Positron emission tomography/computerised tomography imaging in detecting and managing recurrent cervical cancer: systematic review of evidence, elicitation of subjective probabilities and economic modelling

Aetiology

Human papillomavirus (HPV) infection of the cervix is a sexually transmitted infection that is necessary for the development of cervical cancer. 2 However, only a relatively small proportion of women who encounter persistent infection from high-risk genotypes (HPV 16 and 18, and some other strains) go on to develop cervical cancer. 3 When HPV is detected, around 17% of women go on to develop cervical intraepithelial neoplasia grade II+ within 3 years. 2 HPV infection is very common; it is estimated that 20% of sexually active girls will contract the virus by the age of 18 years. 4 The risk of infection increases with the age at first sexual intercourse. 5

There are a number of factors that can increase or decrease the risk of developing cervical cancer:

• Age. Cervical cancer is rare before the age of 20 years but the incidence increases rapidly with age, giving a peak incidence of around 17 per 100,000 between the ages of 30 and 39 years. 6 Cervical cancer mortality rates generally increase with age, so that only about 7% of cervical cancer deaths occur in women under 35 years, with the highest rates in women over 70 years. 7 Squamous cell tumours are more common, but the rates of both squamous cell tumours and adenocarcinomas rise sharply from age 20–40 years, after which they plateau until age 80 years. 8

• Sexual behaviour. There is an increased risk of invasive cervical cancer with early age at first sexual intercourse,5,9 early pregnancy5 and current use of hormonal contraceptives. 10

• Smoking. Current smoking intensity is an independent risk factor for high-grade cervical intraepithelial neoplasia in young women, after controlling statistically for cervical HPV infection,11 and may be a risk factor for developing cervical cancer. 12

• HIV infection. HIV infection leads to an increased risk of advanced and early cervical pathology. 13

• Socioeconomic status. Women living in the most deprived areas in the UK have cervical cancer rates that are more than three times as high as those women in the least deprived areas. Data from a longitudinal study, representing 1% of the population from England and Wales, showed that cervical cancer incidence is considerably higher among women of working age in manual occupations than among women in non-manual occupations. 14

Epidemiology

Cervical cancer is a common gynaecological malignancy, with an estimated 31,400 new cases diagnosed each year in the European Union. 15 In the UK, approximately 2800 patients are diagnosed with cervical cancer per year, accounting for around 2% of all female cancer cases. 6 In England, carcinoma of the cervix is rare in women < 20 years of age. 6 Cancer of the cervix is a leading cause of cancer death in women. In 2008, there were 1110 deaths from cervical cancer in the UK, giving a European age-standardised death rate of 2.7 per 100,000 person-years. 15 In the UK population, the 5-year disease-free survival rate for treated stage IA disease is almost 100%, whereas it is 50–70% for stage IB2 and IIB, 30–50% for stage III and 5–15% for stage IV disease. 16 It is estimated that the median survival for stage IVB disease is around 9–10 months, with 30% of patients surviving 1 year and 2–5% surviving 2 years. 17

Initial treatment of cervical cancer

When patients are initially diagnosed with cervical cancer they can be treated with surgery, a combination of chemotherapy and radiotherapy (chemoradiotherapy) or with palliative care. The treatment chosen is based on stage of tumour, fitness of the woman and tumour characteristics, for example greater than one-third stromal invasion, capillary lymphatic space involvement and large tumour diameter.18 Surgery is usually radical hysterectomy but can also be trachelectomy (if the tumour is small), which is the removal of the cervix only rather than the whole uterus and can be performed in younger women with early cervical cancer who wish to retain their fertility.3 Approximately 20–30% of women undergoing surgery also receive adjuvant postoperative chemoradiotherapy for positive tumour margins or positive lymph nodes or because of the tumour size, volume, lymphovascular space invasion or stromal invasion. 19

Recurrent or persistent cervical cancer

Patients can be cured by initial treatment and approximately 70–80% of initially treated cases are cured with surgery. If surgery is not appropriate because of tumour characteristics or lack of fitness in the patient, chemoradiotherapy can be given. However, the initial treatment may not affect a cure and in approximately 15% of patients disease is detected 3 months after treatment, which is called persistent cervical cancer (rather than recurrent). Recurrence is more common within the first 24 months after the initial diagnosis, but can happen up to 15 years after initial treatment. 20

The Scottish Intercollegiate Guidelines Network (SIGN) guideline3 found the rates of recurrence from the three studies reviewed in the guideline to be 13%,21 18.2%22 and 29%. 23 In another study, the proportion with recurrence in early-stage cervical cancer was 6%;24 a further study of locally advanced cervical cancer reported 30% recurrence. 25 Recurrences are more common within the first 24 months after the initial diagnosis – the median disease-free interval was 17 months for symptomatic patients and 16 months for asymptomatic patients in one cohort21 and the median time from surgery to recurrence in another cohort was 17.6 months. 22 The percentage recurrence was higher after radiotherapy (17%) than after surgery (13%),21 but none of the studies compared recurrence after chemoradiotherapy with recurrence after surgery. The proportions of asymptomatic to symptomatic recurrences were 19 : 11421 and 2 : 5. 22

Patients with pelvic recurrence usually present with one or more of vaginal bleeding, discharge, pelvic pain and sciatic pain. Patients with disseminated recurrence eventually develop systemic symptoms associated with cachexia.

Risk factors for recurrence include disease stage, number of positive lymph nodes, parametrial involvement and depth of invasion of the tumour. 24 The squamous cell carcinoma antigen is elevated in 28–88% of patients with cervical cancer and can precede clinical diagnosis of relapse in 46–92% of cases. 26

Patients with recurrence or persistence are described according to the stage they were when they were diagnosed originally, along with some further information on whether or not and how much the cancer has progressed since the original diagnosis. For example, a woman who presented with a stage IIA cancer who now has distant metastases does not become a stage IVB cancer, but is described as a stage IIA cancer with metastases. Occasionally, a new stage can be assigned in addition if the cancer has recurred, particularly in trials, in which case it will be described with a lower case r in front of the new staging, for example stage rIVB. 27

Prognosis

Survival with recurrent or persistent disease is poor – from 6 months to 2 years. 3 Also, patients frequently experience substantial morbidity from local recurrence and distant spread. 3 It is unclear whether or not earlier detection of recurrence (from clinical follow-up or scanning) leads to increased survival rates, but this is a reasonable assumption to make. Worse survival is associated with shorter disease-free interval, being symptomatic and poorer prognostic factors. 28

Computerised tomography and magnetic resonance imaging scanning

Computerised tomography scanning was introduced in the 1970s and is now widely used in the NHS. A CT scan is a series of tomographic radiographic images used to visualise two-dimensional ‘slices’ through the body. Because the X-ray beam emission and the receiving film-intensifying screen are both revolving around a focal point in the body, this focal point can be visualised much more clearly than in a standard radiography film. A very large number of focal points are visualised consecutively and then a computer is used to mathematically reconstruct a two-dimensional matrix to give a digital image of the part of the body being scanned. CT scanning is painless and takes 15–30 minutes. It is non-invasive unless contrast medium is being used. For most whole-body CT scans, intravenous iodinated contrast is now used and there is the risk of allergic reactions. The main disadvantage, however, is the dose of radiation that is absorbed during the scanning. It has been estimated that 40% of all diagnostic radiation exposure in patients comes from CT scanning. 29 CT scanning can also produce artefacts that impede interpretation of the images. These artefacts can come from motion (e.g. patients have to hold their breath when the chest is being scanned) and from high-density objects such as tooth fillings and orthopaedic hardware.

