the horizontal plane).

FIGURE 25. The cost-effectiveness plane for FDG PET/CT in metastatic CRC.

FIGURE 25

The cost-effectiveness plane for FDG PET/CT in metastatic CRC.

The CEAC (Figure 26) illustrates the uncertainty in the cost-effectiveness estimate for metastatic CRC. The figure shows that there was considerable uncertainty surrounding the cost-effectiveness of the FDG PET/CT strategy. At a monetary threshold of £21,000 per QALY, the probability that the FDG PET/CT intervention would be cost-effective was approximately 50%, as was the probability that CT would be cost-effective. At the £30,000 per QALY threshold recommended by NICE, the CEAC indicated that the FDG PET/CT intervention had a slightly greater probability of being cost-effective (52%).

FIGURE 26. The CEAC for metastatic CRC.

FIGURE 26

The CEAC for metastatic CRC.

The EVPI results show that it is potentially worthwhile collecting more information about the use of FDG PET/CT for metastatic CRC. At a willingness-to-pay threshold of £30,000 per QALY, the EVPI per decision is £1328. To determine the overall population value of EVPI we assumed an annual incidence of 4000 cases (derived from the annual incidence of colon and rectal cancers,156 assuming a 70% likelihood of recurrence and a further 40% likelihood that the recurrence would be metastatic,123 and a death rate of 59% prior to metastatic diagnosis) and a time frame of 2 years (i.e. FDG PET/CT in its current form will be considered for imaging for 2 years). As noted above, this time frame was determined in part by the continual development and upgrading of FDG PET/CT, such that the estimates for diagnostic test accuracy are likely to change outside this time frame. Figure 27 details the results from the EVPI analysis for metastatic CRC at a population level. At a willingness-to-pay threshold of £30,000 per QALY, the EVPI for the population was approximately £10.5M; thus it is potentially worthwhile seeking additional information for FDG PET/CT for metastatic CRC.

FIGURE 27. The EVPI for metastatic CRC.

FIGURE 27

The EVPI for metastatic CRC.

Discussion of the economic modelling

Primary colorectal cancer

To our knowledge, this is the first published evidence assessing the cost-effectiveness of FDG PET/CT as an add-on device for staging primary rectal and colon cancer.

Considering the cost per correct diagnosis outcomes, it is apparent that the diagnostic test accuracy estimates used in the models favour the conventional imaging modalities. This is a result of the use of FDG PET/CT as an add-on imaging device. In the primary model, FDG PET/CT is influential as an add-on after the conventional test; therefore, the model assumes any outcome with a positive test will be treated as such. Negative results from the conventional imaging tests that are refuted by the FDG PET/CT test are treated as positive. Results are treated as negative only when both the conventional and the FDG PET/CT test outcomes are negative. This results in an overall larger number of positive outcomes (both TPs and FPs) and a reduction in negative outcomes (fewer TNs identified) in the intervention strategy. Therefore, in both the rectal and colon primary models, the total proportion of correct diagnoses is greater using the conventional strategies, with no addition of FDG PET/CT. Adopting an alternative strategy in which an FDG PET/CT scan is undertaken only if the conventional tests identify negative outcomes may be a more cost-effective strategy. The model outcomes would be the same as in the strategy adopted in our analysis; however, the cost of an FDG PET/CT scan (which is approximately four times that of the conventional scans) would be incurred only when conventional imaging results are negative, reducing the overall cost of the strategy.

The longer-term analyses for the primary models showed that the low specificity of FDG PET/CT scans results in a greater number of FP outcomes, in which patients are overstaged and incur additional costs and suffer quality of life impacts for unnecessary treatments. This is reflected in the extremely high ICER outcomes in the primary models. Further, the therapeutic impact literature discussed earlier in this report27,49,8993,102,109 found that, although FDG PET/CT may potentially affect accurate staging of primary CRC, it had only a minor impact in changing patient management. Both the rectal and colon primary models identified an incremental QALY gain of only 0.01, indicating that FDG PET/CT as an add-on imaging device in primary CRC does not have any overall impact on patient outcomes.

The cost per QALY results were extremely high for both the primary rectal and colon evaluations: > £400,000 and > £170,000, respectively, per QALY. As such, FDG PET/CT is not cost-effective in either primary rectal or primary colon cancer given the NICE recommended QALY threshold of £20,000–30,000 per QALY.

