The diagnostic accuracy and cost-effectiveness of magnetic resonance spectroscopy and enhanced magnetic resonance imaging techniques in aiding the localisation of prostate abnormalities for biopsy: a systematic review and economic evaluation

Diagnosis of prostate cancer

Men with an elevated PSA level and/or abnormal DRE undergo a prostate biopsy, which is normally performed using TRUS/Bx. Some men with negative biopsies will require a repeat biopsy, either immediately [owing to suspicious features on histology, such as atypical small acinar proliferation (ASAP)] or subsequently (owing to a further rise in PSA, persistently raised PSA or rapidly rising PSA). 17 Achieving a diagnosis at repeat biopsy can be challenging either because they have an enlarged central prostate gland due to benign prostatic hyperplasia or because cancer is present in locations difficult to biopsy. 18 Recently, promising alternatives have emerged, which include MRS and enhanced MRI techniques. Lesions identified on MRS/MRI are sampled either by MRI-directed biopsy (tissue obtained under direct MRS/MRI imaging) or by TRUS guidance (TRUS/Bx used to identify and biopsy suspicious lesions on MRS/MRI).

Staging

Staging is performed to determine the extent of disease spread. Information from staging is essential, because it influences treatment decisions and affects prognosis.

Pre-treatment imaging staging of PC is usually individualised according to risk stratification based on clinical parameters that are predictive of the likelihood of extraprostatic disease. These clinical parameters normally include pre-treatment PSA level and rate of rise or doubling time, Gleason score, clinical T staging and volume of disease detected on biopsy. Imaging potentially improves these general estimates of risk by specifically identifying lesions with anatomical abnormalities. The most commonly used imaging modalities for staging of PC are MRI, computed tomography (CT), isotope bone scan and positron emission tomography.

Staging can be divided into local, regional and distant categories. Local staging is usually performed by DRE and MRI; regional staging is performed by either CT or MRI; and distant staging is performed by CT, bone scanning and plain bone radiography. In addition, measurement of PSA level in the blood19–21 and Gleason sum score22 can also yield useful information regarding stage. Pathological staging determines the actual extent of spread (i.e. if it is either confined to, or spread outwith, the prostate gland, or if resected lymph nodes have cancer) through histological examination. The staging system most commonly used is the tumour, node, metastasis (TNM) staging system. 23 This describes the local extent of the primary tumour (T stage), the absence or presence of spread to nearby lymph nodes (N stage) and the absence or presence of metastasis (M stage).

Grading

Grading is the histological assessment of cancer tissue to determine its aggressiveness. This is done on either biopsy tissue, resected tissue (e.g. from transurethral resection of prostate) or surgical specimens. Pathologists usually assign a grade from 1 to 5 to the most common tumour pattern observed and then a second 1–5 grade to the next most common tumour pattern. The Gleason score is the sum of these two grade assignments. 24 This scoring system describes a score between 2 and 10, with ‘2’ being the least aggressive and ‘10’ being the most aggressive,25 although most pathologists now group scores 1–6 as Gleason 6. 26

Use of nomograms to predict treatment outcomes

Nomograms are a means of predicting the probability of important outcomes following treatment using pre-treatment variables as predictors. For PC, several nomograms exist, which predict various outcomes following treatment for men with localised PC, based on pre-treatment variables such as PSA, clinical stage and Gleason score. The outcomes predicted include the probability of biochemical disease recurrence following curative treatment (Kattan nomograms21,27,28 and the D'Amico nomogram29) and the probability of various pathological stages following surgery (Partin tables30). These nomograms may be used by clinicians and health-care professionals with patients and their families to facilitate decision-making. Use of some of these nomograms has enabled the stratification of men with localised PC into risk groups according to their risk of biochemical recurrence if they were treated with radical treatment, such as radical prostatectomy or external beam radiotherapy (EBRT) (Table 1). 31 Studies have shown the added value of MRS and/or MRI in enhancing the value of normograms. 32–34

Monitoring of disease following treatment

Men who have undergone curative treatments are monitored via PSA measurements, to ensure eradication of disease. Patients who develop disease recurrence will have gradual rises in their PSA level (i.e. biochemical recurrence). In addition, men with suspected local recurrence (i.e. in the pelvis) may be imaged with either MRI or CT scans, or undergo TRUS-guided prostate biopsy to confirm local disease recurrence. However, the benefit of these investigations remains controversial. 17 Patients with more rapid rises in PSA level may have disease outside of the pelvis and more extensive investigations are performed, including a bone scan.

Management of localised prostate cancer

A range of treatment options exist for men with localised PC, ranging from active surveillance for low-risk disease, whereby treatment is deferred until the cancer progresses or becomes more aggressive, to minimally invasive treatments that ablate a part of the prostate [such as high-intensity focused ultrasound (HIFU) and cryotherapy] and to immediate curative treatments (including invasive treatments such as radical prostatectomy, radiation treatment or brachytherapy). 35 Curative treatments may result in significant side effects, including urinary incontinence (UI), erectile dysfunction (ED) or troublesome urinary symptoms. 36

Based on current National Institute for Health and Care Excellence (NICE) guidance,9 Table 2 outlines the alternative treatment modalities recommended by PC stage at time of diagnosis. It has been noted that the vast majority of patients identified from second biopsies have localised cancer and few fall into the high-risk group. 2,37,38 Based on routinely collected data on hospital episodes in Scotland (Dr Karina Laing, MSc in Surgical Sciences thesis, University of Edinburgh, May 2012, personal communication), it is estimated that the majority of patients with localised disease receive active surveillance (40%), radical prostatectomy (35%) or EBRT (25%) in the first year following diagnosis.

Management of locally advanced prostate cancer

The vast majority of patients with locally advanced PC will undergo potentially curative hormone manipulation [castration, luteinising hormone-releasing hormone (LHRH) agonists or antagonists] for a minimum of 2 years plus radiotherapy. In the UK, for radical radiotherapy, most men receive 72 grays (Gy) in 36–37 fractions.

A small percentage of men may undergo radical prostatectomy for previously unsuspected T3 disease, T3 disease with severe lower urinary tract symptoms or patient preference where radiotherapy is contraindicated or problematic. Some will be cured by their surgery but those who are not will mostly be offered adjuvant radiotherapy. Men who would not benefit from radical treatment because of comorbidities are usually offered immediate or deferred hormone manipulation.

Management of metastatic disease

Patients who are initially diagnosed with metastatic disease receive first-line treatment with hormone manipulation. When first-line treatment fails, second-line hormone manipulation with the addition of an anti-androgen is usually initiated. If this is unsuccessful, those who are fit enough are offered chemotherapy. If unsuitable for chemotherapy, or after unsuccessful chemotherapy, third-line hormonal treatment may be initiated. Timing of third-line hormonal treatment, with respect to chemotherapy, varies throughout the UK and may change with the introduction of abiraterone (Zytiga^®, Janssen Biotech). 39

Magnetic resonance spectroscopy

Further to imaging of water and lipids, which is normally performed with MRI, MRS is a technique that provides detail on protons of molecules other than water and lipids. MRS makes use of the slight differences in chemical environment of protons attached to small metabolites present in the tissue or organ of interest. Signals of the different protons in these molecules are presented in a spectrum, in which the position on the x-axis is representative for the exact so-called chemical shift of the protons at hand (which molecule), and the intensity on the y-axis represents the amount of that particular proton pool present (how much of that molecule is present). In this way, MRS can give quantitative information on the presence and quantity of metabolites in the prostate. Magnetic resonance spectroscopy imaging (MRSI) does the same, but also provides this information according to spatial location of spectra superimposed on an imaginary two- or three-dimensional grid over the prostate.

In the prostate, three-dimensional MRSI is the current standard of doing spectroscopy, providing spectra of the whole organ with a spatial resolution in the order of 0.5 cc. 45–47 In the prostate the relative concentrations of four metabolites are routinely detectable:

1. citrate, an intermediate of the Krebs cycle, which accumulates in the luminal space of healthy prostate tissue

2. choline, free and phosphorylated choline compounds, which are involved in the phospholipid metabolism of the cell, elevated in cancer tissue

3. creatine, involved in the energy metabolism of cells

4. polyamines (spermine, spermidine and more), accumulating in the luminal space.

As the chemical shifts of choline, polyamines and creatine do not differ greatly, these resonances cannot always be separated, and are therefore incorporated into one clinically useful biomarker for the presence of PC: the choline (+ polyamines) + creatine to citrate ratio (CC/C). After spectral fitting of the different metabolites, this CC/C ratio can be calculated and used either qualitatively48 or quantitatively49 in the so-called standardised threshold approach50,51 to estimate the presence and aggressiveness of cancer in prostate tissue. 52

Differences in the concentrations of these metabolites between normal and malignant prostate tissues allow for increasing the accuracy of staging among less-experienced readers, and decreasing interobserver variability. 53 Furthermore, correlations have been demonstrated between the metabolic signal pattern and a pathological Gleason score, suggesting the potential for a non-invasive assessment of tumour aggressiveness. 52,54

Dynamic contrast enhanced magnetic resonance imaging

Dynamic contrast-enhanced MRI is a fast T1 weighted imaging technique that dynamically measures a bolus pass of an intravenously administrated MR contrast agent through the prostate. For its nutrient and oxygen supply, a tumour forms new vessels made through the process of neoangiogenesis. In tumour tissue these vessels are often leaky or incomplete, which makes it easier for a contrast agent to extravasate into the extravascular extracellular space. In this extracellular space, the gadolinium-based contrast agent increases the signal intensity of T1 weighted images. In this way, tissues with increased perfusion and vessel leakage stand out with respect to normally perfused tissue, which enhances less.