Magnetic resonance imaging scanning was introduced in the 1980s and is now also widely used in major centres in the NHS. It is also a tomographic imaging technique but uses the ability of hydrogen atoms to absorb and emit radio waves (at a similar frequency to FM radio) when placed in a strong magnetic field. Visualisation of tissues can occur because of the different concentrations of hydrogen atoms in different tissues and the characteristics of the atoms in different complex biochemical environments. MRI uses characteristics such as the density of hydrogen atoms, the speed at which they become magnetised and lose their magnetisation and the presence of flow or motion in a tissue. MRI does not use ionising radiation, which is an advantage compared with CT. However, patients are placed in a magnetic field and so metal objects inside and outside the body will be affected. Patients with pacemakers, cochlear implants, shotgun fragments, etc. should not have a MRI scan. The energy generated inside the body can cause hyperthermia, particularly in obese people. The size of the trolley and aperture (MRI machines are longer than CT machines and fit the whole body inside) mean that people who weigh > 20 stone (127 kg) are unlikely to fit inside the machine. The machine is also noisy and a small proportion of patients have anxiety-related reactions. MRI scans can give false-positive results from motion artefacts, interfaces between fat and water and distortions due to magnetic objects inside the body.

Computerised tomography and MRI are high-resolution anatomical imaging techniques that are commonly used in cancer to detect potential tumours. MRI and CT are currently considered first when recurrence is suspected. 17 Whole-body CT and MRI scanning are now rarely performed; imaging for cervical cancer is frequently limited to the pelvis only. CT and MRI have limitations in differentiating recurrent tumours from postradiotherapy or surgical fibrosis and also have limitations in accurately identifying the extent of recurrence as small volume nodal metastasis. If CT or MRI of the pelvic area only is carried out, distant recurrence may not be identified. They can also be unreliable in determining the presence or absence of recurrent disease in the pelvis after radiotherapy, as radiotherapy-induced fibrosis makes tissues indurated and thus potentially conceals recurrent disease.

Positron emission tomography/computerised tomography scanning

Positron emission tomography is an imaging method that can be used to establish the functional parameters of tissue, allowing detection of metabolically active areas in tissues such as tumours. 30 ¹⁸F-fluorodeoxyglucose (¹⁸F-FDG) is the most widely used radiotracer and is intravenously injected 1–2 hours before imaging. It is a glucose analogue and is taken up and actively trapped in the enhanced glycolytic pathway of hypermetabolic areas, demonstrated by high-energy photons emitted as a result of annihilation of positrons emitted by the radioisotope, with nearby negatively charged electrons. PET provides anatomical image resolution of the order of 4–6 mm, significantly better than conventional gamma cameras but inferior to the 1- to 2-mm resolution of CT or MRI. The size of lesion that can be detected by PET is limited by several factors, including the physics of positron emission, the spatial resolution of the scanner (typically 4.5–6.0 mm in the centre of the axial field) and the safe dosing limits of ¹⁸F-FDG. 30

Positron emission tomography/computerised tomography is a combination of PET scanning and CT scanning on the same machine. It precisely aligns and combines metabolic PET imagines with anatomical CT images obtained immediately and consecutively without patient movement, and is being increasingly preferred over PET scanning alone as it allows more precise localisation of active disease sites than either technology separately. The CT scan usually has a lower radiation dose than standard CT scans and contrast media are rarely used. PET-CT in suspected recurrent or persistent cervical cancer can detect metabolically active metastatic lesions in normal-sized nodes and in postsurgical or radiotherapy fibrosis. PET-CT in the follow-up of cervical cancer patients can be used to identify recurrent or persistent disease, assess local tumour extension, evaluate pelvic nodal involvement, detect distant metastases (e.g. lung, supraclavicular lymph nodes and para-aortic lymph nodes), plan radiotherapy and assess response to therapy. 31

There are several disadvantages to PET-CT scanning. First, the machine is very expensive (approximately £2M). Second, 18F-FDG has a short half-life of around 2 hours and therefore can cause throughput difficulties. False-positives are relatively common because the technique is looking for metabolically active regions and not all are cancerous, for example sepsis and inflammation following surgery and radiotherapy may mimic metastases. False-negatives can also occur soon after chemotherapy because the drugs may slow the metabolism of the metastases but not eliminate them altogether. Therefore, PET-CT to find secondary spread is not recommended within 3 months of surgery and radiotherapy and within 6 weeks of chemotherapy.

Survival data from positron emission tomography/computerised tomography studies in cervical cancer

There are two publications34,35 that contain useful information about survival in cervical carcinoma, using PET-CT to differentiate between different groups of patients, including those with persistent and recurrent cervical cancer. In Schwartz et al. ,35 92 women who had been treated with chemoradiotherapy for carcinoma of the cervix (FIGO stages IB1 to IVA) and who had whole-body PET-CT between 8 and 16 weeks after initial therapy were followed up clinically for at least 6 months (range 6–49 months). PET-CT was used to investigate prognosis, linking findings with progression-free survival and cause-specific survival. Among the 92 patients, PET-CT showed a complete response in 65 (71%) and persistent tumour in 15 (16%) and identified new abnormalities in 12 (13%). The survival rates are shown in Figure 1. The 3-year cause-specific rates were 96% for women with a complete response to treatment and 43% for patients with persistent disease, and the 2-year survival rate was 14% for patients with any new sites of disease. The 3-year progression-free survival rates were 78% for patients with a complete response after therapy, 33% for patients with persistent disease and 0% for those with new sites of tumour.

FIGURE 1.

Cause-specific survival rates (a) and progression-free survival rates (b) for patients categorised by PET-CT as having no tumour, persistent tumour or new site of cervical cancer.

Brooks et al. 34 investigated the usefulness of PET-CT imaging in 78 asymptomatic and 25 symptomatic patients following a complete response to initial chemoradiotherapy for cervical cancer. The post-therapy PET-CT was performed at 3 months after treatment completion and patients were followed up for a median of 13 months for asymptomatic patients and 8 months for symptomatic patients. Unfortunately, for the first 2 years only PET was used and for the remaining 4 years PET-CT was used. The number of women in each group is unclear. The survival curves are shown in Figure 2. The 3-year survival for patients with symptomatic recurrence was 19% compared with 59% for patients with asymptomatic recurrence (p = 0.09).

FIGURE 2.

Survival in asymptomatic and symptomatic patients undergoing surveillance PET and PET-CT following one scan at 3 months.

Treatment options for recurrent cervical cancer

Treatment of recurrent cervical cancer depends on the site (central, pelvic, distant), extent of recurrence, type of previous treatment received (surgery, chemoradiotherapy, radiotherapy), time elapsed since primary treatment and patient fitness. Treatment intention is usually curative or palliative. Palliative treatment is used when there are distant metastases or multiple site recurrences and is usually chemotherapy.

Potentially curative disease is defined as:

• confirmed recurrence of the disease confined to the pelvis, provided that the patient has not received previous primary or adjuvant pelvic radiotherapy

• disease confined to the central pelvis, without pelvic side wall or extrapelvic involvement, provided that radiotherapy has been administered before recurrence

• distant recurrences at a single site (such as para-aortic lymph node) that could be completely resected or encompassed by a curative radiotherapy procedure.

