Positron emission tomography (PET) and magnetic resonance imaging (MRI) for the assessment of axillary lymph node metastases in early breast cancer: systematic review and economic evaluation

Breast cancer and axillary metastases

Breast cancer is the most common type of cancer in women, with 38,048 new cases registered in women in England1 and 2457 in Wales2 in 2007. The usual site of spread outside the breast in newly diagnosed cases is to the lymph nodes in the axilla (underarm). The presence of axillary metastases and the extent of their spread are important prognostic factors for staging disease and planning treatment, whilst removal of any spread is essential to prevent recurrence and wider metastatic spread.

Aetiology, pathology and prognosis

Epidemiology and incidence

Incidence varies most with gender. Women are far more likely to get breast cancer than men. For both men and women, incidence also varies with age (Table 3). Approximately 81% of cases occur in women aged 50 years and over. 22

Incidence varies with ethnicity. Asian, Chinese and Black ethnic groups and those with mixed heritage have a lower incidence than the white ethnic group in England. The rate ratios are 0.65, 0.75, 0.49 and 0.58, respectively, when compared with the white group. 23 Incidence is generally > 80 in 100,000 in developed regions compared with < 30 per 100,000 in developing regions of the world. 24 Similarly, within the developed world, incidence varies with socioeconomic status. In both England25 and Wales26 those who are classed as most deprived have a lower incidence of breast cancer. However, there is some evidence to suggest that the trend for mortality is reversed, with better survival for those from more affluent areas. It is unclear why this is, but may be due to lower levels of screening compliance, worse overall general health status and lower levels of treatment due to limited access to health care27 and poorer compliance with treatment regimens.

Significance in terms of ill-health (burden of disease)

Breast cancer is a significant cause of death in England and Wales, ranking among the five leading causes when age-standardised rates are compared. It is currently the second biggest cause of cancer death in women after lung cancer, with an age-standardised mortality rate of 26 per 100,000 women. In 2008 this constituted 10,716 deaths for women in England and Wales. 28 Breast cancer deaths have been steadily falling since a peak in 1988. 29 The fall in mortality may be due to screening (earlier detection and more successful treatment), improvements in social awareness, diagnostic techniques and treatment options such as tamoxifen, chemotherapy and more recently trastuzumab. 30,31 As screening programmes detect more cancers at an earlier stage and women live longer after diagnosis, the long-term morbidity associated with treatment, such as lymphoedema, is becoming more significant.

Measurement of disease

Breast cancer has few obvious symptoms and can easily go undetected for many years. Among the more noticeable symptoms is a palpable lump in the breast, a change in breast shape and skin appearance or changes to the nipple such as inversion, a rash or discharge. Women are encouraged to be breast aware, and to seek medical advice if they notice anything unusual. 32 Screening was introduced in the UK in 1988. 33 Currently, women between the ages of 50 and 70 years are routinely invited to attend. Screening is thought to have reduced breast cancer deaths in the 55–69 years age category by an estimated 6.4% in addition to the effects of tamoxifen, chemotherapy and earlier presentation outside of screening. 30 Screening increases the proportion of tumours detected in the early, more curable stages.

A suspicious breast mass may be identified through screening, or via presentation to a general practitioner. The breast mass and axillary areas are investigated clinically through palpation and mammography or ultrasound for younger women, and the status of the tumour confirmed by histology of biopsied tissue. Staging of the disease depends on tumour size, the number of involved lymph nodes and the presence or absence of distant metastases. Tumour size and axillary metastases can be estimated by clinical examination and imaging techniques, but definitive status is achieved through surgery. Those with small tumours and no axillary metastases have the best prognosis, while those with distant metastases are considered incurable.

Current methods for staging of breast cancer

Three main factors are used to stage breast cancer. These are tumour size, metastases to the regional lymph nodes and distant metastases. The tumour/node/metastases (TNM) staging system was developed and is maintained by the American Joint Committee on Cancer and the Union for International Cancer Control. 34,35 T stage is classified according to size of the tumour and degree of local infiltration; N stage is classified according to the number and location of metastases to the lymph nodes in the axilla, between the ribs (internal mammary nodes) and above or below the collarbone (supraclavicular and infraclavicular nodes); and M stage is classified by the presence of metastases beyond the breast and regional lymph nodes (Table 4). The overall TNM stage of the cancer is defined as in Table 5.

Early breast cancer is generally defined as cancer which has not spread beyond the breast or the ipsilateral axillary lymph nodes and is confined to stages I, II or IIIA. 36–38

Current methods for assessment of axillary metastases

Axillary lymph nodes are the most common site of spread outside the breast (occurring in approximately 41% of cases)17 and removal of lymph nodes affected by tumour is crucial to prevent recurrence. Assessment of whether cancer has spread to the lymph nodes is also important for staging, assessing prognosis and selection of adjuvant therapy. Clinical examination via palpation has a sensitivity of approximately 46% based on pooled data from a number of studies (in other words, of those patients with metastases, 46% can be detected via clinical examination). 41–46 Therefore, of all patients presenting with breast cancer, approximately 19% (41% × 46%) have axillary metastases that can be detected via clinical examination, while 22% (41% × 54%) have occult axillary metastases.

The following steps (see also Figure 1) for assessing the axilla are recommended in the 2009 National Institute for Health and Clinical Excellence (NICE) breast cancer guideline. 17 A clinical examination and an ultrasound scan are carried out. If these examinations suggest nodal metastases on the basis of size or abnormal morphology, ultrasound-guided needle biopsy [either fine-needle aspiration cytology (FNAC) or core biopsy] of abnormal nodes is undertaken, which detects 45% of axillary node metastases. 47 For patients shown by needle biopsy to have nodal metastases, standard management is surgery to remove all the lymph nodes in the axilla, known as axillary lymph node dissection (ALND) or axillary clearance.

FIGURE 1.

Diagnostic pathway for axillary metastases as recommended in the 2009 NICE breast cancer guidelines. 17 *Either fine-needle aspiration cytology or core biopsy.

As well as being the treatment for those with positive nodes, ALND has been considered the ‘gold standard’ procedure for staging the axilla. Typically, 10–15 lymph nodes are removed and at least one section from each is assessed via haematoxylin and eosin (H&E) staining. ALND is very accurate in establishing the presence of axillary disease and has the therapeutic advantage of being associated with a high long-term local disease control rate. However, ALND is associated with significant complications, including a 21% incidence of arm lymphoedema,48–50 a 22% incidence of seromas48,51 and a 14% infection rate. 48,52 In addition, insertion of a surgical drain during surgery is commonplace (79%) and usually necessitates prolongation of hospital stay. 52

For patients with no evidence of lymph node involvement on ultrasound, or for whom the ultrasound-guided biopsy is negative, surgery to remove a sample of axillary lymph nodes is recommended, as opposed to full axillary clearance. 17 There are two axillary sampling techniques in current practice: sentinel lymph node biopsy (SLNB) and 4-node sampling (4-NS). SLNB is a procedure to identify and remove the sentinel lymph node, which is the first axillary lymph node to which lymphatic fluid drains from the breast, and therefore the most likely to be affected by metastases. The sentinel node may be identified via blue dye, radioactive isotope or a combination. Sentinel lymph nodes may be examined via H&E staining, and/or immunohistochemistry for epithelial or cytokeratin markers. 4-NS involves a more random surgical removal of a minimum of four lymph nodes (sometimes more) from the lower axilla, and may involve use of blue dye to aid in identification of nodal tissue in the axilla. The false-negative (FN) rate of SLNB and 4-NS have been estimated at between 0% and 10%. 17 As stated in the NICE guideline, data on 4-NS are limited and there are currently insufficient data to compare the diagnostic accuracy of SLNB and 4-NS. 17

Sentinel lymph node biopsy and 4-NS involve shorter surgical procedures than ALND, and are associated with lower incidence of surgical complications and long-term adverse effects than ALND. Lymphoedema incidence falls from 21%48–50 for ALND to 7%53 for SLNB, seroma incidence falls from 22%48,51 to 7%48,51, surgical drain requirement from 79% to 2%52,54 and infection incidence from 14%48,52 to 2%. 48,52,54

Patients in whom axillary lymph node metastases are identified via SLNB or 4-NS are advised to undergo ALND as a subsequent procedure to remove all axillary lymph nodes. 17 Some units are investigating the use of intraoperative cytology for assessment of axillary lymph nodes, whereby patients undergo axillary sampling (e.g. via SLNB) and the dissected lymph nodes are assessed while the surgical procedure is still ongoing. If any metastases are identified in the sampled nodes then the procedure is converted to a full ALND, bypassing the requirement for a second surgical procedure. However, it is currently unclear whether immediate or delayed ALND differ significantly in terms of adverse effects. Since intraoperative cytology is not currently used as standard, it is not included in this assessment.

Cost of current methods for assessment of axillary metastases

The costs of clinical examination, ultrasound, and ultrasound-guided biopsy are £86, £53 and £147, respectively55,56 ALND when carried out as a stand-alone procedure costs £2448. 56 All costs have been adjusted to 2007 prices.

Sentinel lymph node biopsy and 4-NS are normally performed at the same time as the main breast cancer surgery while ALND may be performed at the same time as the breast surgery or as a stand-alone surgical procedure. No UK costs are identified from previous studies regarding the combined costs of SLNB, 4-NS and ALND when performed at the same time as the breast surgery. However, it is widely accepted that SLNB has a significantly higher cost than 4-NS.

Relevant national guidelines

The 2009 NICE guideline Early and locally advanced breast cancer: diagnosis and treatment includes recommendations on methods for assessment of the axilla. 17

Variation in services and uncertainty about best practice

Until recently, ALND was the gold standard technique for assessing axillary lymph node status. A 2006 audit of 271 UK breast surgeons, asked how they would manage a woman with small, clinically node-negative breast cancer, reported that 27% performed ALND, 21% used 4-NS and 52% used SLNB. 57 The 2009 NICE guideline recommends that axillary sampling techniques (SLNB or 4-NS) should be used instead of ALND for patients with clinically node-negative disease, stating that SLNB is the preferred technique. 17 However, there are insufficient data to allow comparison of SLNB and 4-NS, either in terms of diagnostic accuracy or complication rate. 17 Therefore, there is likely to be variation in practice within the UK in the use of ALND, SLNB or 4-NS to assess axillary status.

Summary of diagnostic tests under assessment (index tests)

This review assesses two imaging techniques: positron emission tomography (PET) and magnetic resonance imaging (MRI).

Identification of important subgroups

Subgroup analyses were undertaken according to the different methods of PET and MRI, the reference standard test used, and the clinical characteristics of included patients (see Chapter 2, Decision Problem for a full list). Sensitivity analyses were also undertaken according to study quality criteria.

Current usage in the NHS

In recent years, imaging techniques such as PET and MRI have been increasingly used for diagnosis and staging in various types of cancer. However, at present they are not routinely used for staging the axilla in breast cancer.

Anticipated costs associated with index tests

The cost of MRI is £232. 56 The cost of PET is £978 according to a study based on UK hospital. 66 All costs have been adjusted to 2007 prices.

Population and relevant subgroups

Diagnostic tests under assessment (index tests)

The index tests assessed in this review are:

• PET, including:

  1. – ET alone

  2. – PET/CT

• MRI, including:

  1. – gadolinium-enhanced MRI

  2. – dynamic gadolinium-enhanced MRI

  3. – USPIO-enhanced MRI

  4. – MR spectroscopy.

Reference standard tests (comparator tests)

The most relevant reference standard tests were considered to be ALND, SLNB and 4-NS. ALND is the ‘gold standard’ method of staging the axilla. SLNB and 4-NS were also thought to be acceptable comparators, since these techniques are now recommended by NICE and are not associated with a survival detriment in the long term. 17 Studies using other reference standard tests, or where the reference standard is not clear, were included, but a sensitivity analysis excluding these studies was also undertaken.

Outcomes

Relevant outcomes include:

• sensitivity and specificity [or data required to calculate these, i.e. numbers of true-positive (TP), false-negative (FN), true-negative (TN) and false-positive (FP) results]

• adverse effects and withdrawals

• health-related quality of life

• cost-effectiveness and cost–utility.

Studies were only included if they report the numbers of TP, FN, TN and FP results for PET or MRI scanning in comparison with a reference standard test. These values can be used to calculate measures of diagnostic accuracy such as sensitivity and specificity.

Patients are classified by the index test (the diagnostic test being investigated) as either positive (axillary metastases) or negative (no axillary metastases). The reference standard (an established diagnostic test) is also undertaken to identify patients’ true health status. The reference standard is assumed to have 100% sensitivity and specificity; however, subgroup analyses are undertaken to test the effect of using different reference standards. Patients fall into one of four groups. Where the index test is positive, patients may be TP where both tests agree that they have metastases, or FP where the index test indicates that they have metastases but the reference standard does not. Where the index test is negative, patients may be TN where both tests agree they are metastasis free, or FN where the index test incorrectly classifies them as metastasis free. This can be represented in a 2 × 2 table (Table 6). In the clinical setting, FPs can result in patients receiving unnecessary treatment, while FNs can result in people not receiving treatment they require. Sensitivity indicates the effectiveness of the index test in correctly identifying metastases (TPs divided by all persons with metastases). Specificity indicates the effectiveness of the index test in correctly classifying people as metastasis free (TNs divided by all persons without metastases). Sensitivity and specificity can be calculated as simple percentages, as shown in Table 6. In practice, diagnostic tests often have a high sensitivity at the expense of a low specificity and vice versa. Ideally, a test would have both high sensitivity and high specificity.

Study design

Studies of a cohort design (prospective or retrospective) were included. Studies that were both prospective and consecutive were examined in a separate sensitivity analysis as part of the assessment of study quality. Case–control studies (where the test is evaluated in a group of patients already known to have the outcome and a separate group of patients without the outcome) were excluded; however, no studies of this type were identified within this review.

Identification of studies

Inclusion and exclusion criteria

Data extraction strategy

Data were extracted by one reviewer using a standardised data extraction form and checked by a second reviewer. Discrepancies were resolved by discussion.

Critical appraisal strategy

Study quality was considered with the aid of the quality assessment of diagnostic accuracy studies (QUADAS) checklist. 69 Quality assessment was performed by one reviewer and checked by a second. The quality assessment criteria were scored as detailed in Appendix 2. Three items from the published checklist were not used within our assessment, as follows. The ‘description of selection criteria’ item was omitted as this was thought to be covered by the ‘representative patient spectrum’ item, where studies with insufficient description of the selection criteria scored ‘unclear’. The ‘partial verification bias’ item was omitted because, in almost all studies included in this review, all patients received a reference standard. The ‘incorporation bias’ item was also omitted because, in this review, the reference standard was always independent of the index test (it was noted by Whiting et al. 69 that the above two items are not relevant in every review).