Three-dimensional DCE-MRI measures the time course of the contrast agent passing through the prostate by repeatedly acquiring three-dimensional T1 weighted images at high temporal resolution (in the order of seconds), providing a signal enhancement curve for every voxel of the three dimensional MRI data sets. These time-curves can be described semiquantitatively or modelled into pharmacokinetic parameters, which gives either descriptive measures of the enhancement curve (start of enhancement, wash-in gradient, maximum enhancement, time to peak, washout gradient, area under the gadolinium curve, etc.) or model parameters (forward leakage rate, washout rate constant and leakage space) usual after the fitting to a pharmacokinetic model. 55 For an accurate assessment of the model parameters, an arterial input function (AIF) is required that describes the shape of the contrast bolus arriving at the prostate. The semiquantitative parameters do not need such an AIF. Tumour tissue in the prostate is characterised by increased pharmacokinetic parameters compared with healthy tissue. Unfortunately, especially in the TZ of older men, benign diseases such as proliferative benign prostatic hyperplasia or prostatitis also show marked enhancement after contrast agent administration, making DCE-MRI less specific in the TZ of the prostate. Very recently, recommendations have been published on how this technique can best be used. 56

Dynamic contrast-enhanced MRI has been shown to be of use in detection and staging of PC within a multiparametric protocol57–59 and is especially useful in follow-up after treatment, when normal prostate anatomy is either not present60 or disturbed after radiotherapy. 61

Diffusion-weighted magnetic resonance imaging

Diffusion-weighted MRI is a technique that evaluates the microscopic mobility of water molecules in tissue. Impeded water movements within cellularly dense tissues, such as tumours, appear as high-signal regions on diffusion-weighted images and as darker signals on apparent diffusion coefficient (ADC) maps. In glandular spaces (healthy prostate luminal spaces) or large extracellular spaces, water motion is less impeded, leading to larger signal attenuation (low signal on diffusion-weighted images) and to higher ADC values. In addition to its value in the detection of cancer,62,63 DW-MRI has also been shown to be a promising marker of tumour aggressiveness, with good correlation between ADC values and Gleason score in the PZ of the prostate. 64

Patient-level analysis

Ten studies74,76,79,101,105,106,108,112,115,120 involving 438 patients reported the diagnostic accuracy of MRS and provided sufficient information for inclusion in a meta-analysis. All used a (10- or 12-core) TRUS-guided approach plus additional targeted cores on MRS equivocal or suspicious areas, apart from the study by Wetter et al. ,108 which used a MRI-guided approach. Four studies reported the CC/C ratio used as the cut-off for a positive test result, which ranged from > 0.6108 to > 0.86. 79

Across the studies the median (range) prevalence of PC was 34.5% (9.5% to 48.9%). The number of previous biopsy sessions the participants had undergone ranged from one105 to two to six. 101 Most studies reported that participants had undergone one to three, or one to four, previous biopsy sessions. The number of cores extracted in the previous biopsy session ranged from six76,108 to (a mean of) 16. 106

The studies were judged to have low risk of bias for the patient selection, index test and flow of timing domains. All studies were judged to have a high risk of bias for the reference standard domain owing to a lack of follow-up. All studies were judged to have low applicability concerns for the patient selection, index test and reference standard domains, apart from, for the index test domain, Cirillo et al. 79 (only PZ assessed) and Prando et al. 101 (central gland not assessed).

Figure 5 shows the sensitivity and specificity of the individual studies, pooled estimates and SROC curve. The pooled (95% CI) estimates for sensitivity and specificity were 92% (86% to 95%) and 76% (61% to 87%), respectively.

FIGURE 5.

Magnetic resonance spectroscopy – patient-level analysis: sensitivity, specificity, pooled estimates and SROC curve.

All of the studies reported sensitivity of ≥ 88% apart from Yuen et al. 112 (71%). Yuen et al. 112 suggested that contributory factors to the low sensitivity reported might have been (1) difficulties in ensuring the correspondence of TRUS biopsy spatial accuracies to suspicious areas on MRS and (2) that MRS did not cover the entire PZ of the gland. The studies by Prando et al. 101 and Testa et al. 106 reported low specificity (both 44%). Prando et al. 101 reported results either when a voxel score of 4 or 5, or just 5, was used as a cut-off for a positive test result. The results using the cut-off of 4 or 5 were included in the pooled estimates. However, if the results using a cut-off of just 5 had been used this would have increased the specificity to 84% but reduced the sensitivity from 100% to 70.6%. Testa et al. 106 suggested that the low specificity in their study was probably determined by the lower CC/C ratio used (actual value not reported) compared with cut-offs used by other studies.

A sensitivity analysis comparing pooled estimates of the results of earlier studies (pre 2007) with those of studies published more recently (2007 onwards) found no significant differences between the two subgroups. The pooled (95% CI) estimates for sensitivity and specificity were pre 2007, 93% (80% to 98%) and 71% (43% to 89%); 2007 onwards, 91% (84% to 95%) and 79% (60% to 90%) (see Appendix 9).

Biopsy-level analysis

Six studies76,79,100,101,106,112 reported the diagnostic accuracy of MRS at biopsy or other non-patient-level analysis and provided sufficient information for inclusion in a meta-analysis. Figure 6 shows the sensitivity and specificity of the individual studies, pooled estimates and SROC curve. The units of analyses reported by the studies included biopsy,76 site,79 segment,100 region106 and core. 112 The pooled (95% CI) estimates for sensitivity and specificity were 66% (46% to 82%) and 89% (86% to 92%), respectively.

FIGURE 6.

Magnetic resonance spectroscopy – biopsy-level analysis: sensitivity, specificity, pooled estimates and SROC curve.

Testa et al. 106 also reported region-based analysis separately for the PZ (sensitivity 64.9%, specificity 85.8%) and the TZ (sensitivity 72.2%, specificity 93.2%).

Patient-level analysis

Three studies78,90,105 involving 209 patients reported the diagnostic accuracy of DCE-MRI and provided sufficient information for inclusion in a meta-analysis. All used a (10-core or at least 12-core) TRUS-guided approach plus additional targeted cores from suspicious areas on the imaging test.

Across the studies the median (range) prevalence of PC was 48.9% (24.7% to 53.8%). The number of previous biopsy sessions the participants had undergone ranged from one105 to one to twelve. 90 The number of cores extracted in the previous biopsy session was 10,105 and (a mean of) 12.6,78 although this information was not reported by Lattouf et al. 90 The studies were judged to have low risk of bias and applicability concerns for all domains, apart from the reference standard domain where all three were judged to be at high risk of bias due to a lack of follow-up.

Figure 7 shows the sensitivity and specificity of the individual studies, pooled estimates and SROC curve. The pooled (95% CI) estimates for sensitivity and specificity were 79% (69% to 87%) and 52% (14% to 88%), respectively. Compared with the other two studies, the study by Sciarra et al. 105 reported high specificity (91%). This study actually reported sensitivity of 84.6% and specificity of 82.3%; however, using the actual 2 × 2 data presented in the paper led to a calculation of 79.5% for sensitivity and 91.3% for specificity, and these were the data used in the pooled estimates. However, there was no obvious explanation for the large difference in specificity values between this study and the other two studies.

FIGURE 7.

Dynamic contrast-enhanced MRI – patient-level analysis: sensitivity, specificity, pooled estimates and SROC curve.

Biopsy-level analysis

Four studies78,100,105,133 reported the sensitivity and/or specificity of DCE-MRI at biopsy or other non-patient-level analysis (Table 5). Across these studies the median (range) sensitivity and specificity was 64.0% (29.3% to 80.0%) and 83.5% (76.7% to 93.5%), respectively.

Patient-level analysis

One study, by Park et al. ,96 reported a patient-level analysis for DW-MRI. This study, involving 43 patients, employed an MRI-directed, TRUS-guided approach, with at least two cores from suspicious DW areas followed by a 6-, 8- or 10-core biopsy. The study reported a sensitivity of 100% (specificity not reported).

Biopsy-level analysis

Three studies reported DW-MRI at biopsy or other non-patient-level analysis. 96,100,133 The study by Portalez et al. 100 used a 12- to 34-core TRUS-guided approach plus two biopsies of suspicious MRI areas. In a segment level analysis (n = 408) they reported a sensitivity of 39.0% and specificity of 96.0%. Valentini et al. 133 used a 24-core TRUS/Bx (transperineal) approach plus additional biopsies of suspicious MRI areas. In a biopsy-level analysis (number of biopsies not stated) they reported a sensitivity of 60% (specificity not reported). In the study by Park et al. 96 reporting a core level analysis (number of cores not stated), from the information provided it was possible to calculate PPV (78.9%) but not sensitivity or specificity.

Patient-level analysis

Fifteen studies57,74,76,78,79,84,90,101,106,108,112,128,134,136,137 involving 620 patients reported the diagnostic accuracy of T2-MRI and provided sufficient information for inclusion in a meta-analysis. All used a (mostly 10- or 12-core) TRUS-guided approach plus additional targeted cores on T2-MRI equivocal or suspicious areas, apart from the studies by Franiel et al. 84 and Wetter et al. ,108 which used a MRI-guided approach.