In women with recurrence who had surgery for their primary tumour, radiotherapy is the treatment of choice. This may also include chemotherapy, which is often single-agent cisplatin. 3 In women who had chemoradiotherapy or radiotherapy and who have persistent cervical cancer, salvage surgery is generally considered if the patient is sufficiently fit, if the disease is localised to the pelvis only and if surgery has a high chance of completely removing the disease with clear margins. 3 Surgery can be radical hysterectomy or pelvic exenteration. Surgery for relapsed disease after radiotherapy is often associated with high morbidity as radiation fibrosis makes surgery difficult and, to enhance cure rates, surgical excision of disease often involves removal of the bladder, uterus, cervix and various amounts of the vagina (anterior exenteration) or the uterus, vagina and portions of the rectosigmoid colon and anus (posterior exenteration) or a complete pelvic clearance (exenteration). In a small number of patients, radical hysterectomy will suffice if the disease is highly localised. As exenterations are morbid surgical procedures resulting in alteration of body image and loss of bladder and/or bowel control, patients require extensive preoperative psychosocial counselling.

Search strategy

A sensitive search was conducted to identify all relevant published and unpublished studies and studies in progress. All databases were searched from inception to May 2010. Search strategies were designed from a series of test searches and discussions of the results of those searches among the review team. Both medical subject heading (MeSH) terms and text words were used and included ‘cervical cancer’, ‘PET-CT’, ‘CT’ and ‘MRI’. The strategies from MEDLINE, EMBASE and The Cochrane Library can be found in Appendix 6. Literature was identified from several sources including:

• general health and biomedical databases: MEDLINE (Ovid), EMBASE (Ovid), Science Citation Index, The Cochrane Library [Cochrane Central Register of Controlled Trials (CENTRAL)], Medion

• checking of reference lists of systematic and narrative review articles

• searching a range of relevant databases including ClinicalTrials.gov and the UK Clinical Research Network Portfolio to identify information about studies in progress, unpublished research or research reported in grey literature

• specialist search gateways (OMNI and the National Cancer Institute), general search engine (Google) and meta-search engine (Copernic) from March to May 2010

• hand-searching of Gynecologic Oncology from 1980 to May 2010

• authors of included studies contacted for information on relevant published or unpublished studies.

Inclusion and exclusion criteria

Quality assessment

Test accuracy quality assessment followed the Quality Assessment of Diagnostic Accuracy Studies (QUADAS) guidelines40 and diagnostic and therapeutic impact quality assessment followed guidelines suggested by Meads and Davenport. 41 The items of methodological quality listed in the QUADAS guidelines40 are a representative spectrum, selection criteria clearly described, acceptable reference standard, acceptable delay between tests, partial verification avoided, differential verification avoided, reference standard independent of the index test, index test described in sufficient detail, reference standard described in sufficient detail, index test results blinded, reference standard results blinded, relevant clinical information available, uninterpretable results reported, and withdrawals explained.

These items were tailored to assess the included studies because different aspects of quality are applicable to different topic areas. The actual quality items used for this report are listed in Table 2. For acceptable delay between tests, this included delay between the index test and the comparator test (within 1 month) and between the index test and PET-CT (with 1 month). There will inevitably be a delay between the index test and clinical follow-up (as this had to be > 6 months). Differential verification was omitted because it was inevitable that the test positives would have a different reference standard (histology) to the test negatives (clinical follow-up).

Study quality was summarised in a table. Additional issues (e.g. study design characteristics, method of patient enrolment, technique of data collection) were also collected. Technical quality was assessed by a consultant radiologist with considerable experience in current cancer imaging techniques.

Rationale

Subjective probabilities were elicited from clinicians representing the disciplines of radiology, oncology and gynaecology. Eliciting subjective probabilities from clinicians had three roles in the planned investigation of the clinical effectiveness of PET-CT imaging in the detection and management of recurrent cervical cancer:

1. Providing data to populate the economic model in the absence of information found in the literature.

2. Supplementing information found in the literature. Literature may be sparse, of poor quality or not transferable to the UK setting. Information gained from clinicians in the form of subjective probabilities may be used to supplement information found in the literature and to enable sensitivity analyses to be performed as part of the economic model.

3. Planning the dissemination strategy for the results of the research. If there is wide variation in accuracy estimates elicited from clinicians, or if elicited estimates of accuracy are very discrepant with those found in the literature, this may impact on the successful dissemination of the research findings to clinicians.

Probabilities elicited

Informed by the preliminary results of the systematic reviews of test accuracy (and effectiveness), the research team decided on the data priorities for elicitation as follows:

1. To determine the prevalence of recurrence in women with an initial diagnosis of stage IB–IVA cervical cancer, who are assumed to be disease free for a minimum of 3 months post completion of primary treatment:

   1. presenting with symptoms suggestive of recurrence

   2. in the absence of symptoms

2. To determine the test accuracy of chest, abdominal and pelvic CT and/or MRI performed at the discretion of clinicians in women with an initial diagnosis of stage IB–IVA cervical cancer, who are assumed to be disease free for a minimum of 3 months post completion of primary treatment:

   1. presenting with symptoms suggestive of recurrence

   2. in the absence of symptoms (CT and/or MRI used for surveillance)

3. To determine the test accuracy of CT and/or MRI performed at the discretion of clinicians and of PET-CT (performed regardless of the result of initial imaging) in women with an initial diagnosis of stage IB–IVA cervical cancer, who are assumed to be disease free for a minimum of 3 months post completion of primary treatment:

   1. presenting with symptoms suggestive of recurrence

   2. in the absence of symptoms (CT and/or MRI + PET-CT used for surveillance).

Information on rate of recurrence in women post completion of primary treatment as distinct from rate of recurrence in women following imaging was absent in the literature reviewed. Elicitation of accuracy data was necessary because of a lack of disaggregation of women with and without symptoms in the literature and because of the very limited accuracy data available. Elicitation also provided the opportunity to investigate the coherence of subjective probabilities elicited with estimates in the literature.

Methods used

Subjective probabilities were elicited by two project members (CD and CM) during an educational meeting of the West Midlands Gynaecology Oncology Specialist Group on 1 July 2011 at the City Hospital, Birmingham, UK. Following the success of this initial elicitation, as judged by the face validity of the findings, the results were supplemented by purposive sampling by clinicians in the project team and by two further meetings – a gynae-oncology multidisciplinary meeting at Barts Hospital, London, UK, on 17 August 2011 and at the British Gynaecological Cancer Society Scientific Meeting at the International Convention Centre, Birmingham, UK, on 18 November 2011.

The initial elicitation exercise was preceded by a presentation outlining the aims of the project, the role of elicitation in the project, an overview of definitions of prevalence and test accuracy metrics to be elicited and a practice non-clinical elicitation exercise. Subsequent elicitations achieved by purposive sampling used a written description of the task and a printed elicitation example, except at the scientific meeting where a poster on the project was also displayed.