Methods of data synthesis

Sensitivity and specificity are presented for each study. Meta-analysis was undertaken to calculate a mean sensitivity and specificity across studies. Sensitivity and specificity are linked, so that changing the threshold at which a test is considered positive will tend to increase the sensitivity but decrease the specificity, or vice versa. Therefore, sensitivity and specificity were meta-analysed using a bivariate random effects method within stata (StataCorp^©, College Station, TX, USA). This approach assumes a bivariate normal distribution for the logits of sensitivity and specificity, which allows the correlation between them to be accounted for in the meta-regression model; covariates may be used to adjust the (marginal) logits of both sensitivity and specificity. 70,71 Where significant heterogeneity was observed, the random effects method was used in order to account for variation both within and between studies. Forest plots and receiver operating characteristic (ROC) plots were generated within review manager (version 5.0, Cochrane Collaboration^©, Copenhagen, Denmark). 72 To explore possible sources of bias, all study quality variables were added as covariates in univariate regression models for sensitivity and specificity for PET and MRI, in order to test whether any variables had a significant effect (p < 0.10) on sensitivity or specificity.

Quantity and quality of research available

The search identified 658 citations (646 from the literature search and 12 from other sources such as relevant reviews; Figure 2). Of these, 520 were excluded at the title/abstract stage and 138 were obtained for examination of the full text. Ninety-three citations were excluded at the full text stage (Appendix 3). In total, 45 citations relating to 35 studies were included in the review: 26 studies of PET and nine studies of MRI.

FIGURE 2.

Preferred reporting items for systematic reviews and meta-analyses flow chart of included and excluded studies.

Study quality

Figures 3 and 4 provide an overview of the methodological quality of the 35 included studies. 41,42,44,64,73–97,108–113 Of the PET studies, quality items that scored poorly overall were representative patient spectrum, availability of relevant clinical information, handling/reporting of uninterpretable results and interpretation of the reference standard with blinding to index test results. Patient spectrum scored negatively either because the study included up to 20% participants who were not early stage or newly diagnosed (12 studies)73,74,77–79,81,82,90,91,93,95,96 or because the patients were not recruited both prospectively and consecutively (two studies). 76,85 It was unclear if participants were recruited prospectively and consecutively in 12 cases. 73,78,81,83,84,89,91,93–97 The index test was interpreted blind to reference standard results in most studies. The index test was often interpreted blind to other clinical data in addition to the reference standard, possibly to ensure a robust evaluation of PET; however, as this is likely to differ from expected clinical practice, studies where this occurred were scored negatively for the ‘relevant clinical information’ item. Uninterpretable results were dealt with well in only eight studies,75,77,79,80,86–88,92 and were not discussed in 16 cases. 73,74,76,78,81–85,87,89–91,93,94,96 Similarly, blinding of the reference standard results was under-reported, with 20 studies scoring ‘unclear’. 73–75,77,78,80–85,87–89,92–96 It is difficult to know what impact these have on study quality and transferability to real life practice. The reference standard was adequate in nearly all cases, with only three studies failing to give sufficient detail,85,90,93 and two not performing ALND on those patients with large or locally advanced disease. 94,96 The delay between reference standard and index test was acceptable in 14 studies44,76,78,79,83–88,91,92,94,95 and not reported in 12. 73–75,77,80–82,87,89,90,93,96 Where patients were given a different reference standard depending on the index test results (six studies),75,76,78,79,81,83 this was either ALND or SLNB. SLNB is thought to have slightly lower accuracy so differential verification bias may occur in these studies. The index test was usually very well described, while the reference standard was often only partially described, probably due to the widespread routine use of these techniques. Most studies did not have any withdrawals to explain, so this item scored well overall.

FIGURE 3.

Methodological quality: summary across all studies. (a) PET studies; (b) MRI studies.

FIGURE 4.

Methodological quality: summary for each individual study. (a) PET studies; (b) MRI studies.

All nine MRI studies used an acceptable reference standard and avoided differential verification bias. 41,42,64,108–113 All described the index test in detail, probably due to the novel nature of the techniques, and usually described the reference standard well. Withdrawals were explained or did not occur in eight cases. 41,42,64,108,109,111–113 However, as with PET studies, the spectrum of patients was mixed, with over half not recruiting only early stage and newly diagnosed, or not recruiting prospectively or consecutively. Relevant clinical information was not available to the image interpreter in five cases41,42,110,111,112 and not reported in the other four,64,108,109,113 and uninterpretable results were dealt with well by only four studies. 108–110,112 None of the studies reported whether the reference standard results were interpreted blindly.

Assessment of diagnostic accuracy

Positron emission tomography studies: diagnostic accuracy results

Summary findings

Across 26 studies of PET or PET/CT (n = 2591 patients),44,73–97 the mean sensitivity was 63% [95% confidence interval (CI) 52% to 74%; range 20%–100%] and the mean specificity was 94% (95% CI 91% to 96%; range 75%–100%) (Table 9 and Figures 5 and 6). For the seven studies of PET/CT (n = 862),73–79 the mean sensitivity was 56% (95% CI 44% to 67%) and the mean specificity was 96% (95% CI 90% to 99%). For the 19 studies of PET only (n = 1729),44,80–97 the mean sensitivity was 66% (95% CI 50% to 79%) and the mean specificity was 93% (95% CI 89% to 96%). Therefore PET/CT gave a slightly lower mean sensitivity than PET only. The reason for this is not clear, as a concurrent CT scan is generally thought to enhance the accuracy of PET. As sensitivity varied widely between studies, this finding may have been due to chance. It may also have been due to differences in populations or study methods; for example, PET/CT studies used ALND/SLNB for all patients, whereas some PET-only studies used another reference standard for some patients, which may have led to overestimates of sensitivity (see Effect of reference standard). Also, three of seven PET/CT studies73,76,79 (compared with 5,83,84,87,88 of 19 PET-only studies) were restricted to clinically node-negative patients, which may have decreased the overall sensitivity estimate for the PET/CT studies (see Effect of clinical nodal status).

TABLE 9 - Summary of pooled sensitivities and specificities for PET studies

Diagnostic test	No. of studies	No. of patients	Sensitivity, % (95% CI)	Specificity, % (95% CI)
All PET studies
All PET studies44,73–97	26	2591	63 (52 to 74)	94 (91 to 96)
PET studies with or without CT
PET/CT73–79	7	862	56 (44 to 67)	96 (90 to 99)
PET (no CT)44,80–97	19	1729	66 (50 to 79)	93 (89 to 96)

FIGURE 5.

Forest plot of all PET studies. Brackets show 95% CIs. The figure shows the sensitivity and specificity for each study (squares) and 95% CIs (horizontal lines).

FIGURE 6.

Receiver operating characteristic plots for PET. (a) All PET studies, showing ROC curve (solid line), mean sensitivity/specificity (black spot) and 95% confidence region (dashed ellipse); (b) studies of PET only (rectangles) and PET/CT (diamonds). Rectangles/diamonds indicate sensitivity and specificity for individual studies (size proportional to sample sizes).

Effect of date of publication

Although the majority of studies do not show a clear trend in results according to date of publication, the earliest six studies (published between 1996 and 2001)92–97 do appear to report higher sensitivities than later studies (see Figure 5). This may reflect differences in methodology; for example, four93,94,96,97 of six of these early studies did not report using ALND or SLNB for all patients, which may have led to overestimates of sensitivity. Also, none of these early studies restricted inclusion by clinical nodal status, whereas a subset of the later studies included clinically node-negative patients only, which may have decreased the overall sensitivity estimate for the later studies.

Effect of size and number of axillary metastases

A few studies analysed the sensitivity according to the size and number of axillary metastases. Thresholds for size were not consistent across studies and the number of analysed patients was small. However, there was a trend for lower sensitivities where metastatic lymph nodes were smaller or fewer in number (Table 10; Figure 7). Axillary micrometastases (≤ 2 mm) were associated with a mean sensitivity of 11% (95% CI 5% to 22%) based on data from five studies (n = 63 patients),79,80,86–88 while macrometastases (> 2 mm) were associated with a mean sensitivity of 57% (95% CI 47% to 66%) based on data from four studies (n = 111 patients). 73,80,87,88 In addition, some studies reported the mean size and number of axillary metastases in TP patients (i.e. those detected by PET) and FN patients (i.e. those which PET failed to detect). As shown in Table 11, the cases which PET failed to detect tended to have smaller and fewer axillary metastases, although there was variation between studies. The smallest metastatic nodes detected by PET measured 3 mm,87,88 while PET failed to detect some nodes measuring > 15 mm86,94 (including one case of 25 mm in one study94).

TABLE 10 - Positron emission tomography sensitivity by size and number of axillary metastases and nodal stage

Specificity not calculable as these patients all had axillary metastases.

Subgroup	No. of studies	No. of patients	Sensitivity, % (95% CI)	Specificitya, % (95% CI)
Size of axillary metastases
≤ 2 mm79,80,86–88	5	63	11 (5 to 22)	Not calculable
< 5 mm78	1	15	33 (15 to 59)	Not calculable
> 2 mm73,80,87,88	4	111	57 (47 to 66)	Not calculable
> 5 mm79	1	51	57 (43 to 70)	Not calculable
> 10 mm78	1	28	79 (60 to 90)	Not calculable
Number of axillary metastases
1 metastatic node78,91,96	3	52	27 (7 to 63)	Not calculable
Multiple metastatic nodes91	1	12	50 (21 to 79)	Not calculable
2–5 metastatic nodes78,96	2	23	61 (40 to 78)	Not calculable
> 5 metastatic nodes78,96	2	24	77 (53 to 91)	Not calculable
Nodal stage
pN180,81	2	147	63 (24 to 91)	Not calculable
pN280,81	2	53	83 (51 to 96)	Not calculable
pN380,81	2	21	88 (66 to 97)	Not calculable

TABLE 11 - Size and number of metastatic nodes in TP and FN cases

ITCs, isolated tumour cells; micro, micrometastasis; NR, not reported.

Smith et al. 94 suggest that the two FN cases may have been due to patient positioning with arms at sides rather than above head.

Study	TP cases (PET detected)			FN cases (PET failed to detect)
	No. of patients in analysis	Mean	Range	No. of patients in analysis	Mean	Range
Size of largest metastatic node per patient (mm)
Heusner 200974	13	9.0	4–22	9	3.0	0.8–6
Kim 200975				8	2.6	1–7
Weir 200582				13		All ≤ 10
Fehr 200484	2	11.0	10–12	8	7.5	Micro–15
Lovrics 200486	9	22.0		16	13.0	NR (> 15 in 7 cases)
Wahl 200444	66	15.6		43	11.5
Barranger 200387	3	12.0	3–20	12	4.2	0.1–10
Guller 200288	6	13.5	3–30	8	2.9	ITCs–13
Smith 199894	19		8–NR	2	18.5a	12–25
Number of metastatic nodes
Heusner 200974	13	4.3	1–11	9	1.8	1–4
Kim 200975				8	1.4	1–3
Fuster 200877	14	5.2	1–19	6	1.8	1–3
Lovrics 200486	9	5.8		16	2.1	NR (≥ 3 in 5 cases)
Wahl 200444	66	5.0		43	2.7
Greco 200192				4	1.3	1–2
Avril 199696				5		1–4

FIGURE 7.

Forest plot of sensitivity of PET according to (a) size of axillary metastases; (b) number of axillary metastases; and (c) nodal stage. Brackets show 95% CIs. The figure shows the sensitivity and specificity for each study (squares) and 95% CIs (horizontal lines).

Effect of clinical nodal status

Studies in which all patients were clinically node negative, which generally referred to non-palpable axillary nodes, had a trend towards lower sensitivity and similar specificity when compared with studies including both clinically node-negative and node-positive patients (Table 12). This may reflect the fact that clinically negative axillary metastases are likely to be smaller and more difficult to detect via PET. This analysis was limited by the fact that, even in studies with a mixed population, the majority of patients were clinically node negative (see Table 7). One study reported separate results according to clinical nodal status: among clinically node-negative patients the sensitivity was 39/42 (93%) and the specificity 76/87 (87%), while among clinically node-positive patients the sensitivity was 29/30 (97%) and the specificity 6/8 (75%). 92 The mix of clinically node-positive and node-negative patients within the included studies was thought to be representative of clinical practice.

TABLE 12 - Positron emission tomography results according to clinical variables

IBC, inflammatory breast cancer.

Subgroup	No. of studies	No. of patients	Sensitivity, % (95% CI)	Specificity, % (95% CI)
Clinical nodal status
All clinically negative73,76,79,81,83,84,87,88	8	869	48 (32 to 64)	94 (90 to 97)
Mix of positive and negative44,74,77,78,80,85,86,91–94,96,97	13	1423	72 (54 to 85)	93 (88 to 96)
Nodal status not reported75,82,89,90,95	5	299	65 (44 to 81)	95 (82 to 99)
Patient sample
All patients early stage and newly diagnosed44,75,76,80,83–89,92,94,97	14	1413	63 (44 to 79)	94 (89 to 96)
Not all patients early stage and newly diagnosed73,74,77–79,81,82,90,91,93,95,96	12	1178	63 (49 to 76)	95 (91 to 98)
T stage (size) of breast tumour
T1 (≤ 2 cm)74,79,81,84,88,89,92,94,96,97	10	451	53 (34 to 72)	87 (82 to 91)
T2 (> 2 cm, ≤ 5 cm)74,77,79,81,84,88,89,92,94	9	343	67 (40 to 86)	86 (78 to 92)
T2 or above (> 2 cm)96,97	2	82	96 (84 to 99)	82 (13 to 99)
T3 (> 5 cm)74,77,79,84,89,94	6	41	65 (44 to 82)	88 (58 to 98)
T4 (tumour spread/fixed to skin/chest wall, or IBC)94	1	10	100 (54 to 100)	100 (40 to 100)
Reference standard
100% ALND44,73,77,84,87,89,92,95	8	773	59 (36 to 78)	90 (80 to 95)
ALND and/or SLNB74–76,78–83,86,88,91	12	1467	52 (40 to 63)	95 (93 to 97)
Not all ALND/SLNB, or not reported85,90,93,94,96,97	6	351	88 (68 to 96)	94 (85 to 98)
Prevalence of axillary metastases
< 40%44,73,75–78,80,86,95,97	10	1334	66 (49 to 80)	92 (85 to 96)
40%–49%74,79,82–85,87–89,91,92,94	12	874	51 (34 to 68)	94 (91 to 96)
≥ 50%81,90,93,96	4	383	84 (78 to 88)	98 (94 to 99)

Effect of patient sample

The mean sensitivity and specificity were not affected by whether all analysed patients were early stage (stage I, II or IIIA), newly diagnosed and non-DCIS; this may be explained by the fact that all included studies had a majority of patients who were early stage, newly diagnosed and non-DCIS (see Table 12).