Across the studies the median (range) prevalence of PC was 35.7% (9.5% to 53.8%). The number of previous biopsy sessions the participants had undergone ranged from 1108 to 1–12. 90 Most studies reported that participants had undergone somewhere in the region of between one to six previous biopsy sessions. The number of cores extracted in the previous biopsy session ranged from 4 or 657 to (a mean of) 12–14. 112

The studies were judged to have low risk of bias for the patient selection, index test and flow of timing domains, apart from, for the flow and timing domain, Beyersdorff et al. 57 (not all patients were included in the analysis). 57 All studies were judged to have a high risk of bias for the reference standard domain owing to a lack of follow-up. All studies were judged to have low applicability concerns for the patient selection, index test and reference standard domains, apart from, for the index test domain, Bhatia et al. 76 (both normal and equivocal results were classed as negative), Cirillo et al. 79 (only PZ assessed) and Prando et al. 101 (central gland not assessed).

Figure 8 shows the sensitivity and specificity of the individual studies, pooled estimates and SROC curve. The pooled (95% CI) estimates for sensitivity and specificity were 86% (74% to 93%) and 55% (44% to 66%), respectively.

FIGURE 8.

T2-weighted magnetic resonance imaging – patient-level analysis: sensitivity, specificity, pooled estimates and SROC curve.

Four studies74,78,108,112 reported sensitivity of 60% or lower. There was no obvious explanation for this in the studies by Amsellem-Ouazana et al. 74 or Cheikh et al. 78 In the study by Wetter et al. 108 only six patients had undergone a previously negative biopsy and this study also extracted biopsies transgluteally. 108 Yuen et al. 112 suggested that contributory factors to the low sensitivity reported might have been (1) difficulties in ensuring the correspondence of TRUS biopsy spatial accuracies to suspicious areas on MRS and (2) that MRS did not cover the entire PZ of the gland. Four studies57,84,90,101 reported specificity of 35% or lower. There was no obvious explanation for this in the studies by Franiel et al. 84 and Prando et al. 101 Beyersdorff et al. 57 reported results either when suspicious and inconclusive, or just suspicious, were used as a cut-off for a positive test. The results using the cut-off of suspicious and inconclusive were included in the pooled estimates. However, if the results using a cut-off of just suspicious had been used this would have increased the specificity to 61.5% but reduced the sensitivity from 100% to 83.3%. The study by Lattouf et al. 90 reported sensitivity of 40% and specificity of 69.5%; however, using the actual 2 × 2 data presented in the paper led to a calculation of 92.9% for sensitivity and 16.7% for specificity, and these were the data used in the pooled estimates. 90

A sensitivity analysis comparing pooled estimates of the results of earlier studies (pre 2007) with those of studies published more recently (2007 onwards) found no significant differences between the two subgroups. The pooled (95% CI) estimates for sensitivity and specificity were pre 2007, 83% (63% to 94%) and 56% (39% to 71%) and 2007 onwards, 88% (72 to 95%) and 55% (41 to 69%) (see Appendix 10).

Biopsy-level analysis

Eight studies57,76,78,79,84,100,106,112 reported the diagnostic accuracy of T2-MRI at biopsy or other non-patient-level analysis and provided sufficient information to be included in a meta-analysis. Figure 9 shows the sensitivity and specificity of the individual studies, pooled estimates and SROC curve. The units of analyses reported by the studies included biopsy,57,76 sector,78 site,79 region,84,106 segment100 and core. 112 The pooled (95% CI) estimates for sensitivity and specificity were 54% (42% to 66%) and 87% (75% to 94%), respectively.

FIGURE 9.

T2-weighted magnetic resonance imaging – biopsy-level analysis: sensitivity, specificity, pooled estimates and SROC curve.

Testa et al. 106 also reported region-based analysis separately for the PZ (sensitivity 27.0%, specificity 95.8%) and the TZ (sensitivity 61.1%, specificity 98.9%).

Patient-level analysis

Twenty-one studies57,75,77,80,81,83,88,91–93,96–99,102,103,107,110,111,113,120 involving 8393 patients reported the sensitivity and/or specificity of systematic TRUS-guided biopsies. See Appendix 11 for the individual study results.

Eleven of these studies57,75,80,81,83,92,93,96,103,107,111 included the use of TRUS as an imaging test, of which six,57,75,80,83,93,111 involving 782 patients, provided sufficient information for inclusion in a meta-analysis. The number of cores extracted ranged from 8 or 1280 to 15. 93

Across the six studies57,75,80,83,93,111 the median (range) prevalence of PC was 26.5% (14.4% to 31.6%). The number of previous biopsy sessions the participants had undergone ranged from 180,93 to 1–6. 57 The number of cores extracted in the previous biopsy session ranged from 4 or 657 to 12. 93

Most studies were judged to have low risk of bias for the patient selection, index test and flow of timing domains. The study by Babaian et al. 75 was judged to be of unclear risk of bias for patient selection (did not report whether or not participant sample was consecutive or provide exclusion criteria) and for the index test (did not report whether or not the test was interpreted by an experienced person). The study by Eskicorapci et al. 83 was also judged to be of unclear risk of bias for patient selection (did not report whether or not participant sample was consecutive or provide exclusion criteria). The study by Beyersdorff et al. 57 was judged to be at high risk of bias for the flow and timing domain (not all participants received a reference standard and not all were included in the analysis). All studies were judged to have a high risk of bias for the reference standard domain owing to a lack of follow-up. All studies were judged to have low applicability concerns for the patient selection, index test and reference standard domains, apart from, for the index test domain, Yuen et al. ,111 which was judged to have high applicability concerns (only PZ assessed).

Figure 10 shows the sensitivity and specificity of the individual studies, pooled estimates and SROC curve. The pooled (95% CI) estimates for sensitivity and specificity were 27% (16% to 42%) and 81% (77% to 85%), respectively.

FIGURE 10.

Transrectal ultrasonography – patient-level analysis: sensitivity, specificity, pooled estimates and SROC curve.

Six large-scale population screening studies enrolling 86,749 participants provided information on the performance of systematic biopsies using a TRUS-guided approach on a subset of their populations who had a previously negative biopsy (n = 5771). 81,88,97,99,103,113 The number of cores taken ranged from 4–688,103 to 16–21. 97 It was not possible to calculate specificity because the procedure merely extracted cores for histopathological assessment and therefore there were no positive or negative test results as such. It was possible to calculate sensitivity on the basis that, for participants with a first negative biopsy, cores taken during the second biopsy session and assessed histopathologically as positive were considered true-positive. For those patients negative on the second biopsy, cores taken during subsequent biopsy sessions and assessed histopathologically as positive were considered to be false-negative on the second biopsy session, thereby allowing sensitivity to be calculated. Across these studies the median (range) sensitivity was 72.5% (60.6% to 96.3%). In effect these studies provided an indication of the sensitivity of the reference standard, which is influenced by the method by which tissue samples are obtained. Across all of the 10 non-imaging TRUS/Bx studies,77,88,91,97–99,102,110,113,120 the median (range) sensitivity was 72.5% (59.3% to 96.3%). The number of cores taken ranged from 4–688 to 16–21. 97

Biopsy-level analysis

No studies reported sensitivity or specificity at a biopsy or other non-patient-level analysis. In the study by Lee et al. ,126 from the information provided it was possible to calculate PPV (3.5%) but not sensitivity or specificity. This study did not report the number of cores taken per patient but did report the overall number of cores sampled (n = 903 from 87 patients, average of 10 cores per patient).

Magnetic resonance spectroscopy compared with T2-weighted magnetic resonance imaging

Six studies74,76,79,106,108,112 involving 201 patients reported MRS compared with T2-MRI and provided sufficient information for inclusion in a meta-analysis. All used a (10- or 12-core) TRUS-guided approach plus additional targeted cores on MRS/T2-MRI equivocal or suspicious areas, apart from the study by Wetter et al. ,108 which used a MRI-guided approach. 108 Three studies reported the CC/C ratio used as a cut-off for a positive test result for MRS, which ranged from > 0.6108 to > 0.86. 79

Across the studies the median (range) prevalence of PC was 32.4% (9.5% to 40.7%). The number of previous biopsy sessions the participants had undergone ranged from 1108 to 1–4. 74,106 The number of cores extracted in the previous biopsy session ranged from 676,108 to (a mean of) 12–14. 112 The studies were judged to have low risk of bias for the patient selection, index test and flow of timing domains. All studies were judged to be at high risk of bias for the reference standard domain owing to a lack of follow-up. All studies were judged to have low applicability concerns for patient selection, index test and reference standard domains, apart from, for the index test domain, Bhatia et al. 76 (unclear concern for applicability: both normal and equivocal test results categorised as negative) and Cirillo et al. 79 (high concern for applicability: only PZ assessed).

For the HSROC analysis, we made the assumption that the underlying shape parameter varies with the threshold and accuracy parameters. This is because using the original assumption of a common underlying shape made our models unstable. We provide the results of a sensitivity analysis with the original assumption in Appendix 13.

Figure 11 shows the sensitivity and specificity of the individual studies, pooled estimates and SROC plot with 95% confidence region. The pooled (95% CI) estimates for sensitivity and specificity were 89% (79% to 95%) and 71% (51% to 85%) for MRS, and 77% (55% to 90%) and 68% (59% to 75%) for T2-MRI.

FIGURE 11.

Magnetic resonance spectroscopy compared with T2-MRI patient-level analysis: sensitivity, specificity, pooled estimates and SROC curve.