For the clinicians carrying out the first elicitations, the face-to-face pre-elicitation training, questions and discussion were conducted as a group to facilitate a common understanding of the problem and task and to allow participants to benefit from group discussion and interaction. Following the presentation and the non-clinical elicitation exercise (on estimated distance from London to Birmingham), participants were asked for written consent before undertaking the elicitation exercise. Participants were free to leave at any point in the exercise. Participants were instructed to undertake the elicitation exercise itself independently to ensure that variation within and across disciplines could be captured if there were sufficient numbers of respondents to allow subgroup analysis. In addition, mathematical aggregation (as opposed to behavioural aggregation) mitigates against the possibility of ‘consensus’ estimates being biased by the views of a minority. 44

The elicitation exercise comprised an 11-page anonymous self-administered questionnaire (see Appendix 7). The questionnaire included background information on the length of time that participants had practised in their speciality, their use of current imaging techniques and their use of PET-CT. To be eligible participants did not have to have hands-on experience of using PET-CT. Use of PET-CT is not routine in this patient group and beliefs are shaped by factors other than first-hand experience, such as interaction with colleagues, published estimates of accuracy and knowledge of the technology. In addition to the probabilities elicited, participants were also asked to state the minimum important clinical difference in accuracy between imaging with CT and/or MRI and imaging with CT and/or MRI with the addition of PET-CT that they would require before choosing to use one or other imaging strategy routinely.

Accuracy data were elicited in the form of the proportion of test errors (false-positives and false-negatives) that would be expected with the use of the combinations of imaging technologies outlined above. The choice of test errors as a metric of accuracy is based on research suggesting that test accuracy metrics with test result as reference class are more intuitive45 and that the clinical utility of a test is commonly conceptualised using test errors. 46 Test errors were used to derive positive predictive values (PPVs) and negative predictive values (NPVs). Elicited estimates of prevalence in combination with PPVs and NPVs were used to derive estimates of sensitivity and specificity for use in the economic model.

Elicitation of prevalence and test accuracy information was undertaken using the allocation of points technique whereby respondents are asked to indicate the likelihood of a value range being a true estimate by allocating a proportion of 100 points to that value range (the sum of allocated points across each value range summing to 100). In this way probability functions were obtained for each individual and were aggregated mathematically to derive an average distribution for the sample. An aggregated mean value was estimated using the average distribution and the midpoint of each value range. The variability of this aggregated mean was estimated by calculating the standard deviation (SD) across the value ranges. Microsoft Excel was used for calculations and graphical display of results.

Search strategy

A sensitive search was conducted to identify all relevant published and unpublished trials and trials in progress. All databases were searched from inception to August 2010. Search strategies were designed from a series of test searches. Both MeSH terms and text words were used and included a variety of synonyms for recurrent cervical cancer and the interventions (chemotherapy, radiotherapy, palliative treatment, surgery). Strategies for MEDLINE, EMBASE and The Cochrane Library can be found in Appendix 8. Trials were identified from several sources including:

• general health and biomedical databases: MEDLINE (Ovid), EMBASE (Ovid), CENTRAL

• database searches for systematic reviews, from which primary studies could be identified, including MEDLINE (Ovid), EMBASE (Ovid) and The Cochrane Library [Cochrane Database of Systematic Reviews (CDSR), Database of Abstracts of Reviews of Effects (DARE) and Health Technology Assessment (HTA) database]

• searches for studies in progress, unpublished research or research reported in the grey literature in a range of relevant databases including ClinicalTrials.gov and the UK Clinical Research Network Portfolio

• specialist search gateways (OMNI and the National Cancer Institute), general search engine (Google) and meta-search engine (Copernic) from March to May 2010

• hand-searches of Gynecologic Oncology from 1980 to May 2010

• reference lists of review articles and papers

• authors of the included studies, who were contacted for information on relevant published or unpublished studies.

Inclusion/exclusion criteria

Randomised controlled trials

Quality assessment of included RCTs was performed using the criteria outlined in the Cochrane Handbook for Systematic Reviews of Interventions. 37 Each study was assessed for adequate sequence generation, adequate allocation concealment, all methods of blinding used and whether or not they were effective, whether or not there was incomplete outcome data presented (attrition and exclusions from analysis), non-selective outcome reporting, and freedom from other biases. In all cases ‘yes’ indicated a low risk of bias and ‘no’ indicated a high risk of bias. ‘Unclear’ was used if there was insufficient detail reported. The quality of studies was summarised in tables, which were then used to create quality diagrams.

Case series

Quality assessment of case series was performed using the checklist developed by the National Institute for Health and Clinical Excellence (NICE). 47 Each study was then awarded an overall study quality grading for internal validity and an overall study quality grading for external validity:

• ++: all or most of the checklist criteria have been fulfilled; where they have not been fulfilled the conclusions are very unlikely to alter.

• +: some of the checklist criteria have been fulfilled; where they have not been fulfilled, or not adequately described, the conclusions are unlikely to alter.

• −: few or no checklist criteria have been fulfilled and the conclusions are likely or very likely to alter.

Population characteristics

The characteristics of the patient populations in the included studies are shown in Tables 5–7. The total number of patients in the studies ranged from 20 to 75 but some of the studies included any gynaecological cancers and others reported imaging results for both recurrent and primary cervical cancer. Therefore, the tables also report the number of patients with recurrent cervical cancer only and with imaging results. Many of the studies did not report summary patient characteristics for the patients with recurrent cervical cancer and imaging results only but for the full patient group, which is not relevant here and so has not been reported. When stated, most patients had squamous cell carcinoma; fewer had adenocarcinoma. In some studies, such as that by Chung et al. ,20 it was stated that histologically confirmed squamous cell carcinoma, adenocarcinoma or adenosquamous carcinoma of the uterine cervix was a requirement for study eligibility, but for others it was unclear.

All included studies except those by Mittra et al. 51 and Hatano et al. 53 described only women who had undergone treatment for histopathologically proven cervical cancer and who had suspected recurrence based on the presence of clinical signs and/or symptoms. The Mittra et al. study51 included both symptomatic and asymptomatic patients undergoing routine follow-up. The Hatano et al. 53 study verified whether MRI could provide accurate information to evaluate residual tumours after radiotherapy (persistent disease) and the MRI findings were compared with cytology/histopathology before and after radiotherapy.

Six studies20,49,50,52,56,58 described grounds on which the recurrence was suspected. Abnormal imaging and physical examination during follow-up were the main indications for performing PET-CT in the Chung et al. 20 study. Each patient in the Grisaru et al. 49 study had undergone a comprehensive evaluation of her clinical status and was scheduled for routine staging or follow-up imaging studies for suspected recurrence (but results were given only for suspected recurrence). Recurrence in Kitajima et al. 50 was suspected on the basis of physical examination, elevated levels of tumour markers and abnormal findings of conventional imaging, including CT and/or MRI, or an abnormal cervical smear. In Sironi et al. ,52 suspicion of tumour recurrence was based on follow-up procedures (physical examination, serum tumour markers and morphological imaging studies, such as CT or MRI). In Park et al. ,56 recurrence was suspected also on the basis of increased levels of serum squamous cell carcinoma antigen and carcinoembryonic antigen, pain in the lower abdomen and back, oedema of the lower leg and oliguria. The suspicion of recurrence in Williams et al. 58 was based on the clinical features of pelvic pain, vaginal discharge, vaginal bleeding, lower limb swelling or a palpable mass on pelvic examination.

Imaging characteristics

All six PET-CT studies20,48–52 were evaluations of PET-CT after patients had received conventional imaging (MRI and/or CT) or CT only. Of the PET-CT studies, only Amit et al. 48 focused on extracervical lesions, whereas the other five studies evaluated any recurrence. Only Park et al. 56 used CT to evaluate any recurrence and the other five MRI and CT studies evaluated local recurrence in the pelvis only.