Effect of T stage (size) of the breast tumour

Some studies reported sensitivity and specificity according to the T stage (size) of the primary breast tumour (see Table 12; Figure 8). The pattern was difficult to interpret due to the wide variation in sensitivity and specificity between studies and small patient numbers per subgroup. Data from some of the individual studies suggest a trend for lower sensitivity in patients with smaller breast tumours (e.g. between T1 and T274,81,92 and between T2 and T379,94), although the pattern for the meta-analysed data is less clear.

FIGURE 8.

Forest plot of sensitivity of PET according to T stage (size) of breast tumour. Brackets show 95% CIs. The figure shows the sensitivity and specificity for each study (squares) and 95% CIs (horizontal lines). IBC, inflammatory breast cancer.

Effect of reference standard

In terms of reference standard, studies in which all patients received ALND had similar mean sensitivity and specificity values to studies in which some patients received ALND and some SLNB. Studies in which the reference standard was not stated (or some patients did not receive ALND or SLNB) had a higher sensitivity, possibly due to the poorer-quality reference standard in these studies (see Table 12).

Effect of prevalence of axillary metastases

There was no clear correlation between prevalence of axillary metastases and sensitivity or specificity (see Table 12).

Sensitivity analysis for study quality

Eleven quality items from the QUADAS checklist69 were used to assess study quality (see Appendix 2). For PET, study quality variables which had a significant effect (p < 0.10) on either sensitivity or specificity were as follows. Studies with a representative patient spectrum (all early stage, newly diagnosed and non-DCIS, with prospective and consecutive recruitment) had a higher sensitivity than those in which the patient spectrum was unclear (p = 0.001). Studies that used the same reference standard regardless of index test results (differential verification avoided) had a lower sensitivity than studies in which differential verification occurred or those in which this was unclear (p < 0.001). Studies in which the index test was interpreted blind to reference standard results had a higher sensitivity than studies in which this was unclear (p < 0.001). Studies in which the reference standard was interpreted blind to index test results had a lower sensitivity than studies in which this was unclear, but a higher sensitivity than studies where this was unblinded (although there were only two studies76,90 known to be unblinded) (p < 0.001). Studies in which uninterpretable test results did not occur, or occurred and were included in the analysis, had a lower sensitivity than studies in which this was unclear (p < 0.001). No study quality variables had a significant effect on the specificity of PET. These results show a mixed pattern. Studies which score ‘unclear’ on reporting of quality variables may be expected to be of poorer quality, but this could theoretically lead to either underestimates or overestimates of diagnostic accuracy.

Magnetic resonance imaging studies: diagnostic accuracy results

Summary findings

Of the nine studies evaluating MRI (n = 307 patients),41,42,64,108–113 several reported more than one set of results on diagnostic accuracy, according to different criteria for defining whether axillary metastases were present. Results are first presented across all studies using the highest reported sensitivity and specificity for each study (Table 13 and Figures 9 and 10). Results are then presented according to each criterion for positivity (Table 14 and Figures 11 and 12).

TABLE 13 - Summary of pooled sensitivities and specificities for MRI studiesa

Where studies report results using more than one set of criteria for positivity, these analyses use data corresponding to the criteria with the highest reported estimates of diagnostic accuracy per study.

Diagnostic test	No. of studies	No. of patients	Sensitivity, % (95% CI)	Specificity, % (95% CI)
All MRI studies
All MRI studies41,42,64,108–113	9	307	90 (78 to 96)	90 (75 to 96)
MRI studies by type of MRI
USPIO-enhanced MRI108–112	5	93	98 (61 to 100)	96 (72 to 100)
Gadolinium-enhanced MRI41,42,113	3	187	88 (78 to 94)	73 (63 to 81)
MR spectroscopy64	1	27	65 (38 to 86)	100 (69 to 100)

TABLE 14 - Magnetic resonance imaging results according to criteria for positivity

Gd, gadolinium.

Criteria for positivity	No. of studies	No. of patients	Sensitivity, % (95% CI)	Specificity, % (95% CI)
USPIO-based criteria
USPIO uptake108–111	4	75	98 (63 to 100)	94 (69 to 99)
USPIO uptake, size > 10 mm, round shape112 (not clear if ‘and’ or ‘or’)	1	18	82 (48 to 98)	100 (59 to 100)
Gadolinium-based criteria
Gd uptake, size > 5 mm42 (not clear if ‘and’ or ‘or’)	1	75	90 (76 to 97)	83 (66 to 93)
Dynamic Gd signal intensity increase41,113	2	112	86 (68 to 94)	59 (45 to 72)
Dynamic Gd + positive washout41	1	65	71 (49 to 87)	90 (77 to 97)
Dynamic Gd + size > 4 sq-mm113	1	47	100 (69 to 100)	54 (37 to 71)
Dynamic Gd + size > 5 mm + abnormal morphology41	1	65	63 (41 to 81)	93 (80 to 98)
MR spectroscopy
MR spectroscopy64	1	27	65 (38 to 86)	100 (69 to 100)
Size and/or morphological criteria
Size > 4 sq-mm113	1	47	100 (69 to 100)	19 (8 to 35)
Size > 5 mm109	1	33	100 (85 to 100)	10 (0 to 45)
Size > 10 mm109	1	33	43 (23 to 66)	80 (44 to 97)
Abnormal morphology109	1	33	96 (78 to 100)	20 (3 to 56)
Size > 5 mm + abnormal morphology109	1	65	63 (41 to 81)	80 (65 to 91)
Size > 10 mm and/or round shape110	1	22	83 (36 to 100)	31 (11 to 59)

FIGURE 9.

Forest plot of all MRI studies. Brackets show 95% CIs. The figure shows the sensitivity and specificity for each study (squares) and 95% CIs (horizontal lines). Where studies report results using more than one set of criteria for positivity, these analyses use data corresponding to the criteria with the highest reported estimates of diagnostic accuracy per study. The criteria used for each study are shown on the plot.

FIGURE 10.

Receiver operating characteristic plots for MRI. (a) All MRI studies, showing ROC curve (solid line), mean sensitivity/specificity (black spot) and 95% confidence region (dashed ellipse); (b) all MRI studies, showing type of MRI. Shapes indicate sensitivity and specificity for individual studies (size proportional to sample sizes, legend shows type of MRI). Studies appear only once, using the highest reported sensitivity and specificity per study.

FIGURE 11.

Forest plots of MRI studies (using various criteria for defining a node as metastatic; many studies appear more than once). (a) Based on ultrasmall super-paramagnetic iron oxide uptake (and size/morphology); (b) based on gadolinium uptake (and size/morphology); (c) Based on size and/or morphology; (d) based on MR spectroscopy. Brackets show 95% CIs. The figure shows the sensitivity and specificity for each study (squares) and 95% CIs (horizontal lines).

FIGURE 12.

Receiver operating characteristic plot for MRI (using various criteria for defining a node as metastatic). Shapes indicate sensitivity and specificity for individual studies (size proportional to sample sizes, legend shows criteria used for defining a node as metastatic). Many studies appear more than once.

Across all nine MRI studies, using the highest sensitivity and specificity for each study, the mean sensitivity was 90% (95% CI 78% to 96%; range 65%–100%) and the mean specificity was 90% (95% CI 75% to 96%; range 54%–100%) (see Table 13 and Figures 9 and 10). According to the type of MRI, the mean estimates of sensitivity and specificity were as follows (see Table 13). Across five studies of USPIO-enhanced MRI (n = 93),108–112 the mean sensitivity was 98% (95% CI 61% to 100%)and specificity 96% (95% CI 72% to 100%). Across three studies of gadolinium-enhanced MRI (n = 187),41,42,113 the mean sensitivity was 88% (95% CI 78% to 94%) and specificity 73% (95% CI 63% to 81%). In the single study of in vivo proton MR spectroscopy (n = 27),64 the sensitivity was 65% (95% CI 38% to 86%) and the specificity 100% (95% CI 69% to 100%). Therefore, USPIO-enhanced MRI showed a trend towards higher sensitivity and specificity than gadolinium-enhanced MRI.

In addition, the diagnostic accuracy data were analysed according to the criteria for defining whether axillary metastases were present (e.g. based on USPIO or gadolinium uptake, size, morphology, or combinations of these criteria) (see Table 14 and Figures 11 and 12). Within this analysis, some studies appear more than once. The exact combinations of criteria were often not consistent across studies. The use of contrast uptake pattern as the main criterion for defining a node as metastatic appeared to give better combined sensitivity and specificity than size and morphology, although many studies used criteria based on both uptake and size/morphology, and the methods of interpreting uptake patterns varied within and between studies.

Effect of size and number of axillary metastases

No MRI studies presented data allowing calculation of sensitivity according to size and number of axillary metastases.

Effect of clinical nodal status

It was difficult to assess the effect of clinical nodal status as six42,109–113 of nine studies did not report these data (Table 15).

TABLE 15 - Magnetic resonance imaging results according to clinical variables

N/A, not applicable.

Subgroup	No. of studies	No. of patients	Sensitivity, % (95% CI)	Specificity, % (95% CI)
Reference standard
100% ALND41,42,64,109–113	8	297	91 (78 to 97)	88 (73 to 95)
ALND and/or SLNB108	1	10	100 (16 to 100)	100 (63 to 100)
Not all ALND/SLNB or not reported	0	0	N/A	N/A
Cancer stage
All patients early stage and newly diagnosed108,110,113	3	79	90 (63 to 98)	61 (46 to 74)
Not all patients early stage and newly diagnosed41,42,64,109,111,112	6	228	83 (74 to 89)	86 (78 to 92)
Clinical nodal status
All clinically negative108	1	10	100 (16 to 100)	100 (63 to 100)
Mix of positive and negative41,64	2	92	75 (69 to 79)	91 (86 to 94)
Nodal status not reported42,109–113	6	205	94 (83 to 98)	85 (63 to 95)
Prevalence of axillary metastases
< 40%41,108,110,113	4	144	86 (71 to 94)	71 (59 to 81)
40%–49%	0	0	N/A	N/A
≥ 50%42,64,109,111,112	5	163	94 (89 to 97)	84 (73 to 91)

Effect of patient sample

Studies in which all analysed patients were early stage (stage I, II or IIIA), newly diagnosed and non-DCIS had a trend towards a higher sensitivity, and a significantly lower specificity, than studies in which not all patients were early stage, newly diagnosed and non-DCIS; however, there was wide variation in results between studies (see Table 15).

Effect of T stage (size) of the breast tumour

There were insufficient consistently reported data to assess the relationship between individual size or stage categories for the primary tumour, and the sensitivity and specificity of MRI in detecting axillary metastases.

Effect of reference standard

It was not possible to assess the effect of reference standard as, in eight41,42,64,109–113 of nine studies, all patients received ALND (see Table 15).

Effect of prevalence of axillary metastases

There was no clear correlation between prevalence of axillary metastases and sensitivity or specificity (see Table 15).

Sensitivity analysis for study quality

Eleven quality items from the QUADAS checklist69 were used to assess study quality (see Appendix 2). For MRI, no study quality variables had a significant effect on sensitivity or specificity, although this analysis was limited by the small number of studies and the fact that there was little variation in scoring between studies for some quality variables.

Discussion for clinical effectiveness

Conclusions for clinical effectiveness

The studies in this review demonstrate a significantly higher sensitivity for MRI than for PET, with USPIO-enhanced MRI providing the highest sensitivity. However, since none of the included studies directly compared PET and MRI, caution should be taken when comparing these estimates. Sensitivity of PET is reduced for smaller metastases. Specificity was similar for PET and MRI. However, this analysis is limited by the small number and size of MRI studies, and the wide variations between and within studies in terms of the MRI method used and the criteria for defining a node as positive. The sensitivity and specificity of both PET and MRI vary widely between studies, which is likely to reflect differences in imaging methods and interpretation. Therefore, caution should be taken when interpreting these results, particularly for MRI.

Positron emission tomography and MRI have lower sensitivity and specificity than the current surgical diagnostic techniques of SLNB and 4-NS, but are associated with fewer adverse events. The potential use of PET or MRI, either as a replacement for SLNB or 4-NS or as an additional test prior to SLNB or 4-NS, was evaluated using a decision model (described further within Assessment of cost-effectiveness). As the sensitivity of PET is only moderate and is similar to that of ultrasound, it appears unlikely that PET could entirely replace SLNB or 4-NS in the assessment of axillary metastases.

As data on MRI are currently in the experimental stage, further large, well-conducted studies using up-to-date MRI methods are required to obtain more accurate data on the sensitivity and specificity of MRI in the assessment of axillary metastases. Further studies of USPIO-enhanced MRI would be valuable in order to gain more robust data on sensitivity and specificity, adverse effects and which are the best criteria for defining a node as metastatic.

Objective

The aim of the model was to evaluate the effects on patient outcomes and cost-effectiveness of enhanced imaging techniques (MRI and PET) compared with standard techniques in the assessment of axillary lymph node metastases in women with early-stage breast cancer. Two axillary sampling techniques, 4-NS and SLNB, are used currently in the UK. It is beyond the remit of this assessment to compare 4-NS and SLNB. The enhanced imaging techniques (PET and MRI) are therefore compared with the two baseline sampling techniques separately.

Diagnostic methods

The diagnostic methods that were evaluated in the model (MRI, PET, 4-NS and SLNB) were discussed in detail in Chapter 1, Current methods for assessment of axillary metastases and Chapter 1, Description of technology under assessment.

Structure of the model

A probabilistic discrete-event simulation model has been developed in simul8 (SIMUL8^®, Boston, MA, USA) to explore the costs and health outcomes associated with the assessment of axillary lymph node metastases and the treatment of women with early breast cancer.

Discrete-event simulation concerns the modelling of a system as it evolves over time by a representation in which the state variables change instantaneously at separate and countable points in time. 121 In the context of health-care modelling, patients are individually represented in a discrete-event simulation model, and normally have associated attributes indicating their distinctive demographical information and diagnostic and/or disease history.