Magnetic resonance spectroscopy compared with dynamic contrast-enhanced magnetic resonance imaging

Two studies100,105 involving 158 patients reported MRS compared with DCE-MRI (Table 6). Portalez et al. 100 reported segment-level but not patient-level analysis, whereas Sciarra et al. 105 reported both patient- and core-level analysis. 105 In the study by Portalez et al. 100 the sensitivity of the tests was similar but low, whereas specificity was also similar but high. In the study by Sciarra et al. 105 MRS had higher sensitivity than DCE-MRI at both patient- and core-level analysis, with broadly similar specificity, which was higher for patient-level analysis compared with core-level analysis. With regard to the low sensitivity reported by Portalez et al. ,100 the authors stated that in order to visualise the early phase of cancer enhancement and retain spatial resolution, they resorted to the shortest possible time with their MRI unit, which proved to yield adequate specificity but suboptimal sensitivity. 100

Dynamic contrast-enhanced magnetic resonance imaging compared with T2-weighted magnetic resonance imaging

Three studies78,90,100 involving 187 patients compared DCE-MRI with T2-MRI (Table 7). In the two studies reporting patient-level analysis,78,90 DCE-MRI had higher sensitivity and lower specificity than T2-MRI in one, with lower sensitivity and higher specificity in the other. Cheikh et al. 78 reported low specificity for DCE-MRI, whereas Lattouf et al. 90 reported low specificity for both DCE-MRI and T2-MRI. The test combination ‘DCE-MRI or T2-MRI’ resulted in similar or increased sensitivity compared with the individual tests but reduced specificity, whereas the combination ‘DCE-MRI and T2-MRI’ reduced sensitivity with a moderate increase in specificity. In the two studies reporting non-patient-level analysis, DCE-MRI had higher sensitivity and slightly lower specificity than T2-MRI in one and lower sensitivity and slightly higher specificity in the other.

Dynamic contrast-enhanced magnetic resonance imaging compared with diffusion-weighted magnetic resonance imaging

Two studies100,133 involving 79 patients compared DCE-MRI with DW-MRI (Table 8). Both reported non-patient-level analysis. DCE-MRI had higher sensitivity than DW-MRI in one study133 and lower sensitivity100 in the other, whereas the sensitivity reported for both DCE-MRI and DW-MRI was much higher in the study by Valentini et al. 133 than it was in the study by Portalez et al. 100 Portalez et al. 100 reported similarly high specificity for both tests.

Magnetic resonance spectroscopy or T2-weighted magnetic resonance imaging

Patient-level analysis

Eight studies74,79,84,106,108,112,118,130 involving 316 patients reported the diagnostic accuracy of MRS or T2-MRI and provided sufficient information for inclusion in a meta-analysis. All used a (mostly 10- or 12-core) TRUS-guided approach plus additional targeted cores on MRS/T2-MRI equivocal or suspicious areas, apart from Franiel et al. 84 and Wetter et al. ,108 who used a MRI-guided approach. In the study by Destefanis et al. 118 all participants (n = 26) had ASAP. Four studies reported the CC/C ratio used as the cut-off for a positive test result, which ranged from > 0.6108 to > 0.86. 79,118

Across the studies the median (range) prevalence was 35.2% (29.2% to 40.7%). The number of previous biopsy sessions the participants had undergone ranged from 1108,118 to 1–6. 84 The number of cores extracted in the previous biopsy session ranged from 6108 to (a mean of) 16. 106

The studies were judged to have low risk of bias for the patient selection, index test and flow of timing domains. All studies were judged to be at high risk of bias for the reference standard domain owing to a lack of follow-up. All studies were judged to have low applicability concerns for the patient selection, index test and reference standard domains, apart from, for the index test domain, Cirillo et al. 79 (only PZ assessed).

Figure 12 shows the sensitivity and specificity of the individual studies, pooled estimates and SROC curve. The pooled (95% CI) estimates for sensitivity and specificity were 96% (90% to 98%) and 31% (21% to 42%), respectively.

FIGURE 12.

Magnetic resonance spectroscopy or T2-MRI patient-level analysis: sensitivity, specificity, pooled estimates and SROC curve.

Biopsy-level analysis

Three studies79,84,106 reported MRS or T2-MRI and provided sufficient information for inclusion in a meta-analysis. The units of analyses reported by the studies included site79 and region. 84,106 Figure 13 shows the sensitivity and specificity of the individual studies, pooled estimates and SROC curve. The pooled (95% CI) estimates for sensitivity and specificity were 79% (71% to 86%) and 74% (45% to 90%), respectively. Testa et al. 106 also reported region-based analysis separately for the peripheral and TZs. For the PZ (540 regions analysed), sensitivity was 70.3% and specificity 83.3%, whereas for the TZ (108 regions analysed) sensitivity was 72.2% and specificity 92.2%.

FIGURE 13.

Magnetic resonance spectroscopy or T2-MRI biopsy-level analysis: sensitivity, specificity, pooled estimates and SROC curve.

Magnetic resonance spectroscopy and T2-weighted magnetic resonance imaging

Patient-level analysis

Five studies76,106,108,112,134 involving 129 patients reported the diagnostic accuracy of MRS and T2-MRI and provided sufficient information for inclusion in a meta-analysis. All used a (mostly 10- or 12-core) TRUS-guided approach plus additional targeted cores on MRS/T2-MRI equivocal or suspicious areas, apart from Wetter et al. ,108 which used a MRI-guided approach and extracted cores transgluteally. None of the studies reported the CC/C ratio value used as the cut-off for a positive test result.

Across the studies the median (range) prevalence was 33.3% (9.5% to 41.7%). The number of previous biopsy sessions the participants had undergone ranged from 1108 to 1–4. 106 The number of cores extracted in the previous biopsy session ranged from 676,108 to (a mean of) 16. 106

The studies were judged to have low risk of bias for the patient selection, index test and flow of timing domains. All studies were judged to be at high risk of bias for the reference standard domain owing to a lack of follow-up. All studies were judged to have low applicability concerns for the patient selection, index test and reference standard domains, apart from, for the index test domain, Bhatia et al. 76 (normal and equivocal tests were categorised as negative for malignancy).

Figure 14 shows the sensitivity and specificity of the individual studies and pooled estimates (this analysis required to be undertaken without random effect parameters as otherwise the model would not converge and consequently it was not possible to produce a ROC curve). The pooled (95% CI) estimates for sensitivity and specificity were 60% (46% to 75%) and 74% (65% to 84%), respectively.

FIGURE 14.

Magnetic resonance spectroscopy and T2-MRI patient-level analysis: sensitivity, specificity and pooled estimates.

Magnetic resonance spectroscopy or dynamic contrast-enhanced magnetic resonance imaging

Magnetic resonance spectroscopy and dynamic contrast-enhanced magnetic resonance imaging

Other combinations involving magnetic resonance spectroscopy

Franiel et al. 84 reported other combinations of tests involving MRS, both at patient-level (n = 54) and region-level (n = 178) analysis (Table 9).

In the study by Roethke et al. 104 reporting MRS or T2-MRI or DCE-MRI or DW-MRI (n = 100), from the information provided it was possible to calculate PPV (52%), but not sensitivity or specificity.

Dynamic contrast-enhanced magnetic resonance imaging or T2-weighted magnetic resonance imaging

Patient-level analysis

Three studies78,84,90 involving 173 patients reported the diagnostic accuracy of DCE-MRI or T2-MRI and provided sufficient information for inclusion in a meta-analysis. The studies by Cheikh et al. 78 and Lattouf et al. 90 used a 12-core TRUS-guided approach plus additional targeted cores from suspicious areas on the imaging tests. 78,90 The study by Franiel et al. used a MRI-guided approach. 84

Across the studies the median (range) prevalence of PC was 38.9% (24.7% to 53.8%). The number of previous biopsy sessions the participants had undergone ranged from 1–578 to 1–12. 90 Only Cheikh et al. 78 reported the number of cores extracted in the previous biopsy session (mean of 12.6). The studies were judged to have low risk of bias and applicability concerns for all domains apart from the reference standard domain, for which all three were judged to be at high risk of bias owing to a lack of follow-up.

Figure 15 shows the sensitivity and specificity of the individual studies, pooled estimates and SROC curve. The pooled (95% CI) estimates for sensitivity and specificity were 88% (80% to 96%) and 14% (8% to 20%), respectively.

FIGURE 15.

Dynamic contrast-enhanced MRI or T2-MRI patient-level analysis: sensitivity, specificity, pooled estimates and SROC curve.

Dynamic contrast-enhanced magnetic resonance imaging and T2-weighted magnetic resonance imaging

Dynamic contrast-enhanced magnetic resonance imaging or diffusion-weighted magnetic resonance imaging

Dynamic contrast-enhanced magnetic resonance imaging or diffusion-weighted magnetic resonance imaging or T2-weighted magnetic resonance imaging

Dynamic contrast-enhanced magnetic resonance imaging and diffusion-weighted magnetic resonance imaging and T2-weighted magnetic resonance imaging

Diffusion-weighted magnetic resonance imaging or T2-weighted magnetic resonance imaging

Patient-level analysis

For the patient-level estimates of tests with three or more studies, comparing T2-MRI against all the other tests showed statistically significant differences (p < 0.001). Compared with DCE-MRI, T2-MRI was observed to have lower sensitivity and significantly higher specificity. Sensitivity (95% CI) for DCE-MRI was 87% (74% to 94%) compared with 83% (75% to 89%) for T2-MRI (p = 0.499). Specificity (95% CI) for DCE was 40% (25% to 56%) compared with 57% (47% to 67%) for T2-MRI (p = 0.041).