The cost-effectiveness plane shows that there is uncertainty surrounding both the incremental costs and incremental effects associated with FDG PET/CT; however, the CEACs show that this uncertainty does not translate into uncertainty about the cost-effectiveness of FDG PET/CT. For both primary rectal and colon cancer, the probability that FDG PET/CT as an add-on imaging device for staging is cost-effective is zero given the NICE recommended QALY threshold range of £20,000–30,000 per QALY. This translates into very small values for the EVPI. The results therefore show that the use of FDG PET/CT as an add-on imaging device for staging primary CRCs is not cost-effective and that there is no value associated with the collection of further information.

Earlier findings from the systematic review indicated that, as FDG PET/CT technology develops, there will be an increased potential in the future for this improved technology to be used as a lone device, replacing contrast-enhanced CT, as opposed to being utilised as an add-on imaging device. In primary rectal cancer, contrast-enhanced FDG PET/CT could potentially replace contrast-enhanced CT with the addition of an MRI scan, and in primary colon cancer, contrast-enhanced FDG PET/CT could be used alone as a replacement for contrast-enhanced CT. The two scenario analyses undertaken to explore this in primary CRC indicated that such an improved contrast-enhanced FDG PET/CT device is unlikely to be cost-effective for use in primary rectal cancer, but is likely to be very cost-effective for use in colon cancer. In primary rectal cancer, improved FDG PET/CT technology will not negate the necessity for an MRI scan, and therefore the potential incremental value of improved contrast-enhanced FDG PET/CT is limited by the strong diagnostic test accuracy achievable with MRI scanning. The colon cancer scenario analysis indicates substantial improvement in diagnostic test accuracy from contrast-enhanced FDG PET/CT compared with contrast-enhanced CT and improved efficiency through eliminating the need for an add-on test, thereby giving a highly cost-effective outcome. There remains considerable uncertainty in both these outcomes, which is highlighted in the value of information analyses, indicating potential value in further research with a population EVPI of £1.7M for the primary rectal population, and a value of £12.3M for the primary colon population. It must be noted that EVPI analysis can provide only an indication of potential worth for further information as any research undertaken will only reduce, rather than eliminate, uncertainty.

Recurrent colorectal cancer

The recurrent cancer models found FDG PET/CT as an add-on imaging device to have an ICER of £21,409 for rectal cancer and £6189 for colon cancer. Considering the NICE monetary threshold of £20,000–30,000 per QALY, these can be considered to be cost-effective.

The ICER for the recurrent colon cancer model is considerably lower than that for the recurrent rectal cancer model, indicating that FDG PET/CT is more cost-effective in the assessment of colon recurrence than in rectal recurrence. This difference is likely to be due to the sensitivity estimate for the CT diagnostic test parameter, which is considerably lower than the FDG PET/CT sensitivity estimate. The wide difference favours the accuracy of FDG PET/CT, and even though uncertainty around both these estimates was incorporated into the model, the strong influence of the choice of diagnostic test accuracy parameters on model outcomes is evident. The FDG PET/CT intervention does not have the same diagnostic test accuracy sensitivity advantage in the recurrent rectal model, as the MRI scan diagnostic test accuracy estimates are also incorporated. The MRI diagnostic test accuracy was superior to that of CT; therefore, in the recurrent rectal model, the conventional imaging diagnostic test accuracy estimates are closer to those of FDG PET/CT, limiting the incremental value of FDG PET/CT.

Meta-analyses were undertaken using relevant papers identified from the systematic review to elicit pooled diagnostic test accuracy estimates of FDG PET/CT for recurrent CRC. Because of inadequacies and reporting bias in the identified papers, the pooled estimates for FDG PET/CT were considered to be an inaccurate reflection of the diagnostic test accuracy of FDG PET/CT, and the CIs were tight around the pooled means, which was considered to be restrictive in terms of capturing a wide range of uncertainty. Therefore, expert judgement was used to determine point estimates and wide uncertainty intervals from the literature.

Most previous economic evaluations undertaken for recurrent CRC have been specifically interested in hepatic metastases. Two papers were identified that were interested in assessing recurrence. Sloka et al.110 undertook a cost-effectiveness analysis of FDG PET/CT in comparison with CT for diagnosing colorectal recurrence, and the Medical Services Advisory Committee58 undertook a cost–consequence analysis of PET versus no PET for diagnosing local recurrence. Our models add to this literature, providing an assessment of the cost-effectiveness of FDG PET/CT as an add-on imaging device for diagnosing both recurrent rectal and recurrent colon cancer.