The model differs from classical Markov state transition models, in which the state transition probabilities of patients are evaluated during fixed time intervals and patients may remain in one state after each evaluation. In discrete-event simulation, the time spent in each health state is sampled from a distribution when an individual patient enters this state. If one state can transit to multiple states, then the timing to each subsequent state is sampled and compared, and the patient will transit to the state that has the shortest transition delay.

The discrete-event simulation model consists of two main parts: the diagnostic pathway, which represents current and alternative diagnostic strategies (including PET and MRI), and the treatment pathway, which represents the disease progression among various health states and the management of patients with early breast cancer after the diagnosis of axillary lymph node metastases. A hypothetical cohort of 5000 early breast cancer patients was modelled. Each individual patient follows a specific diagnostic path and will obtain one of the four diagnostic results: TN, FP, TP or FN. Short- and long-term adverse events associated with sampling diagnostic techniques (4-NS and SLNB) and the associated cost and utility implications are also determined for each patient. The diagnostic results will influence the time spent in each of the subsequent health states, which is sampled from exponential distributions based on yearly transition probabilities. The starting age of patients was 56 years, which is based on the clinical effectiveness review within this assessment. The model was run for the remaining lifetime of patients.

Resource use and utilities are mainly taken from published literature. Input parameters are assigned probability distributions to reflect their imprecision and Monte Carlo techniques are performed to reflect this uncertainty in the results. Results are presented in terms of net health benefit and cost per incremental QALY gained.

Diagnostic pathway

The status of axillary lymph node metastases was assessed for all early breast cancer patients. Current methods of assessment consist of clinical examination followed by ultrasound. 17 If the result of ultrasound is positive, an ultrasound-guided fine-needle or core-needle biopsy will be conducted. If axillary metastases are identified via the biopsy (TP or FP), then the patients will be managed as node-positive patients (i.e. patients who have axillary lymph node metastases) and ALND will be performed, normally at the same time as the main breast cancer surgery is carried out.

For those women whose axillary metastases have not been identified by ultrasound or biopsy (i.e. either ultrasound or biopsy result is negative), current management involves the surgical removal of only some of the axillary lymph nodes (axillary sampling via either 4-NS or SLNB) for histological examination. Axillary sampling is normally performed at the same time as the main breast cancer surgery. Depending on the results of axillary sampling, patients will be managed either as node-negative patients (i.e. patients who do not have axillary lymph node metastases), who will receive no further investigation at that time, or as node-positive patients, who will undergo ALND to remove all axillary lymph nodes. Figure 13 illustrates the current standard diagnostic pathway.

FIGURE 13.

Standard and alternative imaging replacement diagnostic pathway in the School of Health and Related Research (ScHARR) model.

Apart from the two baseline 4-NS and SLNB strategies, six alternative strategies that involve either MRI or PET were evaluated. Two alternative strategies are to replace the axillary sampling methods with either MRI or PET (scenarios 1 and 2). Another four alternative strategies are to add MRI or PET before 4-NS (scenarios 3 and 4), and to add MRI or PET before SLNB (scenarios 5 and 6) in the diagnostic pathway. In theory, a biopsy (fine-needle or core-needle biopsy) could be undertaken in the event of a positive MRI or PET result in the alternative diagnostic pathways. However, MRI-guided or PET-guided biopsy is not currently available in most centres in the UK and no clinical studies were identified to provide data on this. Therefore, these techniques were not included in our assessment. The eight diagnostic strategies were as follows:

• baseline 1: 4-NS

• baseline 2: SLNB

• scenario 1: replace sampling with MRI

• scenario 2: replace sampling with PET

• scenario 3: add MRI before 4-NS

• scenario 4: add PET before 4-NS

• scenario 5: add MRI before SLNB

• scenario 6: add PET before SLNB.

The cost-effectiveness of the alternative MRI or PET replacement strategies was evaluated using the diagnostic pathway illustrated in Figure 13, in which MRI or PET replaces the current axillary sampling procedures. In order to evaluate the other four alternative strategies, the standard pathway was modified to create an alternative diagnostic pathway (Figure 14). In the alternative pathway, enhanced imaging techniques will be carried out for patients who have negative ultrasound or biopsy results. If the results of the imaging techniques are positive, no axillary sampling is performed and the patients are regarded as node positive and go on to receive ALND. If the results of imaging techniques are negative, further axillary sampling (4-NS or SLNB) is still performed as in the standard pathway.

FIGURE 14.

Alternative imaging addition diagnostic pathway in the School of Health and Related Research (ScHARR) model.

Health states

There are 10 health states within the disease pathway part of the model:

• adjuvant therapy – node (true) negative patients

• adjuvant therapy – node (false) positive patients

• adjuvant therapy – node (true) positive patients

• adjuvant therapy – node (false) negative patients

• post-adjuvant therapy

• locoregional relapse

• remission (post-locoregional relapse)

• metastatic relapse

• death from breast cancer

• death from other causes.

The disease pathway is shown in Figure 15. Each individual patient starts from one of the four adjuvant therapy states, depending on the previous diagnostic result.

FIGURE 15.

Treatment pathways in the School of Health and Related Research (ScHARR) model.

Model assumptions

The model employs a number of simplifying assumptions, which are detailed below.

• The sensitivity and specificity of biopsy, axillary sampling (4-NS and SLNB), and imaging techniques (MRI and PET) are independent of preceding diagnostic results.

• The sensitivity and specificity of MRI is based on all identified studies, with no distinction made between different types of MRI (see Chapter 3, Quantity and quality of research available). This assumption is tested in the sensitivity analysis.

• Seroma, surgical drain and infection are the short-term adverse events associated with diagnostic techniques (4-NS, SLNB and ALND) considered by the model.

• Lymphoedema is the only long-term adverse event considered. Lymphoedema is classified as either mild/moderate or severe.

• Studies have reported adverse events associated with SLNB, whereas no studies were identified which quantify the short-term adverse events associated with 4-NS or compare the probability of adverse events between SLNB and 4-NS. Therefore, the probability of adverse events is assumed to be equal for 4-NS and SLNB.

• Short-term adverse events increase the costs, but do not affect quality of life.

• Long-term adverse events (i.e. lymphoedema) affect both costs and quality of life for the rest of the patient’s life.

• During the adjuvant therapy period, node-positive patients receive chemotherapy plus hormonal therapy (where appropriate) and node-negative patients receive hormonal therapy (where appropriate).

• Patients receive adjuvant therapy for a fixed 5-year period. Node-positive patients receive chemotherapy for half a year, followed by hormonal therapy for 4.5 years. Node-negative patients receive hormonal therapy for 5 years. This is the maximum time patients may stay in this state. Patients may, owing to model dynamics, spend < 5 years in the adjuvant therapy state if the sampled time to locoregional relapse or death from other causes is < 5 years.

• Following locoregional relapse patients cannot experience further locoregional relapse; they can only experience metastatic relapse.

• Death rates for non-breast cancer causes are based on UK mortality statistics and applied across all health states. These are not adjusted to exclude breast cancer mortality, and so may overestimate the risk of dying due to non-breast cancer causes.

Model inputs: accuracy and costs of diagnostic techniques

The model inputs of accuracy and costs of diagnostic methods were summarised in Table 18. The sensitivity and specificity of clinical examination and biopsy were based on previous published studies. 41,42,47,94 The sensitivity and specificity of ultrasound for clinically negative patients were calculated based on either the size criterion (60.9% and 77.3%, respectively) or morphology criterion (43.9% and 92.4%, respectively) from a systematic review by Alvarez et al. 47 We used the averages to represent overall sensitivity and specificity assuming both size and morphology are used to assess axillary lymph node metastases. The sensitivity and specificity of ultrasound for clinically-positive patients were estimated based on expert opinion, as no data were identified from published studies.

The sensitivity of 4-NS was based on two identified studies with data from 335 patients. 122,123 The sensitivity of SLNB was based on a systematic review and meta-analysis of 69 studies undertaken by Kim et al. 125 with data from over 8000 patients. All studies were included in the NICE guideline. 17 The mean sensitivity of 4-NS is slightly higher than that of SLNB (94.5% vs 93%) according to literature reviewed within the NICE guideline. However, it is important to note that the sensitivity of SLNB is based on a significantly larger sample size and should be more robust. The specificities of 4-NS and SLNB are set at 100% as, by definition, there should be no FP cases when histological methods are used.

The costs of clinical examination, ultrasound, biopsy, MRI and ALND were obtained from NHS reference costs. 55,56 Apart from scanning the axilla, MRI is currently also used in the pre-surgical evaluation of breast tumours for some patients. However, the procedure is performed in a small proportion of women with breast cancer (e.g. lobular cancers and patients having neoadjuvant chemotherapy) and it is likely that axillary and breast MRI scans would be undertaken separately, as pre-contrast scans would be required for both procedures. Therefore marginal cost savings of axillary MRI scans due to breast MRI scans are not considered and we assumed the full NHS reference costs for MRI in the model.

The costs of PET were calculated based on a published UK study. 66 The costs of breast surgery were calculated based on NHS costs of total mastectomy surgery [Healthcare Resource Group (HRG) codes are JA07A, JA07B and JA07C] and intermediate breast surgery (HRG codes are JA09A and JA09B),56 assuming that two-thirds of patients receive intermediate breast surgery and one-third of patients receive mastectomy.

Sentinel lymph node biopsy and 4-NS are assumed to be carried out at the same time as the breast surgery. ALND is assumed to be carried out at the same time as the breast surgery, if no sampling is performed beforehand (e.g. patients are diagnosed as node positive by biopsy). No UK costs of SLNB, 4-NS and ALND (at the same time as breast surgery) were identified from published studies. One study from the USA was identified which reported the costs of breast surgery alone, SLNB together with breast surgery and ALND together with breast surgery. 20 The relative ratios between breast surgery and SLNB together with breast surgery, and between breast surgery and ALND together with breast surgery, can be calculated. These ratios were used to adjust the UK costs of breast surgery to obtain the costs of SLNB and ALND together with breast surgery. The procedure of 4-NS together with breast surgery was assumed to increase the costs of breast surgery alone by 10%.

Model inputs: probability and costs of short-term adverse events

The model inputs of probability and costs of short-term adverse events were summarised in Table 19. The probabilities of developing short-term adverse events due to SLNB and ALND were estimated based on published studies. No studies were identified that quantify the short-term adverse events associated with 4-NS. Based on expert opinion, it is assumed that 4-NS is associated with the same probabilities of developing short-term adverse events as SLNB. Sensitivity analysis was carried out to test the assumption that patients are more likely to develop adverse events for 4-NS than SLNB. The costs of short-term adverse events were based on a study in the US, as no UK costs were identified.

Model inputs: costs and utilities of health states

The model inputs of costs and utilities of health states are summarised in Table 20. Patients with node-negative results (TN and FN) are assumed to receive hormonal therapy for 5 years (where appropriate). For patients with FP results, the TN nodal status will be picked up by the following ALND. Therefore, patients with FP diagnoses also receive hormonal therapy. It is assumed that 81% of patients are oestrogen receptor positive, which means that they will respond to hormonal therapy. 126 Among these patients, it is assumed that 90% receive aromatase inhibitors [anastrozole (Armidex^®, AstraZeneca), exemestane (Aromasin^®, Pfizer) or letrozole (Femara^®, Novartis)] and 10% receive tamoxifen (Nolvadex, Istubal, Valodex). It is also assumed that each patient has one clinic visit and one mammogram per year in the adjuvant therapy state. The annual cost of aromatase inhibitors, tamoxifen and mammography is estimated to be £1002 based on a published study. 129

Patients with TP results are assumed to receive chemotherapy for 6 months (£8788) followed by hormonal therapy for 4.5 (£1002 per year). It is assumed docetaxel (Taxotere^®, Sanofi-aventis)-based chemotherapy is used. 17 The costs of chemotherapy were calculated based on a published study. 129

Annual costs for the post-adjuvant therapy state are assumed to be £0. Annual costs for the locoregional and metastatic relapse states and the death state were calculated based on a published study and are £4737, £10,196 and £3321, respectively. 129 For patients in the remission state after locoregional relapse, the annual costs for the first 5 years were assumed to be the same as for patients in the adjuvant therapy state receiving hormonal therapy (£1002). The costs are assumed to be £0 after 5 years in the remission state.

The source of utility data used in the model is from Tengs and Wallace,130 which is a systematic review of health-related quality of life estimates from publicly available source documents. Given that the health-related quality of life in the general population decreases with age, it is important to take this into account in the model. General population utility estimates from Ara and Brazier131 were applied using a regression analysis of utility versus age. The age-related utility is calculated by the following formula (Equation 1):

where A = –0.0001728, B = –0.000034 and C = 0.9584588.

The utilities for all health states are multiplied by this age-related utility value for each year of the model.

Model inputs: transition probabilities between health states

Transition probabilities between health states are summarised in Table 21. When a patient starts adjuvant therapy, the expected life expectancy of the patient is determined by the life table. 132 The patient will die from other causes once the expected life expectancy is met. The transition probabilities from the adjuvant therapy state to the locoregional recurrence state depend on the diagnostic results for lymph node metastases (i.e. TN, FP, TP and FN). Node-negative patients, including TN and FP, will have lower transition probabilities and node-positive patients, including TP and FN, will have higher transition probabilities. In particular, patients with FN results will have the highest transition probability because they have been denied ALND and chemotherapy, needed to reduce the risk of recurrence. The annual transition probabilities for locoregional recurrence were based on a study by Orr et al. ,133 which provided the estimates of annual transition probabilities of recurrence in patients with negative results (0.03), TP results (0.09) and FN results (0.14). The study assumed that patients with TP results receive chemotherapy and patients with FN results do not receive chemotherapy (and only receive hormonal therapy). The same assumption is used in this assessment.

When a patient enters the adjuvant therapy state, the model uses exponential distributions to sample the time to locoregional relapse according to the above annual transition probabilities. The sampled time to locoregional relapse will then be compared with the time to death (according to the life table) and the 5-year maximum period for adjuvant therapy. Depending on which event happens first (locoregional relapse, death due to other causes or finishing the adjuvant therapy), the patient will transit to the corresponding state after the time delay. The methodology used to determine which state the patient transits to is the same for other health states.