Magnetic resonance spectroscopy was observed to have higher sensitivity and specificity than T2-MRI. Sensitivity for MRS was 93% (87% to 97%) compared with 83% (75% to 89%) for T2-MRI (p = 0.008). Specificity for MRS was 64% (52% to 75%) compared with 57% (47% to 67%) for T2-MRI (p = 0.194).

When T2-MRI was used in combination with MRS (‘T2-MRI and MRS’, both tests had to be suspicious for the combination to be considered positive) T2-MRI used alone was observed to have higher sensitivity but significantly lower specificity. Sensitivity for ‘T2-MRI and MRS’ was 71% (50% to 85%) compared with 83% (75% to 89%) for T2-MRI (p = 0.172). Specificity for ‘T2-MRI and MRS’ was 73% (58% to 85%) compared with 57% (47% to 67%) for T2-MRI (p = 0.011).

When T2-MRI was used in combination with DCE-MRI (‘T2-MRI or DCE-MRI’, if either test is suspicious the combination is considered positive), T2-MRI used alone was observed to have lower sensitivity but significantly higher specificity. Sensitivity for ‘T2-MRI or DCE-MRI’ was 92% (81% to 97%) compared with 83% (75% to 89%) for T2-MRI (p = 0.046). Specificity for ‘T2-MRI or DCE-MRI’ was 24% (13% to 39%) compared with 57% (47% to 67%) for T2-MRI (p < 0.001).

When T2-MRI was used in combination with MRS (‘T2-MRI or MRS’, if either test is suspicious the combination is considered positive), this combination had significantly higher sensitivity than T2-MRI alone but significantly lower specificity. Sensitivity for ‘T2-MRI or MRS’ was 97% (91% to 99%) compared with 83% (75% to 89%) for T2-MRI (p = 0.001). Specificity for ‘T2-MRI or MRS’ was 34% (23% to 46%) compared with 57% (47% to 67%) for T2-MRI (p < 0.001).

Compared with TRUS used as an imaging test, T2-MRI was observed to have significantly higher sensitivity but significantly lower specificity. Sensitivity for TRUS was 24% (13% to 39%) compared with 83% (75% to 89%) for T2-MRI (p < 0.001). Specificity for TRUS was 88% (79% to 94%) compared with 57% (47% to 67%) for T2-MRI (p < 0.001).

These differences are based on between-study comparisons, so may have been due to differences between the studies rather than true differences between the tests.

For the estimates comparing T2-MRI with other tests, in terms of relative sensitivity, the direction of effect favoured (1) MRS; (2) ‘T2-MRI or DCE’; and (3) ‘T2-MRI or MRS’ over T2-MRI, while favouring T2-MRI over (1) DCE-MRI; (2) ‘T2-MRI and MRS’; and (3) TRUS, although the only results that were statistically significant were for ‘T2-MRI or MRS’ compared with T2-MRI (‘T2-MRI or MRS’ better) and T2-MRI compared with TRUS (T2-MRI better). See Appendix 13.1 for further details.

In terms of relative specificity the direction of effect favoured (1) MRS; (2) ‘T2-MRI and MRS’; and (3) TRUS over T2-MRI, while favouring T2-MRI over (1) DCE-MRI; (2) ‘T2-MRI or DCE-MRI’; and (3) ‘T2-MRI or MRS’, although the only results that were statistically significant were for ‘T2-MRI and MRS’ compared with T2-MRI (‘T2-MRI and MRS’ better), T2-MRI compared with TRUS (TRUS better), ‘T2-MRI or DCE-MRI’ compared with T2-MRI (T2-MRI better) and ‘T2-MRI or MRS’ compared with T2-MRI (T2-MRI better). See Appendix 13.1 for further details.

Biopsy-level analysis

The highest sensitivity (95% CI) was for the combination ‘T2-MRI or MRS’ at 75% (61% to 86%), whereas the highest specificity was for T2-MRI at 87% (78% to 93%), with MRS also reporting a similarly high specificity at 84% (72% to 91%). See Appendix 13.2 for further details.

For the estimates comparing T2-MRI with other tests, in terms of relative sensitivity both (1) MRS and (2) ‘T2-MRI or MRS’ had statistically significantly higher sensitivity than T2-MRI, whereas for specificity the direction of effect favoured T2-MRI over both (1) MRS and (2) ‘T2-MRI or MRS’, although only the comparison with ‘T2-MRI or MRS’ was statistically significant. See Appendix 13.2 for further details.

Relevant patient population

The modelled cohorts consisted of men with suspected PC with a prior negative/inconclusive biopsy, with indications for repeat biopsy (i.e. sustained suspicion of PC as a result of clinical and/or pathological findings). 40 We carried out several analyses applying cancer prevalence rates consistent with those observed in the literature for different subgroups defined by factors that influence disease prevalence at repeat biopsy. The base-case analysis was carried out using a prevalence of 24%, which is consistent with cancer detection rate (with 24-core saturation biopsy) reported for a cohort of patients with a previous benign biopsy result but persistently elevated PSA (> 4 ng/ml) and/or abnormal DRE. 38 Further analyses were carried out, with the prevalence of underlying cancer set at a higher level consistent with that reported for patients with ASAP or percentage free to total PSA level of < 10% (i.e. 50%). Further, we also set the cancer prevalence at the lower level of 10% to represent a lower risk cohort selected for repeat biopsy.

Analyses were conducted separately for men aged 60 years and men aged 70 years at time of repeat biopsy, as age influences the cost-effectiveness of diagnosing and treating PC. It was assumed that a PC diagnosis would not be aggressively pursued in men aged 75 years and over. Men with cancer were initially spread across the undiagnosed cancer states in the model (Table 16), based on the reported Gleason scores of tumours detected during the studies included in the systematic review, and other available data on the clinical and/or pathological stages/grades of cancers detected at second biopsy. 2,37,38 Although the frequency of higher grade cancer may increase with age and underlying prevalence, data limitations precluded adjustment of the proportions by selection criteria for repeat biopsy. As such, results of subgroup analyses (by age and prevalence) should be treated with caution. If a higher proportion of older men have more advanced or higher risk tumours, this would serve to improve the cost-effectiveness of more sensitive strategies in comparison with the base-case estimates we provide for 70-year-old men.

Diagnostic strategies to be evaluated

The experimental strategies chosen for evaluation were selected based on the availability of data from the systematic review of diagnostic accuracy (Table 17). It was not possible to obtain comparable pooled sensitivity/specificity estimates for all sequences of clinical interest. Further, the majority of studies included in the diagnostic accuracy review assessed the accuracy of MRI sequences for directing TRUS-guided biopsies, rather than for directly guiding the biopsy. Thus, the economic analysis focused on evaluating the use of MRS/MRI in this context, i.e. using it to identify areas of the prostate for targeting in a subsequent TRUS/Bx. As the patient-level sensitivities obtained from the diagnostic accuracy meta-analysis reflect detection rates achieved when both targeted cores and a number of systematic cores (8 to 12) are taken from patients with positive findings on MRS/MRI, we modelled targeted biopsies to proceed in this same manner in the economic model. Insufficient data were available to ascertain how patient-level sensitivity would be affected if only targeted cores were obtained from patients positive on imaging. As a consequence, we assumed that MRI/imaging prior to biopsy would not alter the cost of the biopsy procedure in the base-case analysis. We also assumed that patients with no visible pathology on MRS/MRI would not proceed to biopsy. The model was specified to simulate the use of MRS/MRI in the index repeat biopsy only. It was assumed that any patients missed by the index repeat biopsy (false-negatives), would have persistently elevated PSA level, which would trigger the offer of a saturation biopsy (> 24 cores) 12 months later, and that acceptance would be high (assumption based on attitudes to repeat biopsy reported by Rosario et al. 150). These conservative assumptions, which favour less-sensitive cancer detection strategies, were subjected to sensitivity analysis.

A systematic TRUS-guided extended-cores (14–16 cores) biopsy for all, carried out in an outpatient setting, was selected as the base comparator against which to assess the cost-effectiveness of using MRS/MRI (see Table 17). A limitation of the available literature is that no existing studies have directly assessed the relative sensitivity/specificity of MRI-directed biopsies in comparison with systematic biopsy sampling schemes with different numbers of cores. Thus, in modelling the comparison we were forced to rely on diagnostic accuracy data for the comparator and index tests derived from different sources using different reference standards. The sensitivity of the systematic extended-cores biopsy was derived from a study assessing the proportion of cancers detected by systematic biopsy schemes with variable numbers of cores,38 using the results of a TRUS-guided saturation biopsy as the reference standard. The MRS/MRI sensitivities/specificities were derived from the systematic review (see Chapter 4), where the reference standard was histopathological assessment of biopsied tissue but the number of cores taken varied from study to study. Sensitivity analysis was performed to assess the sensitivity of findings to variation in these parameters.

How care pathways were determined and modelled, including an illustration of the model

The diagnostic and care pathways were determined based on a review of guidelines, expert opinion, and the availability of data. A schematic review of the diagnostic pathways was provided in Chapter 1 (see Figure 2). Figure 16 shows the tree structure used to model the index repeat biopsy within the economic model.

FIGURE 16.