All six PET-CT studies used ¹⁸F-FDG as a radioisotope tracer, with doses of 370–555 MBq,48 555–740 MBq,20 370–666 MBq,49 4.0 MBq/kg,50 400–555 MBq51 and 370 MBq. 52 The time between injection of ¹⁸F-FDG and the PET scan ranged from 30 minutes to 3 hours. The PET-CT scanning was performed mostly with a GE Discovery LS PET-CT scanner (GE Medical Systems, Milwaukee, WI, USA). In Amit et al. ,48 a hybrid PET-CT system combining a third-generation multislice spiral CT system [GE LightSpeed Plus (GE Medical Systems, Milwaukee, WI, USA)] with a dedicated full bismuth germanium oxide (BGO) ring PET scanner [GE Advance NXi (GE Medical Systems, Milwaukee, WI, USA)] was used. In Chung et al. 20 a GEMINI PET-CT system (Philips, Guildford, UK) was used, and in Kitajima et al. 50 all imaging and data acquisitions were performed with a Biograph Sensation 16 PET-CT scanner (Siemens Systems, Erlangen, Germany). Two studies48,50 measured glucose levels before administration of ¹⁸F-FDG.

In the three MRI studies53,54,58 T1-weighted spin-echo and T2-weighted turbo spin-echo were used. Of the four CT studies,55–58 two55,58 used optional intravenous contrast medium to elucidate problems identified on initial scans. Intravenous contrast medium was used routinely in the other two studies: non-ionic contrast (150 mg)56 and Reno-M-DIP® contrast (400 ml of 4% oral meglumine diatrizoate) (Squibb, Princeton, NJ, USA). 57

Quality of studies

The results of the quality assessment are provided in Table 8. Four studies48,49,52,53 collected patients' data prospectively (77 patients in total), seven studies20,50,51,54,56–58 collected data retrospectively (260 patients in total) and in one of the studies55 there was no information on the method of enrolment. Three studies20,51,52 clearly described their inclusion criteria such as presence of symptoms indicating recurrence, new lesions on surveillance imaging, elevated serum tumour markers with or without abnormal imaging and abnormal results on physical or cytological examination on routine surveillance. Relevant clinical information such as age, FIGO stage, histology type of tumour and primary treatment were described in all studies except for those by Amit et al. ,48 Grisaru et al. 49 and Park et al. 56

In all of the included studies the reference standard for diagnosis of cervical cancer was histopathology with or without clinical/radiological follow-up. Four of the studies48,53,57,58 used only histopathology as the reference standard, whereas in the other studies diagnosis was supported by clinical follow-up. Selection bias (using the imaging study being investigated as part of the inclusion criteria into the study) was present in at least four studies. 20,50–52

Information to judge the presence of incorporation bias (in which the index test forms part of the reference standard) was unclear in almost all of the studies, but in Kitajima et al. 50 the index test (PET-CT) was clearly part of the reference standard when the final diagnosis of 12 patients was based on the results of tumour marker level and PET-CT findings. Two studies reported the mean time between index test and reference standard, which was 2.3 weeks52 and 1 week. 57 Readers of PET-CT, MRI and CT studies were reported to be blind to patients' clinical details and final diagnosis in only four studies. 49,50,54,58

With regard to technical quality, the methods used in the more modern studies were similar to currently used imaging methods, whereas the methods used in the older studies were not. In the PET-CT studies there was slight variation found in whether or not and how much oral hydration was used as well as slight differences in acquisition times and injected doses. Chung et al. 20 used oral contrast for CT, but this should not affect the PET interpretation or results. Heron et al. 55 incorporated lymphangiography, which is now no longer used.

Test accuracy

The numerical results for all included studies are shown in Table 9.

Positron emission tomography/computerised tomography

Six PET-CT test accuracy studies were found. 20,48–52 Five studies20,49–52 evaluated local recurrence and distance metastasis and one study48 evaluated extrapelvic recurrence only. The sensitivities and specificities and their 95% CIs are shown in Figure 5 and a SROC space plot is shown in Figure 6. The sensitivities and specificities of local and distant recurrence were 83–100% and 71–100%, respectively, and the sensitivity and specificity of distant recurrence only were 86% and 100%. The summary estimates of the sensitivity and specificity of PET-CT for the detection of cervical cancer recurrence were 92.2% (95% CI 85.1% to 96.0%) and 88.1% (95% CI 77.9% to 93.9%), respectively. Sensitivity analysis, omitting one study48 that reported accuracy for distant recurrence only, did not affect accuracy estimates to any significant degree [sensitivity 92.6% (95% CI 85.3% to 96.4%); specificity 87.3% (95% CI 76.6% to 93.5%)]. The results tables of the univariate random-effects regression model for the meta-analysis and sensitivity analysis are in Appendix 11.

FIGURE 5.

Sensitivity and specificity of the PET-CT studies. FN, false-negative; FP, false-positive; TN, true-negative; TP, true-positive.

FIGURE 6.

Summary receiver operating characteristic space plot for the PET-CT studies.

Magnetic resonance imaging

Three MRI test accuracy studies were found and all evaluated the pelvis only. 53,54,58 Weber et al. 54 and Williams et al. 58 included women with clinical suspicion of recurrence and Hatano et al. 53 included women with residual, advanced-stage cervical cancer (stage IB2–IV). Previous treatment was radiotherapy in Hatano et al. 53 and Weber et al. 54 and 50% surgery and 50% radiotherapy in Williams et al. 58 All three studies investigated local recurrence in the pelvis only. Distant recurrence was noted in Williams et al. 58 (4/20), but these women were not included in the numerical results for sensitivity and specificity. Distant metastases are also mentioned in Hatano et al. 53 Because of clinical heterogeneity between these studies, no meta-analysis was conducted. The sensitivities and specificities and their 95% CIs are shown in Figure 7 and a SROC space plot in Figure 8. The sensitivities and specificities of MRI in pelvic recurrence varied between 82% and 100% and 78% and 100% respectively.

FIGURE 7.

Sensitivity and specificity of the MRI studies. FN, false-negative; FP, false-positive; TN, true-negative; TP, true-positive.

FIGURE 8.

Summary receiver operating characteristic space plot for the MRI studies.

Computerised tomography

Four CT test accuracy studies were found. 55–58 Heron et al. ,55 Walsh et al. 57 and Williams et al. 58 investigated local recurrence only, whereas Park et al. 56 investigated local and distant recurrence. [As mentioned in the MRI section, Williams et al. 58 also mentioned 4 (of 20) women with distant recurrence, who were not included in the sensitivity and specificity statistics.] There is little information available on the patients included in Heron et al. 55 and Walsh et al. 57 Also, both Heron et al. 55 and Walsh et al. 57 have equivocal results. For six patients in the Heron et al. 55 study, the CT findings were classified as equivocal; all of these patients had undergone radiotherapy, making differentiation between radiation fibrosis and recurrence difficult. For two patients in Walsh et al. ,57 CT images could not differentiate radiation sequelae from tumour. The sensitivities and specificities and their 95% CIs are shown in Figure 9 and a SROC space plot is shown in Figure 10. Because of clinical heterogeneity and lack of information about patients, no meta-analysis was conducted. The sensitivities and specificities of CT in pelvic recurrence (excluding the equivocal results) varied between 78% and 93% and 0% and 95% respectively. Sensitivity analysis around the equivocal results for Heron et al. 55 varied the sensitivity from 75% to 94% and the specificity from 82% to 95%. Sensitivity analysis around the equivocal results for Walsh et al. 57 varied the sensitivity from 87% to 94% and the specificity from 0% to 50%.