When patients enter the post-adjuvant therapy state, they may experience locoregional or metastatic relapse, or they may die from other causes. The annual transition probabilities of locoregional relapse and the time to death from other causes are assumed to be the same as under the adjuvant therapy state. The transition probabilities to metastatic relapse are 0.0023 for node-negative patients (TN and FP), 0.0052 for TP patients, and 0.0094 for FN patients. 120

If patients enter the locoregional relapse state, they may enter a subsequent remission state, may have metastatic relapse, or may die from other causes. It is assumed that patients can stay in the locoregional relapse state for a maximum of 1 year. The annual probability of developing metastatic cancer in the first year of locoregional relapse is 0.18,134 which is much higher compared with disease-free survival states (i.e. adjuvant therapy and post-therapy states). Orr et al. 133 suggested that the annual probability of death for patients with locoregional relapse is 0.30. This includes death due to both breast cancer and other causes. The model does not distinguish between death from breast cancer and death from other causes once a patient enters the locoregional relapse state. The maximum lifetime of a patient is still bounded by the life expectancy of the patient (i.e. death due to other causes).

If patients enter the remission state, they may still experience metastatic relapse before death. The average annual probability of developing metastatic cancer from the remission state is 0.13134 and the probability of death (for all reasons) is assumed to be the same as in the locoregional relapse state, which is 0.30.

Metastatic cancer is assumed not to be curable and the annual probability of death from metastatic relapse is 0.37. 129

Model inputs: probability, costs and utilities of long-term adverse events

The model inputs of probability and costs of long-term adverse events (i.e. lymphoedema) are summarised in Table 22. The probabilities of developing lymphoedema due to SLNB and ALND were estimated based on published studies. 48–50,53 The probability of having lymphoedema due to 4-NS was assumed to be the same as SLNB, as no data were identified for 4-NS.

Lymphoedema was classified as either mild/moderate or severe. A literature search was undertaken, but no studies reporting utility for patients with lymphoedema were identified. The proportion of patients within each category and the utility decrements of each category were therefore estimated from a published study reporting quality of life using the FACT-B + 4 (Functional Assessment of Cancer Therapy for Breast Cancer, adding a four-item arm subscale) quality-of-life instrument. 135 The study reported the data regarding the quality of life of breast cancer patients who suffer from different degrees of lymphoedema, using the FACT-B + 4 quality-of-life instrument. The utility decrements were estimated based on these quality-of-life data, therefore the decrements do not represent the true utility decrements due to lymphoedema. Sensitivity analyses were carried out to explore the impact on the cost-effectiveness results caused by changing the estimated utility decrements. The annual additional costs due to lymphoedema were based on expert opinions from the Sheffield Lymphoedema Service.

The utility decrement represents the reduced quality of life for patients with lymphoedema.

Discounting

The economic analysis assumes that both costs and QALYs are discounted at 3.5% per annum, in line with current recommendations from Her Majesty’s Treasury. 136

Univariate sensitivity analysis

In order to explore the impact on the cost-effectiveness results of changes to individual parameters and assumptions, a number of sensitivity analyses were performed.

Probabilistic sensitivity analysis

Probabilistic sensitivity analysis (PSA) was undertaken to demonstrate the impact of uncertainty in the key model parameters and to generate information on the likelihood that each of the diagnostic strategies is optimal.

The sensitivity and specificity of a diagnostic test are generally correlated. To maintain this correlation, the sensitivities of each diagnostic method are sampled from a beta distribution, while the specificities of the test are derived based on the sampled sensitivity. The sensitivity and specificity are linked by the prevalence and the overall accuracy of the test, which are assumed to be constants. Therefore, when the sensitivity is sampled, the specificity can be calculated deterministically.

Equation 2 is the formula for calculating the test accuracy. After rearranging Equation 2, the model applies Equation 3 to derive test specificity from sensitivity. The overall accuracy of each diagnostic method is presented in Table 23, where the prevalence of early breast cancer among breast cancer patients is assumed to be fixed at 41.2%. The beta distributions representing the sensitivity of each diagnostic method are based on trial data from previous literature and the systematic review within this assessment (MRI and PET).

Health utilities and the probabilities of developing short- and long-term adverse events were modelled using beta distributions. Costs were sampled from normal distributions where standard errors can be calculated. Transition probabilities among health states and other costs were sampled from uniform distributions bounded by a 10% increase or decrease in the mean value.

The PSA was carried out by allowing the key model input parameters to vary according to the uncertainty specified in their probability distributions, with 500 sets of random numbers used to generate 500 parameter configurations, which produce 500 sets of model outputs. All model results were based on the PSA model outputs.

To demonstrate that 500 replications is enough to obtain accurate model outputs, the cumulative mean total costs and total QALYs of the baseline 4-NS diagnostic strategy based on 2000 replications were calculated. A significance level of 5% was used to construct the CI around the cumulative means. The analysis suggests that the CI was sufficiently narrow and the cumulative mean stabilises after 500 replications are performed.

Base-case estimates of cost-effectiveness

The base-case estimates given below are the mean estimates from the 500 runs of the PSA. For each strategy, results are presented in terms of the diagnostic results, the total number of diagnostic and surgical procedures performed, the total costs and QALYs, the net benefit and incremental cost-effectiveness.

Diagnostic results

The proportions of patients whose lymph node diagnostic results are TN, FP, TP and FN are presented in Table 24 and Figure 16.

TABLE 24 - Diagnostic results of each strategy

Diagnostic strategy	TN (%)	FP (%)	TP (%)	FN (%)	Total (%)
Baseline 1: 4-NS	58.6	0.2	40.1	1.1	100.0
Baseline 2: SLNB	58.6	0.2	39.9	1.3	100.0
Scenario 1: replace sampling with MRI	52.5	6.3	39.3	1.9	100.0
Scenario 2: replace sampling with PET	55.2	3.6	34.0	7.2	100.0
Scenario 3: add MRI before 4-NS	52.5	6.3	41.1	0.1	100.0
Scenario 4: add PET before 4-NS	55.2	3.6	40.8	0.4	100.0
Scenario 5: add MRI before SLNB	52.5	6.3	41.1	0.1	100.0
Scenario 6: add PET before SLNB	55.2	3.6	40.7	0.5	100.0

FIGURE 16.

Proportion of people with FP and FN diagnostic results under each strategy.

Adverse events

The numbers of short- and long-term adverse events associated with each diagnostic strategy are presented in Table 26 and Figure 17. The numbers of short- and long-term adverse events are proportional to the number of sampling and ALND procedures undertaken (see Table 25). The model assumes that the probabilities of short-term adverse events are the same for both 4-NS and SLNB, while ALND is associated with much higher probabilities of developing adverse events.

TABLE 26 - Adverse event cases associated with each diagnostic strategy

Diagnostic strategy	Short-term adverse events	Long-term adverse events (lymphoedema)
Baseline 1: 4-NS	2781	633
Baseline 2: SLNB	2764	631
Scenario 1: replace sampling with MRI	2642	487
Scenario 2: replace sampling with PET	2172	401
Scenario 3: add MRI before 4-NS	3055	684
Scenario 4: add PET before 4-NS	2925	662
Scenario 5: add MRI before SLNB	3055	684
Scenario 6: add PET before SLNB	2919	661

FIGURE 17.

Adverse event cases associated with each diagnostic strategy.

Survival results

The 5-year survival rates for patients with different diagnostic results are presented in Figure 18. Patients with axillary lymph node metastases (both TP and FN) have lower survival rates than patients with negative nodal status. In particular, patients with FN results are associated with the lowest survival rate, because they do not receive ALND or chemotherapy that may reduce the chance of recurrence.

FIGURE 18.

The 5-year survival rates for patients with different diagnostic results.

The 5-year survival rates of all patients with early breast cancer and the absolute number of deaths per 10,000 patients presenting with early breast cancer are presented in Table 27 for each tested diagnostic strategy. The overall survival rate of each strategy is dependent on the proportion of patients with each diagnostic result (see Table 24) and the individual survival rate for patients with different diagnostic results (see Figure 18). As patients with FN results have the lowest survival rate, diagnostic strategies that are associated with more FN results have lower overall survival rate (e.g. PET replacement strategy). The absolute differences in overall survival rates among tested diagnostic strategies are relatively small (range from 87.52% to 88.1%). This is because the absolute numbers of patients with FN results only account for a small proportion of patients (range from 0.1% to 7.2% as shown in Table 24). The absolute difference in the number of deaths during the first 5 years per 10,000 early-stage breast cancer patients is also presented in Table 27. This shows the change in absolute mortality for each tested diagnostic strategy, using the 4-NS baseline as the reference strategy.

TABLE 27 - The 5-year survival rates and the absolute number of deaths in England for patients with early breast cancer (comparing different diagnostic strategies)

Diagnostic strategy	5-year survival rate (all patients) (%)	Number of deaths during first 5 years (per 10,000 patients)	Absolute difference in number of deaths during first 5 years (per 10,000 patients)
Baseline 1: 4-NS	88.03	1197	0
Baseline 2: SLNB	88.00	1200	3
Scenario 1: replace MRI	87.96	1204	7
Scenario 2: replace PET	87.52	1248	51
Scenario 3: MRI before 4-NS	88.10	1190	–7
Scenario 4: PET before 4-NS	88.08	1192	–5
Scenario 5: MRI before SLNB	88.10	1190	–7
Scenario 6: PET before SLNB	88.07	1193	–4

Cost-effectiveness results (net benefit analysis)

Positron emission tomography and MRI are assumed to be associated with no short- and long-term adverse events. However, due to the lower accuracy of the imaging techniques, more FP and FN cases will be produced, which will lead to increased costs, worse quality of life due to adverse events, and in some cases higher probability of recurrence and subsequent death from breast cancer. Economic modelling provides a systematic way to understand and quantify the complex trade-offs between the advantages and disadvantages of the imaging techniques, so that the overall cost-effectiveness of alternative strategies can be determined.

Net benefit is the increase in effectiveness (Δ E), multiplied by the amount the decision-maker is willing to pay per unit of increased effectiveness (R_T), less the increase in cost (Δ C). The formula for calculating net benefit is:

^{[Equation 4]} Net benefit = R T Δ E − Δ C > 0

A strategy is most cost-effective if it has the highest positive net benefit. Thresholds of willingness to pay per QALY of £10,000, £20,000 and £30,000 were used to calculate the net benefit of each strategy.

Two baseline strategies (4-NS and SLNB) are considered. Both 4-NS and SLNB are currently used in the UK. It is not within the scope of this assessment to compare these sampling methods. Tables 28 and 29 summarise the net benefits of each strategy using either 4-NS or SLNB as the baseline strategy. Note that scenarios 1 and 2 appeared in both tables because they are comparable to both 4-NS and SLNB strategies. The total costs and QALYs of each diagnostic strategy are plotted in Figures 19 and 20.

TABLE 28 - Total costs and QALYs and the net benefit of each diagnostic strategy using 4-NS as baseline

	Costs associated with diagnostic tests or short-term adverse events (£)	Costs associated with health states or long-term adverse events (£)	Total costs (£)	Total QALYs	Net benefit (£10,000 per QALY)	Net benefit (£20,000 per QALY)	Net benefit (£30,000 per QALY)
Baseline 1: 4-NS	3157	16,571	19,728	8.122		Baseline
Scenario 1: replace sampling with MRI	3030	16,295	19,325	8.174	919	1435	1952
Scenario 2: replace sampling with PET	3484	15,835	19,319	8.126	450	490	531
Scenario 3: add MRI before 4-NS	3204	16,655	19,859	8.125	–104	–78	–51
Scenario 4: add PET before 4-NS	3818	16,623	20,440	8.126	–681	–649	–618

TABLE 29 - Total costs and QALYs and the net benefit of each diagnostic strategy using SLNB as baseline

	Costs associated with diagnostic tests or short-term adverse events (£)	Costs associated with health states or long-term adverse events (£)	Total costs (£)	Total QALYs	Net benefit (£10,000 per QALY)	Net benefit (£20,000 per QALY)	Net benefit (£30,000 per QALY)
Baseline 2: SLNB	3642	16,547	20,189	8.119		Baseline
Scenario 1: replace sampling with MRI	3030	16,295	19,325	8.174	1412	1959	2507
Scenario 2: replace sampling with PET	3484	15,835	19,319	8.126	942	1014	1085
Scenario 5: add MRI before SLNB	3546	16,655	20,201	8.124	35	82	129
Scenario 6: add PET before SLNB	4208	16,614	20,822	8.125	–577	–520	–464

FIGURE 19.

Total costs and QALYs of diagnostic strategies using 4-node sampling as baseline.

FIGURE 20.

Total costs and QALYs of diagnostic strategies using sentinel lymph node biopsy as baseline.

The horizontal axis of Figures 19 and 20 represents the total QALYs accrued by each diagnostic strategy and the vertical axis represents the total costs associated with each strategy. The lines in the figures denote the cost-effective frontiers if PET and MRI replacement strategies are excluded based on clinical grounds.

The total and breakdown costs (costs associated with diagnostic tests or short-term adverse events and costs associated with health states or long-term adverse events) of each strategy are illustrated in Figure 21.

FIGURE 21.

Costs associated with each diagnostic strategy.

Cost-effectiveness results (incremental analysis)

The incremental cost-effectiveness analyses were performed assuming that MRI and PET replacement strategies are deemed unacceptable on clinical grounds. Tables 30 and 31 show the incremental cost-effectiveness analyses where 4-NS and SLNB are used as the baseline strategy.

TABLE 30 - Incremental cost-effectiveness analysis of diagnostic strategies using 4-NS as baseline (excluding MRI and PET replacement strategies)

ICER, incremental cost-effectiveness ratio.

	Total cost (£)	Total QALYs	Incremental cost (£)	Incremental QALY	ICER (£)
Baseline 1: 4-NS	19,728	8.122		Baseline
Scenario 3: add MRI before 4-NS	19,859	8.125	131	0.003	48,986
Scenario 4: add PET before 4-NS	20,440	8.126	581	0.000	1,200,212

TABLE 31 - Incremental cost-effectiveness analysis of diagnostic strategies using SLNB as baseline (excluding MRI and PET replacement strategies)

ICER, incremental cost-effectiveness ratio.

	Total cost (£)	Total QALYs	Incremental cost (£)	Incremental QALY	ICER (£)
Baseline 2: SLNB	20,189	8.119		Baseline
Scenario 5: add MRI before SLNB	20,201	8.124	11	0.005	2451
Scenario 6: add PET before SLNB	20,822	8.125	622	0.001	647,415

The cost-effectiveness plane for the MRI before 4-NS strategy versus the baseline 4-NS strategy is presented in Figure 22. The cost-effectiveness plane for the MRI before SLNB strategy versus the baseline SLNB strategy is presented in Figure 23.