Diagnostic pathways for index re-biopsy. The base-case analysis assumed that all patients negative on MRS/MRI would not proceed to biopsy, but we also assessed the impact of assuming these patients would proceed to an extended-cores TRUS/Bx. FN, false-negative; FP, false-positive; sens, sensitivity; spec, specificity; TN, true-negative; TP, true-positive.

The diagnostic pathways were embedded in a Markov model developed to simulate the progression of diagnosed (treated) and undiagnosed PC (Figure 17). Seven basic states were used to model the natural history of PC: (1) no or undetectable cancer; (2) localised (T1–T2) PC (low risk); (3) localised PC (intermediate risk); (4) localised PC (high risk); (5) locally advanced cancer (T3); (6) metastatic cancer; and (7) PC death. Patients with localised and locally advanced disease were modelled to progress towards metastatic disease based on age, tumour risk status, and whether or not their cancer was diagnosed and appropriately treated. To begin with, patients with suspected PC following a first negative biopsy were spread across the undiagnosed states (using the proportions in Table 16). In the first cycle of the model, all patients were modelled to undergo their repeat biopsy, either by standard means or directed by one of the MRS/MRI sequences. Patients with underlying cancer (undetected) identified by the second biopsy as having disease, as determined by the sensitivity of the biopsy procedure, were modelled to transit to the appropriate diagnosed cancer state for the subsequent model cycle. Those with undetected cancer missed by the second negative biopsy remained in the appropriate undiagnosed state. Those remaining undiagnosed faced a higher risk of progression to metastases (based on progression rates observed for patients under watchful waiting), whereas those detected were modelled to progress at rates observed for patients receiving radical treatments. The model was cycled on a 3-monthly basis, such that probabilities of progression and costs of treatment and monitoring were expressed in terms of this constant cycle length.

FIGURE 17.

Model structure. a, Patients with localised cancer were risk stratified by cancer grade (low, intermediate and high) and modelled to progress to metastatic disease at different rates. Prop recurrence, proportion with local recurrence following treatment.

Patients remaining in an undiagnosed cancer state after the index repeat biopsy were modelled to have their PSA levels monitored on a 6-monthly basis. An assumption was made that these patients would have persistently elevated PSA level and would therefore be selected for a further biopsy 12 months later. It was assumed that a saturation biopsy (≥ 24 cores) would be offered at this stage, that there would be 90% uptake,150 and that the biopsy would have 98% sensitivity for detecting the remaining undiagnosed cancers (based on clinical opinion within the team). For patients without underlying PC, it was assumed that no further biopsies would be indicated unless incident PCs developed. These assumptions were subjected to sensitivity analysis.

For every biopsy undertaken, modelled patients also faced an associated risk of complications (bleeding, infection, urinary retention). Patients crossing to the diagnosed states were modelled to receive appropriate staging, treatment and monitoring. In addition, patients receiving treatment faced a risk of experiencing complications, which incurred further health service costs and quality-of-life decrements. Following treatment for localised disease, a proportion of the cohort was modelled to experience tumour recurrence, triggering further treatment and costs.

Costs associated with biopsy procedures, PSA monitoring, staging, treatment and disease monitoring were incorporated into the model based on the application of unit costs to procedures and treatment protocols (derived from expert opinion and current guidelines). Utilities associated with the different cancer states were used to quality adjust the time spent by patients in each state, and utility decrements associated with complications arising from treatment were also applied. Thus the model enabled cumulative costs, LYs and QALYs to be tracked over the lifetime of modelled cohorts under alternative diagnostic strategies. The model captures the potential trade-offs between increased short-term costs associated with incorporating MRI sequencing into the care pathways and any cost savings and potential survival gains resulting from fewer repeat biopsies and earlier cancer treatment. The model also accounts for the fact that treatment may have a detrimental impact on patients’ health-related QoL.

Complications of biopsy

The occurrence of biopsy complications was modelled on the basis of two data sources: the ProtecT trial150 and a cohort study reported by Nam et al. 151 The resultant probabilities are provided in Table 18. Costs associated with these complication events were estimated and incorporated into the model.

Risk of cancer progression (undiagnosed and diagnosed patients)

A simplifying assumption of the model was that all men in a cancer state are at risk of disease progression, and that men progress towards metastatic disease. It was also assumed that all cancer-related deaths occur following transition to metastatic disease. Given a lack of comparable data on the rate of transition from localised to locally advanced disease, and from locally advanced to metastatic disease (and the relative effect of diagnosis and treatment on these transitions), the model structure was simplified such that progression from localised disease to metastases was modelled in a single step (using a Weibull function fitted to observed published data for this transition).

Men were initially spread across the ‘no cancer’, ‘localised cancer’ and ‘locally advanced cancer’ states (see Table 16). They were then modelled to progress according to their cancer and diagnostic status using observed follow-up data on the cumulative incidence of metastatic disease combined with estimates of relative treatment effects (i.e. baseline transition risks were adjusted downwards to reflect the impact of appropriate treatment in those receiving a diagnosis). The progression risk for localised cancer was modelled based on data reported by Bill-Axelson et al. 148 whereas the progression risk for men starting in the locally advanced state was modelled based on data from a European Organisation for Research and Treatment of Cancer (EORTC) study reported by Bolla et al. 152

Regression methods153 were used to fit Weibull functions to the observed metastases-free survival probabilities reported for men receiving watchful waiting over a 15-year follow-up period;148 separate functions were fitted for men of < 65 years of age and men aged ≥ 65 years. The estimated parameters of the Weibull functions were then used to derive 3-monthly transition probabilities for the risk of developing metastatic disease from undiagnosed localised cancer. In order to risk-stratify the probabilities of progression, separate functions were determined for patients with low-, moderate- and high-risk localised cancer. This was achieved by adjusting the rate parameters of the Weibull functions to yield the cumulative incidence of metastases or PC mortality observed for cohorts with low-148 and high-risk154 localised cancer. The cumulative incidence rates of metastatic disease reported for the two age-specific cohorts (< 65/≥ 65) as a whole by Bill-Axelson et al. were taken to represent the risk of progression for moderate risk patients in each respective modelled age group. Transition probabilities for developing metastatic disease following diagnosis and treatment were estimated by multiplying the rate parameters of the Weibull functions by published relative risk estimates associated with radical prostatectomy. 148 The resultant modelled cumulative incidence of metastases in treated and untreated patients is shown in Figure 18 compared with the observed values derived from published sources. As a sensitivity analysis, we calibrated the model transition rates to yield the PC-specific survival probabilities (by risk status) observed for patients (> 12 years of follow-up) in the control group of a recently published randomised controlled trial of radical prostatectomy compared with observation for localised disease. 149 In addition, we applied the corresponding relative risk estimates obtained from this trial.

FIGURE 18.

Modelled and observed cumulative incidence of metastases. WW, watchful waiting; RP, radical prostatectomy.

For those starting the model in the locally advanced cancer stage, a similar approach as above was used to model the risk of progression to metastatic disease. However, given the lack of contemporary data on the risk of developing metastases from untreated locally advanced disease, we applied the metastases-free survival data reported for a cohort of patients treated with EBRT alone. These rates were then adjusted downwards for diagnosed patients using the relative risk reduction associated with EBRT combined with adjuvant hormone therapy. 152 All of the relative risk parameters applied in the model are presented in Table 19.

For those with metastatic disease, a constant 3-monthly risk of death from PC was estimated from English observational data155 and applied in the model. The age-specific risk of death from other causes was also incorporated based on age- and sex-specific interim UK life tables. 156

Standard transrectal ultrasound-guided biopsy

The cost of a TRUS-guided needle biopsy was taken from the NHS reference costs41 using the appropriate HRG (LB27Z). There is some uncertainty as to how hospitals in England and Wales are, or would be, reimbursed for repeat biopsies using the systematic extended-cores or MRI-/MRS-directed approach. Although both approaches can be carried out as outpatient procedures without general anaesthetic, it is likely that at least some organisations commission these as day-case procedures. As a result of outpatient procedure coding being non-mandatory, it is not possible to accurately ascertain the proportion of procedures carried out in each care setting, and this is likely to vary from trust to trust. Thus, adopting a conservative approach in favour of less sensitive less costly diagnostic strategies, we initially assumed that all index repeat biopsies would be carried as outpatient procedures, incurring an average cost of £212.

Note, however, that although this tariff-based cost should reflect the budget impact on NHS primary care trusts of commissioning such procedures, it might not fully capture the opportunity cost that hospitals face in delivering the procedure, particularly for extended-cores biopsies that can substantially increase pathology time over standard TRUS/Bx (10–12 cores). We therefore assessed the impact of increasing this cost through sensitivity analysis.

In the base case we also made the conservative assumption that the use of MRS/MRI would not influence the cost of the biopsy procedure itself. This was due to a lack of certainty as to how the patient-level pooled sensitivity estimates obtained for MRS/MRI imaging (from the systematic review) would be affected if only targeted cores were taken in the subsequent biopsy. However, we explored the impact of increasing the cost of extended-cores biopsies, but not the cost of MRI-/MRS-targeted biopsies.

For patients with underlying cancer missed by the index re-biopsy, we assumed that a saturation biopsy (≥ 24 cores) would be indicated at 12 months, and applied the day-case NHS reference cost for all these procedures (£447). We also explored the impact of increasing this cost to reflect potential underestimation of histopathology costs associated with obtaining larger numbers of cores (for further deterministic scenario analyses, see Results, below).