Sloka et al.110 report cost savings with the FDG PET/CT approach through avoidance of unnecessary surgeries. The paper does not report the number of unnecessary surgeries avoided in each strategy, only the cost savings. After considering the parameter estimates used in their model, it can be seen that the diagnostic test accuracy estimates assigned to FDG PET/CT are superior to those in the CT comparator arm by a wide margin, so it is no surprise that the FDG PET/CT intervention was found to dominate CT. Our recurrent colon model assigned the same specificity values to contrast-enhanced CT and FDG PET/CT, and therefore there was no difference in unnecessary surgeries in our outcomes; however, our recurrent rectal model did indicate reductions in unnecessary surgeries with the FDG PET/CT intervention. The Medical Services Advisory Committee publication58 also reports cost savings through the use of PET in comparison with a no PET strategy; however, few details are provided as to what the no PET strategy entails.

In comparison with other economic evaluations undertaken in this disease area, our models appear to have adopted a more conservative approach in assigning diagnostic test accuracy estimates and through incorporating quality of life impacts and overall survival impacts in a cost per QALY outcome. This conservative approach attempted to minimise bias in the models to avoid unfairly favouring the intervention arm (add-on FDG PET/CT).

At a cost per QALY threshold of £30,000, the probability that the FDG PET/CT intervention will be cost-effective for recurrent colon cancer is 85%; this is lower (70%) for recurrent rectal cancer. This greater level of uncertainty in the recurrent models leads to a non-zero value for the EVPI (at a population level the EVPI is £5.6M for recurrent rectal cancer and £5.1M for recurrent colon cancer). At these levels there is potential worth in collecting further information to inform the decision regarding the use of FDG PET/CT in the future. It must be noted that EVPI analysis can provide only an indication of potential worth for further information, as any research undertaken will reduce rather than eliminate uncertainty.

Metastatic colorectal cancer

The metastatic model found FDG PET/CT as an add-on device to have an ICER of £21,434 per QALY gained. This ICER value is within the NICE monetary threshold range of £20,000–30,000 per QALY for determining cost-effectiveness.

Most of the existing publications that have undertaken economic evaluations of PET for CRC have been specifically interested in hepatic metastases. Park et al.109 developed a decision model to determine the cost-effectiveness of PET and CT imaging in comparison with CT alone. They evaluated outcomes in terms of life-year gains and report an incremental cost per life-year gained of US$16,437. This paper is the most similar to our model, but does not incorporate quality of life impacts.

Other economic evaluations in metastatic CRC have been undertaken. Lejeune et al.51 report cost savings of €2671 (US$3213) with no change in life expectancy when FDG PET/CT was compared with CT in staging metastatic CRC. Zubeldia et al.111 assessed the cost-effectiveness of FDG PET/CT in comparison with CT for identifying the presence of extrahepatic metastases. They report a cost saving of US$5269 as a result of unnecessary surgeries avoided; however, they provide few details of how their model was constructed. Details were not provided of the diagnostic test accuracy estimates used in the model or how the impact on patient management was incorporated. None of these metastatic models used probabilistic analysis to incorporate uncertainty for each of the model parameters.

The CEAC for the metastatic model reflects uncertainty towards the cost-effectiveness of FDG PET/CT as an add-on strategy. The CEAC illustrates that beyond a threshold of £21,000 per QALY FDG PET/CT has a greater probability of being cost-effective than CT, although there is considerable uncertainty, with the probability that FDG PET/CT is cost-effective ranging between 40% and 60%. This level of uncertainty leads to an EVPI of £10.5M for the population. Thus, it is potentially worthwhile collecting further information to inform the decision regarding FDG PET/CT in the future. It must be noted that EVPI analysis can provide only an indication of potential worth for further information, as any research undertaken will reduce rather than eliminate uncertainty.

Future research

There is potential value in undertaking further research into the use of FDG PET/CT for:

  • staging recurrent colon cancer.
  • staging recurrent rectal cancer.
  • staging metastatic CRC.

There is the potential for contrast-enhanced PET/CT technology to be used as a replacement for contrast-enhanced CT in primary CRC, if and when this technology becomes available. Further research in this area is likely to be worthwhile, particularly for use in investigating primary colon cancer.

© 2011, Crown Copyright.

Included under terms of UK Non-commercial Government License.

Bookshelf ID: NBK99948

Views

  • PubReader
  • Print View
  • Cite this Page
  • PDF version of this title (1.7M)
  • Disable Glossary Links

Other titles in this collection

Recent Activity

Your browsing activity is empty.

Activity recording is turned off.

Turn recording back on

See more...