FIGURE 22.

Cost-effectiveness plane for MRI before 4-node sampling (4-NS) strategy versus baseline 4-NS strategy.

FIGURE 23.

Cost-effectiveness plane for MRI before SLNB strategy versus baseline SLNB strategy.

Both Figures 22 and 23 demonstrate that MRI before sampling strategy may lead to either an increase or decrease in QALYs compared with the baseline strategy. The advantage of this strategy in terms of QALYs is therefore uncertain. Figure 22 shows that the MRI before 4-NS typically generates higher total costs than the 4-NS strategy. Figure 23 shows that MRI before SLNB strategy may lead to either an increase or a decrease in total costs compared with the SLNB strategy. The benefits offered by these strategies are not clear-cut.

Univariate sensitivity analysis

Diagnostic results

The baseline (4-NS and SLNB) strategies produce the smallest number of FP cases among all strategies. The number of FN cases produced by the two baseline strategies is also very small (1.1% for 4-NS and 1.3% for SLNB). In the MRI replacement strategy, the number of FP cases is increased significantly from 0.2% to 6.3% and the number of FN cases is increased to a lesser extent, from around 1.0% to 1.9%. In the PET replacement strategy the numbers of both FP and FN cases are increased significantly, from 0.2% to 3.6% for FP cases and from around 1.0% to 7.2% for FN cases. Overall, the PET replacement strategy produces the largest number of FP/FN cases among all the diagnostic strategies tested.

If MRI or PET is placed before sampling methods, the number of FN cases is reduced from around 1.0% to 0.1% if MRI is placed before sampling and to around 0.5% if PET is placed before sampling. The number of FP cases remains the same as for the MRI and PET replacement strategies.

The baseline strategies produce the lowest number of FP cases. The strategies of adding MRI before 4-NS and SLNB produce the lowest number of FN cases. Overall, the baseline strategies produce the smallest number of combined FP and FN cases. The MRI and PET replacement strategies may be considered unacceptable on clinical grounds, as they both generate more FP and FN cases than current standard practice. The strategies of adding MRI or PET before sampling also produces more FP cases, although the number of FN cases is reduced.

Number of procedures

Under the baseline sampling strategies a proportion of patients will need two separate surgical procedures: a sampling procedure (4-NS or SLNB) and subsequent ALND. In the replacement strategies, no sampling procedures are needed and the ALND procedures are carried out for TP and FP cases. Compared with the baseline strategies, the total number of ALND procedures carried out is increased for the MRI replacement strategy (due to more FP cases) and decreased for the PET replacement strategy (due to less TP cases).

If MRI or PET is placed before the sampling methods, the numbers of sampling procedures carried out are reduced compared with the baseline 4-NS and SLNB strategies because both TP and FP cases detected by MRI or PET will receive ALND surgery without sampling procedures. The reduction is more evident when MRI rather than PET is placed before sampling, as MRI is associated with more positive cases (both true and false). However, due to the increase in FP cases, the strategies of adding MRI or PET before sampling are associated with more ALND procedures than the baseline 4-NS and SLNB strategies.

Adverse events

The number of short- and long-term adverse events is proportional to the number of 4-NS, SLNB and ALND surgical procedures carried out. Adverse events are more frequent for ALND than 4-NS and SLNB. Among all diagnostic strategies modelled, the PET replacement strategy is associated with the lowest number of adverse events, followed by the MRI replacement strategy. The PET replacement strategy is associated with the smallest number of ALND procedures. The strategies of adding MRI or PET before sampling produce more adverse events than the baseline 4-NS and SLNB strategies, because more ALND procedures are carried out.

Survival rates

Patients with axillary lymph node metastases have lower survival rates (both TP and FN) than patients with negative nodal status (both TN and FP). Patients with FN results have the lowest survival rates because they do not receive ALND or chemotherapy that may reduce the risk of recurrence.

Compared with the two baseline strategies, the overall survival rates of early-stage breast cancer patients are lower for the MRI or PET replacement strategies and higher for the strategies where MRI or PET is placed before 4-NS and SLNB. The absolute differences in overall survival rate among tested diagnostic strategies are relatively small, as the absolute number of patients with FN results only account for a small proportion of all patients.

Cost-effectiveness analyses – baseline results

The PET and MRI strategies are compared with two baseline techniques – SLNB and 4-NS – which are both used currently in the UK. It is beyond the remit of this assessment to compare 4-NS and SLNB.

The MRI replacement strategy is the most cost-effective strategy and dominates the two baseline strategies. The higher QALYs of the MRI replacement strategy are driven by fewer cases of lymphoedema, which has a lifelong impact on quality of life. The PET replacement strategy is the next most cost-effective strategy and also dominates the two baseline strategies. Compared with the MRI replacement strategy, the PET replacement strategy has significantly lower QALYs, which is driven by more FN cases who are more likely to experience locoregional and metastatic relapse.

The cost-effectiveness results demonstrate that, on the population level, it is beneficial to replace invasive sampling methods with the non-invasive imaging techniques of MRI or PET. If the MRI or PET replacement strategies are used, a small proportion of patients will be wrongly diagnosed as FP or FN, which will impact on life-years gained and quality of life for these patients. However, the majority of patients will be correctly diagnosed by MRI or PET, without the need for sampling procedures. This will improve their quality of life owing to avoidance of short- and long-term adverse events such as lymphoedema. The model results suggest that the health benefits gained by the majority of patients outweigh the negative impact on reduced survival and lower quality of life of a small proportion of patients. The imaging replacement strategies, especially for MRI, also cost less than the baseline 4-NS and SLNB strategies. Overall, the analysis predicts that it is cost-effective to replace 4-NS or SLNB with MRI or PET.

Despite the cost-effectiveness results, the MRI and PET replacement strategies may be considered unacceptable on clinical grounds, due to higher numbers of FP and FN cases. If this is the case, then the most cost-effective strategy is the 4-NS strategy (if 4-NS is used as the baseline) or the addition of MRI before SLNB (if SLNB is used as the baseline). However, these results are less robust than the results for the replacement strategies. The differences in costs and QALYs between strategies are small and therefore the results are sensitive to changes in the input parameters. More robust evidence is needed on the costs of 4-NS and SLNB and the costs of MRI and PET. In addition more robust evidence on the sensitivity and specificity of 4-NS is also needed.

The strategies of placing MRI or PET before sampling may also be rejected on the clinical grounds that they are associated with more FP cases. In order to have a similar level of FP cases to the sampling methods, the specificity of MRI and PET needs to be improved to be close to 100% which, by definition, is the specificity of 4-NS and SLNB. For the MRI or PET replacement strategies, in order to have similar levels of FP and FN cases to the sampling methods, both the sensitivity and specificity have to be improved to the levels of 4-NS and SLNB. The most promising technique is the USPIO-enhanced MRI, which has a mean sensitivity of 98% and specificity of 96%. These figures are, however, based on a limited number of small studies. Further studies are needed to confirm the robustness of these figures.

In general, based on the estimates of sensitivity and specificity of MRI and PET in this assessment, the analysis predicts that it is cost-effective to replace 4-NS or SLNB with MRI or PET which will accrue more QALYs and cost less at the population level. Within the two replacement strategies, it is more cost-effective to replace sampling with MRI than PET.

Cost-effectiveness analyses – sensitivity analysis results

The sensitivity and specificity of MRI have a significant impact on the cost-effectiveness results, and change the results in terms of the most cost-effective strategy. Further evidence on the sensitivity and specificity of MRI is therefore needed.

The utility decrement for lymphoedema has a significant impact on the cost-effectiveness of the PET replacement strategy, as the strategy is associated with the smallest number of lymphoedema cases. If the assumed utility decrement for lymphoedema is reduced, the PET replacement strategy no longer dominates the baseline 4-NS and SLNB strategies. Further studies on the costs and quality of life of lymphoedema are needed.

The probabilities of relapse for FN patients also have a significant impact on the cost-effectiveness results, particularly for the PET replacement strategy. The PET replacement strategy no longer dominates the baseline 4-NS and SLNB strategies when the probabilities of relapse for FN patients are increased.

The change in cost of SLNB has an impact on the cost-effectiveness of the strategy of adding MRI before SLNB. When the cost of SLNB is decreased, the strategy of adding MRI before SLNB becomes less cost-effective than the baseline SLNB strategy.

Sensitivity analyses also suggest that cost-effectiveness results appear to be robust when the model inputs of sensitivity of PET, the additional costs of lymphoedema and the costs of 4-NS are changed.

In general, the MRI replacement strategy remains the most cost-effective strategy and dominates the baseline 4-NS and SLNB strategies in most of the sensitivity analyses undertaken. The PET replacement strategy is not as robust as the MRI replacement strategy, as its cost-effectiveness is significantly affected by the utility decrement for lymphoedema and the probability of relapse for FN patients. When the imaging replacement strategies are excluded, the cost-effectiveness of adding MRI before 4-NS or SLNB is affected by the change of model inputs of MRI sensitivity and specificity, the probabilities of relapse for FN patients and the cost of SLNB.

Limitations of the analysis

The main limitations of the cost-effectiveness analyses include:

• Evidence on the sensitivity and specificity of 4-NS is limited and is less robust than evidence on SLNB. The adverse event rates of 4-NS are assumed to be the same as rates for SLNB, but this may underestimate the adverse event rates of 4-NS.

• The sensitivity and specificity of MRI and PET vary significantly across different studies. The evidence for MRI is less robust than the evidence for PET, given that it based on a limited number of small studies.

• The sensitivity and specificity of MRI and PET are assumed to be independent of previous tests.

• The quality of evidence on the cost of lymphoedema and the impact of lymphoedema on quality of life is poor.

• The quality of evidence on the costs of short-term adverse events is poor.

• More robust costing information for the baseline sampling procedures and for MRI and PET are also needed for the UK setting.

Summary of clinical results

Summary of cost-effectiveness and benefits versus risks for each strategy

The decision model evaluated the benefits, harms and cost-effectiveness of PET and MRI, either as a replacement for SLNB or 4-NS or as an additional test prior to SLNB or 4-NS. Comparison of SLNB with 4-NS was not part of the remit of this assessment, but both techniques are in current use. Therefore, two baseline strategies (SLNB and 4-NS) were modelled and strategies involving SLNB were assessed separately to strategies involving 4-NS.

The results of the decision modelling suggest that the most cost-effective strategy is to replace sampling (SLNB or 4-NS) with MRI. This strategy dominates the baseline SLNB and 4-NS strategies, generating higher QALYs and lower costs. Axillary sampling (SLNB and 4-NS) is avoided for all patients, leading to fewer adverse effects, especially lymphoedema, which has an assumed lifelong impact on quality of life. However, MRI has lower sensitivity than SLNB and 4-NS (leading to more FNs) and a lower specificity (leading to more FPs). Patients with a FN diagnosis will not receive ALND or adjuvant therapy, leading to a higher risk of cancer recurrence. Patients with a FP diagnosis will receive ALND unnecessarily, with the accompanying increased risk of adverse effects. On the population level, the model results suggest that the MRI replacement strategy costs less, and also the health benefits gained by the majority of patients outweigh the negative impact on survival and quality of life of a small proportion of patients. This strategy may, however, be rejected on clinical grounds, owing to the increase in FP and FN cases compared with current practice.

If the replacement strategies are rejected on clinical grounds, the most cost-effective strategy is predicted to be the baseline 4-NS strategy (compared with 4-NS as the baseline) or the use of MRI prior to SLNB (compared with SLNB as the baseline). For strategies that place MRI or PET before sampling, patients with TP PET or MRI results receive immediate ALND without the need to carry out a separate SLNB or 4-NS procedure. However, the need for a second surgical procedure may also be averted through the use of intraoperative cytology, whereby the axillary sampling procedure may be converted immediately to full ALND for patients with positive nodes. Intraoperative cytology is currently under investigation at a number of units. 139 The number of FN cases is also reduced due to the use of two sequential tests. The disadvantage is that, due to the lower specificity of PET and MRI compared with SLNB and 4-NS, there are still more patients with FP results who receive ALND unnecessarily, with the accompanying increased risk of adverse effects. The total QALYs generated by the strategies of adding MRI prior to 4-NS or SLNB are very similar to those in the baseline strategies and these results are not considered to be robust, based on the quality of the data available.

The model results indicate that, as would be expected, patients with axillary lymph node metastases (either TP or FN) have lower survival rates than patients with negative nodal status. Patients with FN results have the lowest survival rates because they do not receive ALND or chemotherapy that may reduce the risk of recurrence. Compared with the two baseline strategies, the overall survival rates are lower for the MRI or PET replacement strategies (due to an increase in FNs) and higher for the strategies where MRI or PET are placed before 4-NS and SLNB (due to a decrease in FNs). However, the absolute differences in overall survival rate among tested diagnostic strategies are relatively small, as the absolute number of patients with FN results only account for a small proportion of all patients.

Sensitivity analysis suggests that the MRI replacement strategy remains the most cost-effective strategy in the majority of the one-way sensitivity analyses undertaken. The sensitivity analyses indicate that the cost-effectiveness of the PET replacement strategy is significantly affected by the assumption relating to utility decrements for lymphoedema and the probability of relapse for FN patients. The quality of data on long-term utility decrements generated by lymphoedema is particularly poor. If the PET and MRI replacement strategies are excluded, the cost-effectiveness of the strategies of adding MRI prior to 4-NS or SLNB is affected by the assumed MRI sensitivity and specificity, the probability of relapse for FN patients, and the costs of SLNB. The results relating to addition of MRI prior to 4-NS or SLNB are not considered to be robust.