FIGURE 9.

Sensitivity and specificity of the CT studies. FN, false-negative; FP, false-positive; TN, true-negative; TP, true-positive.

FIGURE 10.

Summary receiver operating characteristic space plot for the CT studies.

Comparison of standard imaging followed by positron emission tomography/computerised tomography with standard imaging only

One study49 gave results in one table for both standard imaging alone and standard imaging with whole-body PET-CT with the same reference standard of histology or clinical evidence of disease, allowing comparisons to be made. Unfortunately, the part of the body imaged with standard imaging was not provided in the paper. The results are provided in Table 10. This shows that the PET-CT results are closer to the reference standard results than the standard imaging results.

Diagnostic and therapeutic impact

One included PET-CT study20 reported information on the diagnostic and therapeutic impact of the imaging. None of the included MRI or CT studies provided any details on whether or how the management of patients was altered by imaging.

In Chung et al. ,20 the mean age of patients was 53 years (range 32–77 years) and they had primarily stage I (50%) and stage II (40%) cancer. The results of PET-CT imaging were found to have an impact on the management of 12 patients (23%) by initiating previously unplanned treatment (four patients), changing the previously planned therapeutic approach (five patients) or eliminating a previously planned diagnostic procedure (three patients). The PET-CT led to additional invasive diagnostic procedures in nine patients: mediastinoscopic biopsy in three patients, PET-CT-guided pelvic lymph node biopsy in three patients, supraclavicular lymph node biopsy in two patients and bone biopsy in one patient. The PET-CT assisted in the planning of the therapeutic strategy in nine patients.

Chung et al. 20 also reported the prognostic outcomes of patients undergoing PET-CT giving 2-year disease-free survival rates and survival curves for women with positive and negative PET-CT results. The 2-year disease-free survival rates for women with a positive and a negative PET-CT result for recurrence were 10.9% and 85.0% respectively (p = 0002). The survival curves are reproduced in Figure 11.

FIGURE 11.

Two-year disease-free survival of patients with positive and negative PET-CT scans (from Chung et al. 20).

Accuracy

Individual respondents' accuracy results (PPVs, NPVs) for MRI and/or CT and for MRI and/or CT with PET-CT for symptomatic and for asymptomatic women are given in Tables 85–92 in Appendix 12. Note that PPVs are the proportion of women who test positive on either CT or MRI at the discretion of a clinician (and PET-CT if used and performed regardless of the result of initial imaging) who are confirmed as having recurrence of cervical cancer on the basis of histology, and NPVs are the proportion of women who test negative on either CT or MRI at the discretion of a clinician (and PET-CT if used and performed regardless of the result of initial imaging) who are confirmed as not having recurrence on the basis of clinical follow-up. Summary results are shown in Table 12. These are shown graphically in Figures 13 and 14 for symptomatic and asymptomatic women respectively.

FIGURE 13.

Elicited estimates of the accuracy of CT and/or MRI and CT and/or MRI with PET-CT in symptomatic women a minimum of 3 months after completion of primary treatment for cervical cancer.

FIGURE 14.

Elicited estimates of the accuracy of CT and/or MRI and CT and/or MRI with PET-CT in asymptomatic women a minimum of 3 months after completion of primary treatment for cervical cancer.

Minimum important clinical difference in accuracy between imaging with computerised tomography and/or magnetic resonance imaging and imaging with computerised tomography and/or magnetic resonance imaging plus positron emission tomography/computerised tomography

The average minimum important increase in accuracy from the addition of PET-CT to CT and/or MRI that was considered necessary to warrant introduction of PET-CT as a routine investigation in this sample of clinical experts was similar for asymptomatic and symptomatic patients: a 7.7% reduction in false-positives and a 6.4% reduction in false-negatives for symptomatic women and an 8.7% reduction in false-positives and a 6.3% reduction in false-negatives for asymptomatic women. Mean elicited estimates of the differences in test accuracy between CT and/or MRI and CT and/or MRI plus PET-CT were 2.6 and 3.6 for PPV and NPV, respectively, for symptomatic women and 4.6 and 3.4 for PPV and NPV, respectively, for asymptomatic women.

The results suggest that, in our sample of experts, the elicited increase in accuracy as a result of the addition of PET-CT to MRI and/or CT is smaller than the elicited minimum important difference in accuracy required to justify the routine addition of PET-CT for the investigation of women post completion of primary treatment for cervical cancer, that is, all of the differences in false-positives and false-negatives in Table 12 are smaller than the minimum important clinical differences listed in the paragraph above.

Comparison with systematic review results

Comparison of elicited estimates of accuracy with those reported in the literature are complicated because of the age of the CT and MRI studies, the lack of disaggregation between symptomatic and asymptomatic patients and a paucity of estimates of the combined accuracy of CT and MRI. Table 13 illustrates that elicited estimates of the accuracy of CT/MRI and CT/MRI plus PET-CT in symptomatic women are similar to estimates in the literature. For asymptomatic women the elicited specificities of CT/MRI and CT/MRI plus PET-CT are comparable to literature-based estimates but elicited estimates of sensitivity are lower. This is most likely to be a function of the spectrum of patients in included studies; inclusion of symptomatic patients in studies in the literature would be expected to result in higher sensitivity.

Effectiveness of single cisplatin agents

Quality of studies

All studies were RCTs with no blinding, but the description of randomisation was provided in only three trials. 71,78,80 In Long et al. ,71 patients were randomly assigned to the treatment regimens with equal probability using a fixed-block design; patients were stratified by treating institution only. In Moore et al. ,78 randomisation (with equal probability to each of the treatment arms) was carried out using a block design that balanced the sequence of assigned arms within parent institutions. In Omura et al. ,80 patients were prospectively stratified according to whether or not they had received previous radiation-sensitizer treatment (hydroxyurea, cisplatin or fluorouracil) and by Karnofsky performance score, and were then centrally randomised with equal probability to three groups.

Description of allocation concealment was not reported in any of the included studies. Several studies had methodological ambiguities. In Alberts et al. ,60 one of the treatment arms, cisplatin, was dropped early because of poor accrual; the number of patients in the cisplatin group was much lower than in the other two groups (9 vs 54 and 51). In Cadron et al. ,65 the intention had been to include 200 patients in the trial but because of poor accrual the trial was stopped prematurely and only 24 patients were included. In Long et al. ,71 the methotrexate, vinblastine, doxorubicin and cisplatin (MVAC) arm was closed by the Data Safety Monitoring Board after four treatment-related deaths occurred among 63 patients, and results for the MVAC arm were not reported. In Vermorken et al. ,82 45 patients from the cisplatin group received bleomycin, vindesine, mitomycin and cisplatin (BEMP) as second-line treatment. The quality assessment results are shown in Figures 16 and 17.

FIGURE 16.

Methodological quality on individual items for the eight included RCTs: single-agent cisplatin.

FIGURE 17.

Summary of the quality and reporting assessment of the eight included RCTs: single-agent cisplatin.

Effectiveness of cisplatin combinations

Quality of studies

Only two studies63,76 specified the method of randomisation in their reports. Description of allocation concealment was not reported in any of the included RCTs. Until January 2004, the Monk et al. 76 study consisted of only two arms comparing cisplatin plus paclitaxel with cisplatin plus vinorelbine. Primary analyses excluded those 41 patients. In Bezwoda et al. ,62 after a preliminary analysis of the results, the hydroxyurea arm of the study was discontinued and a further 25 patients received the cis-diamminedichloroplatinum(II) (DDP) plus methotrexate regimen. Figures 18 and 19 show the results of the quality assessment.