Magnetic resonance imaging sequences for guiding biopsy

The costs of performing alternative MRI sequences to guide prostate biopsy were estimated using a bottom-up approach. Radiographer and radiologist time associated with the performance of different sequences was estimated by asking all of the radiologists involved in the project to provide estimates of time inputs they deemed to be representative of standard practice. Within these estimates, allowance was made for preparation (getting the patient into the machine) and scanning time (two radiographers) and reading/reporting time (one consultant radiologist). The average reported time inputs for sequences included in the economic model are outlined in Table 20. Unit costs obtained from the Unit Costs of Health and Social Care157 were applied to these resource-use inputs. These unit costs included salaries, on-costs (employer superannuation and national insurance contributions) and an apportionment of capital space and overhead costs to capture the opportunity cost of space and overheads attributable to the alternative procedures. Capital equipment required for the alternative MRI sequences was costed by applying current market prices obtained from NHS Grampian. These initial outlay costs were annuitised over the useful working lifespan of the piece of equipment in question, applying an annual discount rate of 3.5% to account for the opportunity cost of the investment over time. The equivalent annual cost of each piece of equipment was divided through by its estimated running time to give a cost per minute estimate. The scanning time estimates associated with alternative MRI sequences were then multiplied by the appropriate equipment cost per minute estimates to give estimates of the capital equipment costs attributable to each different MRI sequence. Costs of equipment used only for DCE-MRI (pump) or MRS (MRS software) were only allocated to sequences involving these procedures. The annual equivalent costs of these items were divided through by the number of uses per year (Dr Lutfi Kurban, Aberdeen Royal Infirmary, March 2012, personal communication; Dr Anwar Padhani, Mount Vernon Cancer Centre, April 2012, personal communication) to give cost per use estimates, which were then applied to sequences incorporating these procedures. Finally, consumables associated with DCE-MRI (contrast, pump pack, others) were costed using unit prices provide by NHS Grampian.

As the MRI costs represent the opportunity costs to hospitals of providing alternative scan sequences, they are well suited to assessing the relative cost-effectiveness of using alternative sequences. However, the estimated costs for some of the simpler scans may underestimate the costs of commissioning such activity. This makes them somewhat less comparable with the tariff-based cost estimate for TRUS/Bx. However, we did not adjust these costs further in the base-case analysis given the concurrent conservative approach to costing TRUS/Bx. As a sensitivity analysis we adjusted the costs of sequences by setting the cost of T2 + DCE-MRI equal to the NHS reference cost for HRG RA03Z (Magnetic Resonance Imaging Scan, one area, pre and post contrast) (£229)41 and maintained the incremental differences in cost between sequences as estimated from the bottom-up calculations.

Biopsy complication costs

Standard practice for repeat biopsy in the UK is systematic TRUS-guided extended-cores or saturation biopsy. The incidence of adverse events post biopsy was determined from the literature150,151 and categorised into hospital admissions or biopsy-related consultations (see Table 18). A risk of death from biopsy complications was experienced only by patients who developed an infection (p = 0.0009) and all other patients were assumed to recover after initial treatment.

Hospital admissions resulting from biopsy complications were reported by Nam151 and Rosario150 as being due to one of three urological diagnoses: urinary infection; urinary bleeding (haematuria); or urinary obstruction (Table 21). For inpatient admissions due to urinary tract infection (UTI) we applied the NHS reference cost for HRG LA04G (Kidney or Urinary Tract Infections with length of stay 1 day or less) (£401). Admission for haematuria was assumed to require insertion of a haematuria catheter for bladder irrigation HRG LB18Z (Attention to Suprapubic Bladder Catheter) at a cost of £567 per patient. 41 Urinary retention was assumed to be temporary and was modelled to incur the cost of inserting and subsequently removing a urethral catheter: day-case HRGs LB09Z (Ureter Intermediate Endoscopic Procedures) and LB15E (Bladder Minor Procedure 19 years and over) at £652 and £368, respectively. 41 It was further assumed that the NHS would incur the daily cost of an overnight catheter bag and the weekly cost of a leg bag (apart from in the first week when two leg bags would be required) over the course of 1 month (£6.47 and £12.61, respectively).

Rosario et al. 150 also reported the 35-day incidence of consultations with general practitioners, urology department nurses, and ‘other sources of medical advice’ (e.g. NHS Direct). The cost associated with a general practitioner (GP) consultation was derived from the Unit Costs of Health and Social Care. 157 The average duration of a GP consultation is 11.7 minutes157 at a cost of £3.10 per surgery minute,157 giving a unit cost of £36.27 per consultation. The cost of a consultation with a urology department nurse was derived from the relevant NHS tariff – non-consultant-led follow-up attendance, non-admitted, face to face – at cost of £70. 41 The cost per NHS direct contact was derived from the NHS Direct National Health Service Trust Annual Report and Accounts 2009–10, and was based on the total reported staff wages divided by the number of calls logged, giving a cost of £20.98 per call.

Prostate cancer treatment costs for localised disease

Potential treatment pathways by cancer stage were derived from the current NICE guidance. 9 The costs associated with implementing alternative treatment pathways, on an ongoing 3-monthly basis, were estimated using data from a variety of sources including the Department of Health NHS reference costs,41 the Unit Costs of Health and Social Care,157 and a recently completed technology assessment report (TAR) evaluating the cost-effectiveness of robotic radical prostatectomy compared with laparoscopic prostatectomy144 for localised PC. Clinical opinion was relied upon to enable an appropriate estimation of timelines for treatment pathways.

It is typical practice in the UK to monitor PSA level for the duration of the patient's life post treatment; every 3 months for the first 2 years and then every 6 months thereafter (based on clinical opinion within the research team). The cost of PSA level monitoring was thus estimated, based on a consultation with a practice nurse (£12)157 plus £5.91 for laboratory services,144 and included in the model (Table 22).

Patients with localised PC were modelled to follow one of three alternative treatment pathways: (1) active surveillance; (2) radical prostatectomy followed by PSA level monitoring; or (3) EBRT followed by PSA level monitoring. The proportion of patients receiving each management strategy, by D'Amico Risk category,31 was derived from routine Scottish health episode data (Dr Karina Laing, MSc in Surgical Sciences thesis, University of Edinburgh, May 2012, personal communication). It was assumed that the alternative treatment modalities would be applied appropriately based on the risk of progression and that, as such, the observed risk reduction associated with radical prostatectomy 148 could be achieved at the level of the cohort as a whole.

The cost of active surveillance was estimated based on the cost of PSA testing (see Table 22) on a 3-monthly basis, followed by a repeat TRUS/Bx41 at 12 months, and every 3 years thereafter (based on clinical opinion within the research team).

The cost of radical prostatectomy was taken as the NHS reference cost for HRG LB21Z (Bladder Neck Open Procedures – Male). Of the two most common approaches to radical surgery (open and laparoscopic radical prostatectomy) the overall activity reported for open procedures was higher,41 and, as such, the cost for this procedure was applied in the model.

The cost associated with a programme of EBRT was calculated on the basis of 37 sessions within a 7.5-week time frame (expert opinion) at a cost per session of £129. 41 EBRT treatment is generally accompanied by a course of androgen deprivation therapy. Although all patients with localised PC were modelled to receive 3 months of hormone therapy from commencement of EBRT, hormone therapy prior to EBRT treatment was assumed to occur only for those with intermediate- or high-risk disease. Before commencing EBRT, these patients were initially modelled to receive a 21-day course of bicalutamide (Casodex^®, AstraZeneca: £96.00), followed by a 3-month course of the LHRH agonist triptorelin (Decapeptyl^® SR, Ipsen: 11.25-mg 3-month injection) at a cost of £207. 158 As localised low-risk patients do not generally receive hormone therapy prior to EBRT treatment, it was assumed that the costs of hormone treatment for these patients would be incurred in the first 3-month cycle following diagnosis, concurrently with the EBRT sessions. It was assumed in all cases that triptorelin would be administered by a practice nurse in a primary care setting, at a cost of £12 per visit.

Treatment costs associated with locally advanced disease

External beam radiotherapy with adjuvant hormone therapy was identified as the most appropriate treatment option for patients with locally advanced PC upon diagnosis. 9 A small proportion of men were also modelled to receive radical prostatectomy. The cost streams and timelines for these treatments were assumed to be consistent with those outlined above for patients with moderate- to high-risk localised disease, with the exception that hormone therapy was continued to 2 years post EBRT.

Costs associated with local progression following treatment for localised disease

A proportion of the cohort was modelled to experience biochemical recurrence following radical treatment for localised cancer. These patients were modelled to receive either salvage EBRT or hormone therapy alone.

Salvage EBRT is delivered at lower gray, with fewer sessions (33 sessions within a 6.5-week time frame). As such, we applied the NHS reference cost (£107) for the appropriate HRG (SC22Z) to each treatment session. 41 In addition, hormone treatment for patients receiving salvage EBRT for biochemical recurrence was extended for a period of 2 years post EBRT treatment.

The 3-monthly cost of hormone therapy was assumed to correspond to the cost of hormone therapy administered pre and post EBRT (21-day course of bicalutamide, followed by 3-monthly injections of triptorelin). However, treatment was assumed to extend for the duration of the patient's lifetime when initiated for biochemical relapse.