Prioritisation Group

1. Director, NIHR HTA programme, Professor of Clinical Pharmacology, University of Liverpool

2. Professor Imti Choonara, Professor in Child Health, Academic Division of Child Health, University of Nottingham

   Chair – Pharmaceuticals Panel

3. Dr Bob Coates, Consultant Advisor – Disease Prevention Panel

4. Dr Andrew Cook, Consultant Advisor – Intervention Procedures Panel

5. Dr Peter Davidson, Director of NETSCC, Health Technology Assessment

6. Dr Nick Hicks, Consultant Adviser – Diagnostic Technologies and Screening Panel, Consultant Advisor–Psychological and Community Therapies Panel

7. Ms Susan Hird, Consultant Advisor, External Devices and Physical Therapies Panel

8. Professor Sallie Lamb, Director, Warwick Clinical Trials Unit, Warwick Medical School, University of Warwick

   Chair – HTA Clinical Evaluation and Trials Board

9. Professor Jonathan Michaels, Professor of Vascular Surgery, Sheffield Vascular Institute, University of Sheffield

   Chair – Interventional Procedures Panel

10. Professor Ruairidh Milne, Director – External Relations

11. Dr John Pounsford, Consultant Physician, Directorate of Medical Services, North Bristol NHS Trust

    Chair – External Devices and Physical Therapies Panel

12. Dr Vaughan Thomas, Consultant Advisor – Pharmaceuticals Panel, Clinical

    Lead – Clinical Evaluation Trials Prioritisation Group

13. Professor Margaret Thorogood, Professor of Epidemiology, Health Sciences Research Institute, University of Warwick

    Chair – Disease Prevention Panel

14. Professor Lindsay Turnbull, Professor of Radiology, Centre for the MR Investigations, University of Hull

    Chair – Diagnostic Technologies and Screening Panel

15. Professor Scott Weich, Professor of Psychiatry, Health Sciences Research Institute, University of Warwick

    Chair – Psychological and Community Therapies Panel

16. Professor Hywel Williams, Director of Nottingham Clinical Trials Unit, Centre of Evidence-Based Dermatology, University of Nottingham

    Chair – HTA Commissioning Board

    Deputy HTA Programme Director

HTA Commissioning Board

1. Professor of Dermato-Epidemiology, Centre of Evidence-Based Dermatology, University of Nottingham

2. Professor of General Practice, Department of Primary Health Care, University of Oxford Programme Director,

3. Professor of Clinical Pharmacology, Director, NIHR HTA programme, University of Liverpool

1. Professor Ann Ashburn, Professor of Rehabilitation and Head of Research, Southampton General Hospital

2. Professor Deborah Ashby, Professor of Medical Statistics and Clinical Trials, Queen Mary, Department of Epidemiology and Public Health, Imperial College London

3. Professor Peter Brocklehurst, Director, National Perinatal Epidemiology Unit, University of Oxford

4. Professor John Cairns, Professor of Health Economics, London School of Hygiene and Tropical Medicine

5. Professor Peter Croft, Director of Primary Care Sciences Research Centre, Keele University

6. Professor Jenny Donovan, Professor of Social Medicine, University of Bristol

7. Professor Jonathan Green, Professor and Acting Head of Department, Child and Adolescent Psychiatry, University of Manchester Medical School

8. Professor John W Gregory, Professor in Paediatric Endocrinology, Department of Child Health, Wales School of Medicine, Cardiff University

9. Professor Steve Halligan, Professor of Gastrointestinal Radiology, University College Hospital, London

10. Professor Freddie Hamdy, Professor of Urology, Head of Nuffield Department of Surgery, University of Oxford

11. Professor Allan House, Professor of Liaison Psychiatry, University of Leeds

12. Dr Martin J Landray, Reader in Epidemiology, Honorary Consultant Physician, Clinical Trial Service Unit, University of Oxford

13. Professor Stephen Morris, Professor of Health Economics, University College London, Research Department of Epidemiology and Public Health, University College London

14. Professor E Andrea Nelson, Professor of Wound Healing and Director of Research, School of Healthcare, University of Leeds

15. Professor John David Norris, Chair in Clinical Trials and Biostatistics, Robertson Centre for Biostatistics, University of Glasgow

16. Dr Rafael Perera, Lecturer in Medical Statisitics, Department of Primary Health Care, University of Oxford

17. Professor James Raftery, Chair of NETSCC and Director of the Wessex Institute, University of Southampton

18. Professor Barney Reeves, Professorial Research Fellow in Health Services Research, Department of Clinical Science, University of Bristol

19. Professor Martin Underwood, Warwick Medical School, University of Warwick

20. Professor Marion Walker, Professor in Stroke Rehabilitation, Associate Director UK Stroke Research Network, University of Nottingham

21. Dr Duncan Young, Senior Clinical Lecturer and Consultant, Nuffield Department of Anaesthetics, University of Oxford

22. Professor Stephen Morris, Professor of Health Economics, University College London, Research Department of Epidemiology and Public Health, University College London

23. Professor E Andrea Nelson, Professor of Wound Healing and Director of Research, School of Healthcare, University of Leeds

24. Professor John David Norris Chair in Clinical Trials and Biostatistics, Robertson Centre for Biostatistics, University of Glasgow

25. Dr Rafael Perera, Lecturer in Medical Statisitics, Department of Primary Health Care, University of Oxford

26. Professor James Raftery, Chair of NETSCC and Director of the Wessex Institute, University of Southampton

27. Professor Barney Reeves, Professorial Research Fellow in Health Services Research, Department of Clinical Science, University of Bristol

28. Professor Martin Underwood, Warwick Medical School, University of Warwick

29. Professor Marion Walker, Professor in Stroke Rehabilitation, Associate Director UK Stroke Research Network, University of Nottingham

30. Dr Duncan Young, Senior Clinical Lecturer and Consultant, Nuffield Department of Anaesthetics, University of Oxford

1. Dr Morven Roberts, Clinical Trials Manager, Health Services and Public Health Services Board, Medical Research Council

HTA Clinical Evaluation and Trials Board

1. Director, Warwick Clinical Trials Unit, Warwick Medical School, University of Warwick and Professor of Rehabilitation, Nuffield Department of Orthopaedic, Rheumatology and Musculoskeletal Sciences, University of Oxford

2. Professor of the Psychology of Health Care, Leeds Institute of Health Sciences, University of Leeds

3. Director, NIHR HTA programme, Professor of Clinical Pharmacology, University of Liverpool

1. Professor Keith Abrams, Professor of Medical Statistics, Department of Health Sciences, University of Leicester

2. Professor Martin Bland, Professor of Health Statistics, Department of Health Sciences, University of York

3. Professor Jane Blazeby, Professor of Surgery and Consultant Upper GI Surgeon, Department of Social Medicine, University of Bristol

4. Professor Julia M Brown, Director, Clinical Trials Research Unit, University of Leeds

5. Professor Alistair Burns, Professor of Old Age Psychiatry, Psychiatry Research Group, School of Community-Based Medicine, The University of Manchester & National Clinical Director for Dementia, Department of Health

6. Dr Jennifer Burr, Director, Centre for Healthcare Randomised trials (CHART), University of Aberdeen

7. Professor Linda Davies, Professor of Health Economics, Health Sciences Research Group, University of Manchester

8. Professor Simon Gilbody, Prof of Psych Medicine and Health Services Research, Department of Health Sciences, University of York

9. Professor Steven Goodacre, Professor and Consultant in Emergency Medicine, School of Health and Related Research, University of Sheffield

10. Professor Dyfrig Hughes, Professor of Pharmacoeconomics, Centre for Economics and Policy in Health, Institute of Medical and Social Care Research, Bangor University

11. Professor Paul Jones, Professor of Respiratory Medicine, Department of Cardiac and Vascular Science, St George‘s Hospital Medical School, University of London

12. Professor Khalid Khan, Professor of Women’s Health and Clinical Epidemiology, Barts and the London School of Medicine, Queen Mary, University of London

13. Professor Richard J McManus, Professor of Primary Care Cardiovascular Research, Primary Care Clinical Sciences Building, University of Birmingham

14. Professor Helen Rodgers, Professor of Stroke Care, Institute for Ageing and Health, Newcastle University

15. Professor Ken Stein, Professor of Public Health, Peninsula Technology Assessment Group, Peninsula College of Medicine and Dentistry, Universities of Exeter and Plymouth

16. Professor Jonathan Sterne, Professor of Medical Statistics and Epidemiology, Department of Social Medicine, University of Bristol

17. Mr Andy Vail, Senior Lecturer, Health Sciences Research Group, University of Manchester

18. Professor Clare Wilkinson, Professor of General Practice and Director of Research North Wales Clinical School, Department of Primary Care and Public Health, Cardiff University

19. Dr Ian B Wilkinson, Senior Lecturer and Honorary Consultant, Clinical Pharmacology Unit, Department of Medicine, University of Cambridge

1. Ms Kate Law, Director of Clinical Trials, Cancer Research UK

2. Dr Morven Roberts, Clinical Trials Manager, Health Services and Public Health Services Board, Medical Research Council

Diagnostic Technologies and Screening Panel

1. Scientific Director of the Centre for Magnetic Resonance Investigations and YCR Professor of Radiology, Hull Royal Infirmary

2. Professor Judith E Adams, Consultant Radiologist, Manchester Royal Infirmary, Central Manchester & Manchester Children’s University Hospitals NHS Trust, and Professor of Diagnostic Radiology, University of Manchester

3. Mr Angus S Arunkalaivanan, Honorary Senior Lecturer, University of Birmingham and Consultant Urogynaecologist and Obstetrician, City Hospital, Birmingham

4. Dr Stephanie Dancer, Consultant Microbiologist, Hairmyres Hospital, East Kilbride

5. Dr Diane Eccles, Professor of Cancer Genetics, Wessex Clinical Genetics Service, Princess Anne Hospital

6. Dr Trevor Friedman, Consultant Liason Psychiatrist, Brandon Unit, Leicester General Hospital

7. Dr Ron Gray, Consultant, National Perinatal Epidemiology Unit, Institute of Health Sciences, University of Oxford

8. Professor Paul D Griffiths, Professor of Radiology, Academic Unit of Radiology, University of Sheffield

9. Mr Martin Hooper, Service User Representative

10. Professor Anthony Robert Kendrick, Associate Dean for Clinical Research and Professor of Primary Medical Care, University of Southampton

11. Dr Anne Mackie, Director of Programmes, UK National Screening Committee, London

12. Mr David Mathew, Service User Representative

13. Dr Michael Millar, Consultant Senior Lecturer in Microbiology, Department of Pathology & Microbiology, Barts and The London NHS Trust, Royal London Hospital

14. Mrs Una Rennard, Service User Representative

15. Dr Stuart Smellie, Consultant in Clinical Pathology, Bishop Auckland General Hospital

16. Ms Jane Smith, Consultant Ultrasound Practitioner, Leeds Teaching Hospital NHS Trust, Leeds

17. Dr Allison Streetly, Programme Director, NHS Sickle Cell and Thalassaemia Screening Programme, King’s College School of Medicine

18. Dr Alan J Williams, Consultant Physician, General and Respiratory Medicine, The Royal Bournemouth Hospital

1. Dr Tim Elliott, Team Leader, Cancer Screening, Department of Health

2. Dr Catherine Moody, Programme Manager, Medical Research Council

3. Professor Julietta Patrick, Director, NHS Cancer Screening Programme, Sheffield

4. Dr Kay Pattison, Senior NIHR Programme Manager, Department of Health

5. Professor Tom Walley, CBE, Director, NIHR HTA programme, Professor of Clinical Pharmacology, University of Liverpool

6. Dr Ursula Wells, Principal Research Officer, Policy Research Programme, Department of Health

Disease Prevention Panel

1. Professor of Epidemiology, University of Warwick Medical School, Coventry

2. Dr Robert Cook, Clinical Programmes Director, Bazian Ltd, London

3. Dr Colin Greaves, Senior Research Fellow, Peninsula Medical School (Primary Care)

4. Mr Michael Head, Service User Representative

5. Professor Cathy Jackson, Professor of Primary Care Medicine, Bute Medical School, University of St Andrews

6. Dr Russell Jago, Senior Lecturer in Exercise, Nutrition and Health, Centre for Sport, Exercise and Health, University of Bristol

7. Dr Julie Mytton, Consultant in Child Public Health, NHS Bristol

8. Professor Irwin Nazareth, Professor of Primary Care and Director, Department of Primary Care and Population Sciences, University College London

9. Dr Richard Richards, Assistant Director of Public Health, Derbyshire Country Primary Care Trust

10. Professor Ian Roberts, Professor of Epidemiology and Public Health, London School of Hygiene & Tropical Medicine

11. Dr Kenneth Robertson, Consultant Paediatrician, Royal Hospital for Sick Children, Glasgow

12. Dr Catherine Swann, Associate Director, Centre for Public Health Excellence, NICE

13. Professor Carol Tannahill, Glasgow Centre for Population Health

14. Mrs Jean Thurston, Service User Representative

15. Professor David Weller, Head, School of Clinical Science and Community Health, University of Edinburgh

1. Ms Christine McGuire, Research & Development, Department of Health

2. Dr Kay Pattison Senior NIHR Programme Manager, Department of Health

3. Professor Tom Walley, CBE, Director, NIHR HTA programme, Professor of Clinical Pharmacology, University of Liverpool

External Devices and Physical Therapies Panel

1. Consultant Physician North Bristol NHS Trust

2. Reader in Wound Healing and Director of Research, University of Leeds

3. Professor Bipin Bhakta, Charterhouse Professor in Rehabilitation Medicine, University of Leeds

4. Mrs Penny Calder, Service User Representative

5. Dr Dawn Carnes, Senior Research Fellow, Barts and the London School of Medicine and Dentistry

6. Dr Emma Clark, Clinician Scientist Fellow & Cons. Rheumatologist, University of Bristol

7. Mrs Anthea De Barton-Watson, Service User Representative

8. Professor Nadine Foster, Professor of Musculoskeletal Health in Primary Care Arthritis Research, Keele University

9. Dr Shaheen Hamdy, Clinical Senior Lecturer and Consultant Physician, University of Manchester

10. Professor Christine Norton, Professor of Clinical Nursing Innovation, Bucks New University and Imperial College Healthcare NHS Trust