FIGURE 18.

Methodological quality on individual items for the four included RCTs:62,63,76,79 cisplatin combinations.

FIGURE 19.

Summary of the quality and reporting assessment of the four included RCTs:52,63,76,79 cisplatin combinations.

Effectiveness of other platinum agents

Quality of studies

The description of randomisation was provided only in McGuire et al. 74 None of the trials gave any details of the method of allocation concealment. In the study by Lira-Puerto et al.,68 accrual was suspended in two institutions because of termination of support. The results of the quality assessment are provided in Figures 20 and 21.

FIGURE 20.

Methodological quality on individual items for the three included RCTs:68,74,81 other platinum agents.

FIGURE 21.

Summary of the quality and reporting assessment of the three included studies:68,74,81 other platinum agents.

Effectiveness results

Overall and progression-free survival

All RCTs reported overall survival and two reported progression-free survival (Tables 36 and 37). There was little difference in overall survival or progression-free survival between arms in each RCT. Hazard ratio results were not supplied for any of the included studies.

TABLE 36 - Overall survival: other platinum agents

CHIP, iproplatin; NR, not reported; OS, overall survival; T, teniposide.

Weeks.

Study	Comparison	Median OS, months (range)	Hazard ratio (95% CI)
McGuire 198974	CBDCA	6.2	NR
	CHIP	5.5
Lira-Puerto 199168	CBDCA	7.5	NR
	CHIP	7.6
Thomsen 199881	CBDCA	40 (20–49)a	NR
	T	41 (34–56)a

TABLE 37 - Progression-free survival: other platinum agents

CHIP, iproplatin; NR, not reported; PFS, progression-free survival; T, teniposide.

Weeks.

Study	Comparison	Median PFS, months (range)	Hazard ratio (95% CI)
McGuire 198974	CBDCA	2.7	NR
	CHIP	3
Thomsen 199881	CBDCA	20 (11–31)a	NR
	T	17 (12–32)a

Response rate

Two of the RCTs68,74 gave response rates, complete response rates and partial response rates for the same treatment comparisons and so meta-analysis was possible (Figures 22–24). There were no statistically significant differences in terms of frequency of response rate (overall, partial, complete) between these cisplatin agents. However, it should be noted that there were differences between studies in the frequency of response rate in the CBDCA arms (ranging from 15% to 33%) as well as in the iproplatin (CHIP) arms (11–30%), and in the frequency of partial response rate in the CBDCA arms (10–33%) and the CHIP arms (7–25%).

FIGURE 22.

Response rate: other platinum agents.

FIGURE 23.

Complete response: other platinum agents.

FIGURE 24.

Partial response: other platinum agents.

Adverse events

Data on haematological toxicity was supplied for all trials. Lira-Puerto et al. 68 reported drug reactions of Eastern Cooperative Oncology Group (ECOG) grade 2 or more. Meta-analysis of thrombocytopenia rates comparing CBDCA with CHIP indicated no statistical differences between chemotherapeutic agents (Figure 25). There were no differences in leucopenia rates in two RCTs (CBDCA arm 17/176 vs CHIP arm 8/180;74 CBDCA arm 0/12 vs teniposide arm 1/1481). Neurological adverse events (grades 2, 3 or 4) were seen in 1 out of 47 patients treated with CBDCA and 6 out of 41 patients treated with CHIP in the study by Lira-Puerto et al. 68 and in 6 out of 176 and 6 out of 180 patients, respectively, in the study by McGuire et al. 74 Gastrointestinal adverse events such as nausea and vomiting were experienced less often in patients receiving CBDCA (57/176) than in patients receiving CHIP (95/180) [RR 0.61 (95% CI 0.48 to 0.79)] in the study by McGuire et al. ,74 but there was no difference in gastrointestinal adverse events between CBDCA (2/12) and teniposide (2/14) in the study by Thomsen and Pfeiffer. 81

FIGURE 25.

Thrombocytopenia: other platinum agents.

Effectiveness of non-platinum agents

Characteristic of included studies

Four studies gave evidence on the effectiveness of non-platinum agents for the treatment of recurrent, persistent or advanced cervical cancer. 61,67,75,83 Because two studies67,83 were, unfortunately, impossible to obtain, analysis was conducted on the basis of the systematic review by Hirte et al. 59 Baseline characteristics including previous treatment and stage and site of disease are presented in Table 38; however, not all relevant clinical information was presented in all publications.

TABLE 38 - Characteristics of the populations in the included RCTs: non-platinum agents

BLEO, bleomycin; Cyclo, cyclophosphamide; L, lapatinib; NR, not reported; P, pazopanib; V, vincristine.

45 patients underwent previous radiotherapy and/or chemotherapy.

Patients with disease no longer amenable to therapy with surgery or irradiation, or who represented failures of chemotherapy with agents considered effective for their disease.

Parameter	Barlow 197361			Greenberg 197767		Monk 201075			Wallace 197883
	ADM	BLEO	ADM + BLEO	ADM	ADM + BLEO	P	P + L	L	ADM	ADM + V	ADM + Cyclo
Number of patients (randomised)	21 (+ two patients mistakenly randomised)	10	23	9	11	74	76	78	61	61	52
Age (years), median (range)	54.5 (13–83)			NR	NR	49.5 (29–73)	48.5 (28–82)	49 (23–81)	NR	NR	NR
Previous treatment
Chemotherapy	NRa	NRa	NRa	11%	18%	5	3	5	NR	NR	NR
Radiotherapy	NRa	NRa	NRa	100%	100%	24	24	18	NR	NR	NR
Surgery	NR	NR	NR	NR	NR	1	0	1	NR	NR	NR
Chemotherapy and radiotherapy	NR	NR	NR	NR	NR	26	30	30	NR	NR	NR
Surgery and radiotherapy	NR	NR	NR	NR	NR	0	2	2	NR	NR	NR
Stage
IVB	NRb	NRb	NRb	100%	100%	6	6	3	100%	100%	100%
Persistent	NRb	NRb	NRb	100%	100%	NR	NR	NR	100%	100%	100%
Recurrent	100%	100%	100%	100%	100%	95%	95%	92%	100%	100%	100%
Site of disease
Pelvic	NR	NR	NR	0%	45%	NR	NR	NR	NR	NR	NR
Distant	NR	NR	NR	22%	0%	NR	NR	NR	NR	NR	NR
Both	NR	NR	NR	77%	55%	NR	NR	NR	NR	NR	NR

Quality of studies

As full texts were impossible to obtain for Greenberg et al. 67 and Wallace et al. ,83 the quality assessment was based on the systematic review by Hirte et al. 59 A description of the allocation concealment procedure was not reported in any of the RCTs and blinding was not used in any of the RCTs. In Barlow et al. ,61 two patients with squamous cell tumours were mistakenly randomised as non-squamous cell tumours and received adriamycin (ADM) alone (group of 21 + 2 patients). In Monk et al. ,75 patients were initially randomly assigned to combination and monotherapy arms. The protocol was later amended after receiving results of a formal interim analysis and combination therapy was discontinued. In the same study, the unconfirmed response rate (not verified on a second scan) was 19% for pazopanib-treated patients and 9% for lapatinib-treated patients. The results of the quality assessment are shown in Figures 26 and 27.