11. Dr Lorraine Pinnigton, Associate Professor in Rehabilitation, University of Nottingham

12. Dr Kate Radford, Senior Lecturer (Research), University of Central Lancashire

13. Mr Jim Reece, Service User Representative

14. Professor Maria Stokes, Professor of Neuromusculoskeletal Rehabilitation, University of Southampton

15. Dr Pippa Tyrrell, Senior Lecturer/Consultant, Salford Royal Foundation Hospitals’ Trust and University of Manchester

16. Dr Sarah Tyson, Senior Research Fellow & Associate Head of School, University of Salford

17. Dr Nefyn Williams, Clinical Senior Lecturer, Cardiff University

1. Dr Kay Pattison, Senior NIHR Programme Manager, Department of Health

2. Professor Tom Walley, CBE, Director, NIHR HTA programme, Professor of Clinical Pharmacology, University of Liverpool

3. Dr Ursula Wells, Principal Research Officer, Policy Research Programme, Department of Health

Interventional Procedures Panel

1. Professor of Vascular Surgery, University of Sheffield

2. Consultant Colorectal Surgeon, Bristol Royal Infirmary

3. Mrs Isabel Boyer, Service User Representative

4. Mr David P Britt, Service User Representative

5. Mr Sankaran ChandraSekharan, Consultant Surgeon, Breast Surgery, Colchester Hospital University NHS Foundation Trust

6. Professor Nicholas Clarke, Consultant Orthopaedic Surgeon, Southampton University Hospitals NHS Trust

7. Ms Leonie Cooke, Service User Representative

8. Mr Seamus Eckford, Consultant in Obstetrics & Gynaecology, North Devon District Hospital

9. Professor David Taggart, Consultant Cardiothoracic Surgeon, John Radcliffe Hospital

10. Professor Sam Eljamel, Consultant Neurosurgeon, Ninewells Hospital and Medical School, Dundee

11. Dr Adele Fielding, Senior Lecturer and Honorary Consultant in Haematology, University College London Medical School

12. Dr Matthew Hatton, Consultant in Clinical Oncology, Sheffield Teaching Hospital Foundation Trust

13. Dr John Holden, General Practitioner, Garswood Surgery, Wigan

14. Professor Nicholas James, Professor of Clinical Oncology, School of Cancer Sciences, University of Birmingham

15. Dr Fiona Lecky, Senior Lecturer/Honorary Consultant in Emergency Medicine, University of Manchester/Salford Royal Hospitals NHS Foundation Trust

16. Dr Nadim Malik, Consultant Cardiologist/ Honorary Lecturer, University of Manchester

17. Mr Hisham Mehanna, Consultant & Honorary Associate Professor, University Hospitals Coventry & Warwickshire NHS Trust

18. Dr Jane Montgomery, Consultant in Anaesthetics and Critical Care, South Devon Healthcare NHS Foundation Trust

19. Professor Jon Moss, Consultant Interventional Radiologist, North Glasgow Hospitals University NHS Trust

20. Dr Simon Padley, Consultant Radiologist, Chelsea & Westminster Hospital

21. Dr Ashish Paul, Medical Director, Bedfordshire PCT

22. Dr Sarah Purdy, Consultant Senior Lecturer, University of Bristol

23. Professor Yit Chiun Yang, Consultant Ophthalmologist, Royal Wolverhampton Hospitals NHS Trust

1. Dr Kay Pattison, Senior NIHR Programme Manager, Department of Health

2. Dr Morven Roberts, Clinical Trials Manager, Health Services and Public Health Services Board, Medical Research Council

3. Professor Tom Walley, CBE, Director, NIHR HTA programme, Professor of Clinical Pharmacology, University of Liverpool

4. Dr Ursula Wells, Principal Research Officer, Policy Research Programme, Department of Health

Pharmaceuticals Panel

1. Professor in Child Health, University of Nottingham

2. Senior Lecturer in Clinical Pharmacology, University of East Anglia

3. Dr Martin Ashton-Key, Medical Advisor, National Commissioning Group, NHS London

4. Mr John Chapman, Service User Representative

5. Dr Peter Elton, Director of Public Health, Bury Primary Care Trust

6. Dr Peter Elton, Director of Public Health, Bury Primary Care Trust

7. Dr Ben Goldacre, Research Fellow, Division of Psychological Medicine and Psychiatry, King’s College London

8. Dr James Gray, Consultant Microbiologist, Department of Microbiology, Birmingham Children’s Hospital NHS Foundation Trust

9. Ms Kylie Gyertson, Oncology and Haematology Clinical Trials Manager, Guy’s and St Thomas’ NHS Foundation Trust London

10. Dr Jurjees Hasan, Consultant in Medical Oncology, The Christie, Manchester

11. Dr Carl Heneghan Deputy Director Centre for Evidence-Based Medicine and Clinical Lecturer, Department of Primary Health Care, University of Oxford

12. Dr Dyfrig Hughes, Reader in Pharmacoeconomics and Deputy Director, Centre for Economics and Policy in Health, IMSCaR, Bangor University

13. Dr Maria Kouimtzi, Pharmacy and Informatics Director, Global Clinical Solutions, Wiley-Blackwell

14. Professor Femi Oyebode, Consultant Psychiatrist and Head of Department, University of Birmingham

15. Dr Andrew Prentice, Senior Lecturer and Consultant Obstetrician and Gynaecologist, The Rosie Hospital, University of Cambridge

16. Ms Amanda Roberts, Service User Representative

17. Dr Martin Shelly, General Practitioner, Silver Lane Surgery, Leeds

18. Dr Gillian Shepherd, Director, Health and Clinical Excellence, Merck Serono Ltd

19. Mrs Katrina Simister, Assistant Director New Medicines, National Prescribing Centre, Liverpool

20. Professor Donald Singer Professor of Clinical Pharmacology and Therapeutics, Clinical Sciences Research Institute, CSB, University of Warwick Medical School

21. Mr David Symes, Service User Representative

22. Dr Arnold Zermansky, General Practitioner, Senior Research Fellow, Pharmacy Practice and Medicines Management Group, Leeds University

1. Dr Kay Pattison, Senior NIHR Programme Manager, Department of Health

2. Mr Simon Reeve, Head of Clinical and Cost-Effectiveness, Medicines, Pharmacy and Industry Group, Department of Health

3. Dr Heike Weber, Programme Manager, Medical Research Council

4. Professor Tom Walley, CBE, Director, NIHR HTA programme, Professor of Clinical Pharmacology, University of Liverpool

5. Dr Ursula Wells, Principal Research Officer, Policy Research Programme, Department of Health

Psychological and Community Therapies Panel

1. Professor of Psychiatry, University of Warwick, Coventry

2. Consultant & University Lecturer in Psychiatry, University of Cambridge

3. Professor Jane Barlow, Professor of Public Health in the Early Years, Health Sciences Research Institute, Warwick Medical School

4. Dr Sabyasachi Bhaumik, Consultant Psychiatrist, Leicestershire Partnership NHS Trust

5. Mrs Val Carlill, Service User Representative

6. Dr Steve Cunningham, Consultant Respiratory Paediatrician, Lothian Health Board

7. Dr Anne Hesketh, Senior Clinical Lecturer in Speech and Language Therapy, University of Manchester

8. Dr Peter Langdon, Senior Clinical Lecturer, School of Medicine, Health Policy and Practice, University of East Anglia

9. Dr Yann Lefeuvre, GP Partner, Burrage Road Surgery, London

10. Dr Jeremy J Murphy, Consultant Physician and Cardiologist, County Durham and Darlington Foundation Trust

11. Dr Richard Neal, Clinical Senior Lecturer in General Practice, Cardiff University

12. Mr John Needham, Service User Representative

13. Ms Mary Nettle, Mental Health User Consultant

14. Professor John Potter, Professor of Ageing and Stroke Medicine, University of East Anglia

15. Dr Greta Rait, Senior Clinical Lecturer and General Practitioner, University College London

16. Dr Paul Ramchandani, Senior Research Fellow/Cons. Child Psychiatrist, University of Oxford

17. Dr Karen Roberts, Nurse/Consultant, Dunston Hill Hospital, Tyne and Wear

18. Dr Karim Saad, Consultant in Old Age Psychiatry, Coventry and Warwickshire Partnership Trust

19. Dr Lesley Stockton, Lecturer, School of Health Sciences, University of Liverpool

20. Dr Simon Wright, GP Partner, Walkden Medical Centre, Manchester

1. Dr Kay Pattison, Senior NIHR Programme Manager, Department of Health

2. Dr Morven Roberts, Clinical Trials Manager, Health Services and Public Health Services Board, Medical Research Council

3. Professor Tom Walley, CBE, Director, NIHR HTA programme, Professor of Clinical Pharmacology, University of Liverpool

4. Dr Ursula Wells, Principal Research Officer, Policy Research Programme, Department of Health

Expert Advisory Network

1. Professor Douglas Altman, Professor of Statistics in Medicine, Centre for Statistics in Medicine, University of Oxford

2. Professor John Bond, Professor of Social Gerontology & Health Services Research, University of Newcastle upon Tyne

3. Professor Andrew Bradbury, Professor of Vascular Surgery, Solihull Hospital, Birmingham

4. Mr Shaun Brogan, Chief Executive, Ridgeway Primary Care Group, Aylesbury

5. Mrs Stella Burnside OBE, Chief Executive, Regulation and Improvement Authority, Belfast

6. Ms Tracy Bury, Project Manager, World Confederation of Physical Therapy, London

7. Professor Iain T Cameron, Professor of Obstetrics and Gynaecology and Head of the School of Medicine, University of Southampton

8. Professor Bruce Campbell, Consultant Vascular & General Surgeon, Royal Devon & Exeter Hospital, Wonford

9. Dr Christine Clark, Medical Writer and Consultant Pharmacist, Rossendale

10. Professor Collette Clifford, Professor of Nursing and Head of Research, The Medical School, University of Birmingham

11. Professor Barry Cookson, Director, Laboratory of Hospital Infection, Public Health Laboratory Service, London

12. Dr Carl Counsell, Clinical Senior Lecturer in Neurology, University of Aberdeen

13. Professor Howard Cuckle, Professor of Reproductive Epidemiology, Department of Paediatrics, Obstetrics & Gynaecology, University of Leeds

14. Professor Carol Dezateux, Professor of Paediatric Epidemiology, Institute of Child Health, London

15. Mr John Dunning, Consultant Cardiothoracic Surgeon, Papworth Hospital NHS Trust, Cambridge

16. Mr Jonothan Earnshaw, Consultant Vascular Surgeon, Gloucestershire Royal Hospital, Gloucester

17. Professor Martin Eccles, Professor of Clinical Effectiveness, Centre for Health Services Research, University of Newcastle upon Tyne

18. Professor Pam Enderby, Dean of Faculty of Medicine, Institute of General Practice and Primary Care, University of Sheffield

19. Professor Gene Feder, Professor of Primary Care Research & Development, Centre for Health Sciences, Barts and The London School of Medicine and Dentistry

20. Mr Leonard R Fenwick, Chief Executive, Freeman Hospital, Newcastle upon Tyne

21. Mrs Gillian Fletcher, Antenatal Teacher and Tutor and President, National Childbirth Trust, Henfield

22. Professor Jayne Franklyn, Professor of Medicine, University of Birmingham

23. Mr Tam Fry, Honorary Chairman, Child Growth Foundation, London

24. Professor Fiona Gilbert, Consultant Radiologist and NCRN Member, University of Aberdeen

25. Professor Paul Gregg, Professor of Orthopaedic Surgical Science, South Tees Hospital NHS Trust

26. Bec Hanley, Co-director, TwoCan Associates, West Sussex

27. Dr Maryann L Hardy, Senior Lecturer, University of Bradford

28. Mrs Sharon Hart, Healthcare Management Consultant, Reading

29. Professor Robert E Hawkins, CRC Professor and Director of Medical Oncology, Christie CRC Research Centre, Christie Hospital NHS Trust, Manchester

30. Professor Richard Hobbs, Head of Department of Primary Care & General Practice, University of Birmingham

31. Professor Alan Horwich, Dean and Section Chairman, The Institute of Cancer Research, London

32. Professor Allen Hutchinson, Director of Public Health and Deputy Dean of ScHARR, University of Sheffield

33. Professor Peter Jones, Professor of Psychiatry, University of Cambridge, Cambridge

34. Professor Stan Kaye, Cancer Research UK Professor of Medical Oncology, Royal Marsden Hospital and Institute of Cancer Research, Surrey

35. Dr Duncan Keeley, General Practitioner (Dr Burch & Ptnrs), The Health Centre, Thame

36. Dr Donna Lamping, Research Degrees Programme Director and Reader in Psychology, Health Services Research Unit, London School of Hygiene and Tropical Medicine, London

37. Professor James Lindesay, Professor of Psychiatry for the Elderly, University of Leicester

38. Professor Julian Little, Professor of Human Genome Epidemiology, University of Ottawa

39. Professor Alistaire McGuire, Professor of Health Economics, London School of Economics

40. Professor Neill McIntosh, Edward Clark Professor of Child Life and Health, University of Edinburgh

41. Professor Rajan Madhok, Consultant in Public Health, South Manchester Primary Care Trust

42. Professor Sir Alexander Markham, Director, Molecular Medicine Unit, St James’s University Hospital, Leeds

43. Dr Peter Moore, Freelance Science Writer, Ashtead

44. Dr Andrew Mortimore, Public Health Director, Southampton City Primary Care Trust

45. Dr Sue Moss, Associate Director, Cancer Screening Evaluation Unit, Institute of Cancer Research, Sutton

46. Professor Miranda Mugford, Professor of Health Economics and Group Co-ordinator, University of East Anglia

47. Professor Jim Neilson, Head of School of Reproductive & Developmental Medicine and Professor of Obstetrics and Gynaecology, University of Liverpool

48. Mrs Julietta Patnick, Director, NHS Cancer Screening Programmes, Sheffield

49. Professor Robert Peveler, Professor of Liaison Psychiatry, Royal South Hants Hospital, Southampton

50. Professor Chris Price, Director of Clinical Research, Bayer Diagnostics Europe, Stoke Poges

51. Professor William Rosenberg, Professor of Hepatology and Consultant Physician, University of Southampton

52. Professor Peter Sandercock, Professor of Medical Neurology, Department of Clinical Neurosciences, University of Edinburgh

53. Dr Philip Shackley, Senior Lecturer in Health Economics, Sheffield Vascular Institute, University of Sheffield

54. Dr Eamonn Sheridan, Consultant in Clinical Genetics, St James’s University Hospital, Leeds

55. Dr Margaret Somerville, Director of Public Health Learning, Peninsula Medical School, University of Plymouth

56. Professor Sarah Stewart-Brown, Professor of Public Health, Division of Health in the Community, University of Warwick, Coventry

57. Dr Nick Summerton, GP Appraiser and Codirector, Research Network, Yorkshire Clinical Consultant, Primary Care and Public Health, University of Oxford

58. Professor Ala Szczepura, Professor of Health Service Research, Centre for Health Services Studies, University of Warwick, Coventry

59. Dr Ross Taylor, Senior Lecturer, University of Aberdeen

60. Dr Richard Tiner, Medical Director, Medical Department, Association of the British Pharmaceutical Industry

61. Mrs Joan Webster, Consumer Member, Southern Derbyshire Community Health Council

62. Professor Martin Whittle, Clinical Co-director, National Co-ordinating Centre for Women’s and Children’s Health, Lymington