FIGURE 26.

Methodological quality on individual items for the four included RCTs:61,67,75,83 non-platinum agents.

FIGURE 27.

Summary of the quality assessment of the four included RCTs: 61,67,75,83 non-platinum agents.

Study selection

Included in this section are studies in which participants have recurrent or persistent cervical cancer that was initially treated with surgery and who now have evidence of recurrence. The interventions are radiotherapy or radiotherapy with chemotherapy (chemoradiotherapy). The search for relevant studies did not identify any RCTs but 16 case series met the inclusion criteria: nine evaluated radiotherapy84–92 and seven evaluated chemoradiotherapy. 93–99 The results of the quality assessment for the radiotherapy and chemoradiotherapy studies are presented in Appendix 15.

Study selection

Included in this section are studies in which participants have recurrent or persistent cervical cancer that was initially treated with radiotherapy or chemoradiotherapy and who now have evidence of recurrence. The interventions are radical hysterectomy or Wertheim's operation and pelvic exenteration, and these two categories are described separately. The search found no relevant RCTs. Twenty-seven case series100–126 fulfilled the inclusion criteria, most of which were retrospective, based on chart reviews. Most of the excluded papers were case series of gynaecological cancers as a whole without giving separate results for cervical cancer patients, or studies of cervical cancer patients with primary and recurrent tumour characteristics with the results described together. The results of the quality assessment are presented in Appendix 16. No measure of quality of life was provided in any of the included studies.

Statement of principal findings

Accuracy and effectiveness inputs to the economic evaluation

Model assumptions

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

1. Women are followed up with examinations every 3 months for 2 years, then every 6 months for 2 years and then annually for 1 year, with the total follow-up being 5 years.

2. Women who were symptomatic at 3 months and whose cancer has not been detected cannot become asymptomatic.

3. Women with symptoms that they suspect are related to cervical cancer are usually given an urgent appointment or their pre-existing follow-up appointment is brought forward.

4. The sensitivity and specificity of the confirmatory biopsy test were 100% accurate.

5. Women who previously received chemoradiotherapy for primary cervical cancer and who are not diagnosed with persistence at 3 months' follow-up (i.e. not persistent or persistent cases missed) will be treated similarly to women with recurrent cervical cancer when detected.

6. The PET-CT procedure includes both the preparation and the scan of the patient; therefore, the preparation activity is implicit in the PET-CT scan resource use [NHS Reference Cost Team (anonymous) by email, pbrdatacollection@dh.gsi.gov.uk, 12 April 2011, personal communication].

7. Women who received treatment for primary cervical cancer and who have not survived at 5 years have died from recurrent cervical cancer only.

8. The utility for recurrent cervical cancer is equivalent to the average of the utilities for primary stage III and stage IV cervical cancer.

9. There is a constant hazard over 5 years for early-stage recurrent cervical cancer (i.e. the risk of recurrence is the same at 4 years as it is at 1 year).Women treated for recurrent cervical cancer will have the same quality of life following treatment as they had after treatment for initial cervical cancer.

Rates of recurrent cervical cancer

The model was populated with the rates of recurrent cervical cancer derived from the literature and in consultation with clinical experts. The rates of recurrence were calculated using a two-stage process. First, the survival following treatment for primary cervical cancer was derived from disease-free survival curves from Landoni et al. ,135 progression-free survival curves from Keys et al. 136 and overall survival curves following initial treatment from Landoni et al. 135 and Vale et al. 137 Information from these curves was used with the standard assumption of an exponential survival function. Three-month survival results were calculated for women who received surgical treatment, based on the disease-free survival curve presented in Landoni et al. 135 (Figure 29). Similar procedures were used to calculate survival following postoperative chemoradiotherapy and chemoradiotherapy. Second, the rates of recurrence were calculated, based on the initial survival of women in the branch of women who were symptomatic without cancer (see probabilities f1–f4 in Tables 98–101 in Appendix 18), using the conditional probabilities following survival and the formulae presented in Table 57. Table 58 shows the rates of recurrence used in the models.

FIGURE 29.

Disease-free survival from surgery and radiotherapy for stage IB–IIA cervical cancer (after Landoni et al. 135).

Women enter the model at 3 months after initial treatment. If they have cancer at 3 months they enter the state ‘Asymptomatic cancer at 3 months’ or ‘Symptomatic cancer at 3 months’, but if they are free of cancer at this time they enter the state ‘Asymptomatic without cancer’ or ‘Symptomatic without cancer’. In effect, the 3 months before entry in the model can be regarded as being represented by the probability tree shown in Figure 30.

FIGURE 30.

Initial 3 months following treatment (before entry into the model).

Ideally, a separate data source would have been used to determine the proportions of women in each of these four states at the start of the model. In the absence of such a data source, it was necessary to make an assumption about these proportions. It was decided to use the probabilities for women moving from the state ‘Symptomatic without cancer’ (see Tables 98–101 in Appendix 18 for probabilities f2–f4) to give the necessary proportions. The way this was carried out is further detailed in Table 57.

Test accuracy results

Test accuracy results used in the model were based on the values estimated in the subjective elicitation exercise (see Chapter 5 and Table 14). The predictive values, 95% CIs and probability distributions for MRI and/or CT and for PET-CT are shown in Tables 59 and 60 respectively. Using the appropriate formulae, predictive values were converted to sensitivities and specificities to be used in the models. Table 61 shows the accuracy data used in the base-case analysis. For sensitivity analysis, the uncertainty indicated in Tables 59 and 60 was applied to the predictive values before conversion to sensitivity and specificity.

Survival following treatment

The results used in the model for survival following treatment for recurrence or persistence are shown in Table 62. Survival was reported in the systematic review in Chapter 6. From the systematic review, survival data from studies that followed up women following treatment for recurrent cervical cancer prior to 1990 were excluded. In cases in which 5-year survival after 1990 was not reported,96 2- and 3-year survival data were used. Note that these overall survival results are not given separately by the four FIGO stages.

From the results in Table 62, a weighted average of the 3-month survival following treatment was calculated, using weighting based on the percentages of people receiving the different treatments. Table 63 shows the 3-month survival data used in the models.

Costs and resources

The costs of resources used were those that were directly incurred by the NHS. Costs for clinical examination, diagnostic imaging (PET-CT, MRI and CT), confirmatory biopsy and treatment were included (Table 64). Costs that were not considered were those incurred during the primary diagnosis and treatment of cervical cancer. Other costs not included were those for long-term and end-of-life care. In the models, recurrence was assumed to occur only once. Diagnostic procedure costs were taken from the NHS Reference Costs 2009–2010. 139 Cost estimates for chemotherapy and radiotherapy treatment were taken from Clark et al. ,140 and estimates for chemoradiotherapy were taken from Clark et al. 140 and were adjusted to 2010 prices using the Hospital and Community Health Services combined pay and price inflation index. 141 Estimated costs for the diagnosis of recurrent cervical cancer included costs for clinical examination, PET-CT, MRI and CT. These cost estimates were taken from the NHS Reference Costs 2009–2010139 and published sources. 141 As a result of a paucity of cost-effectiveness studies comparing PET-CT as an adjunct with standard practice, an additive procedural cost of PET-CT as an adjunct to standard practice was assumed, as shown in Table 64. All costs were adjusted to 2010 prices and were discounted at 3.5% per annum.