Costs associated with metastatic disease

Upon transiting to the metastatic disease state, it was initially assumed that all patients would be treated with hormone therapy, incurring a continuous 3-monthly cost of £219 (Table 23). Without explicitly modelling the initiation and impact of chemotherapy, we also assumed that 50% of patients developing metastatic disease would undergo a first-line docetaxel-based chemotherapy regimen (£10,450) and that 70% of these patients would go on to receive a second-line abiraterone-based regimen (£24,670) prior to death, as per the assumptions used in the costing template for the NICE abiraterone technical appraisal. 39

Costs of complications arising from treatment

Health measurement and valuation

Discount rate (costs and benefits)

Costs and benefits (LYs and QALYs) were discounted at the treasury recommended rate of 3.5% per annum. 167 We also assessed the impact of discounting benefits at the rate of 1.5% per annum, while maintaining a discount rate of 3.5% for costs, as suggested by NICE in instances where treatment effects ‘… are both substantial in restoring health and sustained over a very long period (normally at least 30 years)’. 167

Internal validation

To assess the internal validity of the model, Figures 19 and 20 show the Markov traces for treated and untreated patients with localised cancer (men aged 60 years) over a 15-year follow-up period. The modelled cumulative incidence of PC death does not match the data reported by Bill-Axelson et al. 148 exactly (23% vs 26% for untreated; 12% vs 16% for treated) owing to the application of UK age-specific rates of death from other causes and the application of a constant UK-specific risk of death from metastatic PC. However, the cumulative PC mortality rate is generally consistent with the data reported by Bill-Axelson and other similar cohorts; for example, Albertson et al. 168 estimated prostate-specific mortality of ∼20% at 15 years in men aged 60–64 years with a Gleason score of 6. 168 However, our modelled rates are significantly higher than those reported in the recently published PIVOT trial,149 which identified men through PSA screening. As such, sensitivity analysis was undertaken to assess the impact of recalibrating the model to yield the PC mortality rates observed by Wilt et al. 149

FIGURE 19.

Markov trace for undiagnosed/untreated local cancer in patients aged 60 years.

FIGURE 20.

Markov trace for diagnosed/treated local cancer for patients aged 60 years.

Mean costs and mean effects, and incremental analysis

Tables 26 and 27 present the mean costs, mean LYs, and incremental cost per LY gained for each strategy in men aged 60 years at the time of repeat biopsy (based on deterministic and the probabilistic analyses, respectively), assuming a prevalence of underlying cancer of 24%. Tables 28 and 29 present the same analyses using QALYs as the measure of effect. A breakdown of strategy costs into diagnosis and pre-diagnosis monitoring costs, biopsy complication costs, and cancer treatment and treatment complication costs is provided in Appendix 17. Appendix 17 also provides a summary of the expected numbers of unnecessary and appropriate biopsies undertaken with each strategy. Figures 21 and 22 present the findings of the cost per LY and cost per QALY analyses graphically on the cost-effectiveness plane. Strategies falling above and behind the lines plotted through the cost-effectiveness planes represent options that are more costly and less effective than other strategies or combinations of strategies. Strategies falling on the lines (the cost-effectiveness frontier) represent potentially cost-effective options, dependent on decision-makers' willingness to pay per LY or QALY gained.

FIGURE 21.

Cost-effectiveness frontier based on the cost per LY analysis (men aged 60 years, underlying cancer prevalence 24%).

FIGURE 22.

Cost-effectiveness frontier based on the cost per QALY analysis (men aged 60 years, underlying cancer prevalence 24%).

The base-case results show systematic extended-core TRUS/Bx to be the least costly option. However, using T2-MRI to determine which patients to biopsy (and to subsequently direct the biopsy) increases the costs by only a very small margin, with corresponding very small survival and QALY gains. Although these differences are very small and insignificant, T2-MRI-directed biopsy does have a favourable incremental cost per LY and QALY gained in comparison with systematic TRUS/Bx.

Using MRS to determine and direct biopsies results in a further cost increase and survival gain over T2-MRI but its incremental cost-effectiveness ratios (ICERs) are somewhat less favourable; although the incremental cost per QALY gained with MRS compared with systematic TRUS/Bx is just < £30,000 (deterministic analysis), it is > £30,000 in comparison with T2-MRI. Using positive findings on T2-MRI or MRS to determine and direct biopsies again increases costs, LYs and QALYs further. However, the ICERs (for LYs and QALYs) for using any visible abnormalities detected on T2-MRI or MRS, compared with only using abnormalities detected on MRS alone, are well above £30,000. This is due to a substantial loss of specificity associated with combined strategy, compared with using the findings on MRS alone to guide biopsy (31% for T2-MRI or MRS vs 76% for MRS alone), for only a small gain in sensitivity (96% vs 92%).

Tables 30 and 31 presents the incremental cost per LY analysis for men aged 70 years at the time of repeat biopsy, assuming a prevalence of underlying cancer of 24%. Tables 32 and 33 present the same analysis but use QALYs as the unit of outcome (see Appendix 17 for a breakdown of strategy costs by component categories). Figures 23 and 24 present the findings of the respective analyses graphically on the cost-effectiveness plane.

FIGURE 23.

Cost-effectiveness frontier based on the cost per LY analysis (men aged 70 years, underlying cancer prevalence 24%).

FIGURE 24.

Cost-effectiveness frontier based on the cost per QALY analysis (men aged 70 years, underlying cancer prevalence 24%).

A similar pattern of results is observed as for the cohort of men aged 60 years, but the survival benefit associated with the more sensitive strategies is smaller in the older cohort, owing to there being a higher risk of death from other causes (a competing risk for death from PC) and a smaller relative risk reduction associated with radical treatment in older men. As a consequence, the additional costs per LY and QALY gained with T2-MRI, MRS, and ‘T2-MRI or MRS’, are higher. However, the ICERs for T2-MRI compared with systematic TRUS/Bx remain < £30,000 despite the very small survival/QALY benefits.

Differential results for subgroups according to disease prevalence

Although few data were available to ascertain how diagnostic accuracy parameters vary by risk status of the cohort and underlying prevalence of cancer, Tables 34 and 35 present the incremental cost per LY and QALY findings for a high-prevalence cohort (50%) and low-prevalence cohort (10%) of men aged 60 and 70 years, respectively (assumes patient-level sensitivities not influenced by cancer prevalence). The findings seem to indicate that, for the 60-year-old cohorts (see Table 34), the more sensitive, more costly strategies have a higher chance of being considered cost-effective if used to direct biopsies in groups at higher risk of harbouring PC (e.g. men with ASAP on first biopsy). In the lower prevalence cohort, the T2-MRI strategy dominates systematic TRUS/Bx as a result of its specificity taking on greater significance with the underlying cancer prevalence set at only 10%.

A similar pattern of results is observed for the older cohort (see Table 35), with the cost-effectiveness of MRS improving relative to systematic TRUS/Bx and T2-MRI in the high-prevalence cohort. However, the ICER for MRS does not drop below £30,000 per QALY in this cohort. Moreover, the cost-effectiveness of T2-MRI rises above £30,000 in this cohort owing to the lower influence of specificity with high cancer prevalence combined with only very small gain in sensitivity with T2-MRI compared with systematic TRUS/Bx. Thus, none of the MRS/MRI sequences appears cost-effective in this older cohort.

For the 70-year-old low-prevalence cohort, the same finding is observed as for the low-prevalence 60-year-old cohort, i.e. T2-MRI dominates systematic TRUS-guided extended-cores biopsy for all.

Illustrative analysis incorporating diffusion-weighted-magnetic resonance imaging-directed biopsy

Although the lack of sensitivity/specificity estimates for DW-MRI in repeat biopsy cohorts precluded its incorporation in the base-case analysis, it was still felt to be a relevant alternative based on evidence from other cohorts coupled with its lower cost compared with MRS. As such, an illustrative analysis was undertaken to assess how it would compare in terms of cost-effectiveness if it could be demonstrated to have sensitivity at least equal to that of MRS (92%) and specificity at least equal to that of T2-MRI (55%). Table 36 presents the cost per QALY findings from this analysis for a 60-year-old cohort.

The findings indicate that if DW-MRI could be shown to achieve this level of diagnostic accuracy then it would be preferred on grounds of cost-effectiveness over MRS in this cohort of patients [see Table 32 (scenario 1) and Figure 25]. Under this scenario, DW-MRI is also borderline cost-effective compared with T2-MRI (incremental cost per QALY gained: £30,298).

FIGURE 25.

Comparison with DW-MRI included (assuming 92% sensitivity, 55% specificity).

When the sensitivities of DW-MRI and MRS are adjusted by cancer grade so that all false-negatives arising with these strategies occur in patients with low-risk cancer – to reflect the observation from other cohorts that MRS and DW-MRI positivity is highly correlated with tumour Gleason score – the ICER for DW-MRI compared with MRI falls to £18,260 [see Table 32 (scenario 2) and Figure 26], i.e. below the £20,000–£30,000 per QALY range often used to make judgements on cost-effectiveness. This analysis gives an indication of sensitivity/specificity requirements for DW-MRI to be considered cost-effective for directing biopsies (holding all other model parameters constant at base-case values).

FIGURE 26.

Comparison with DW-MRI included (assuming 92% sensitivity, 55% specificity, and that DW-MRI and MRS miss only low-risk cancers).

Deterministic sensitivity analysis scenarios (60-year-old cohort)

Several deterministic analyses were carried out to assess the sensitivity of the base-case cost per QALY findings to assumptions surrounding the incorporation of health-state utilities. We assessed the impact of applying a utility decrement of 0.035 (half of the disutility associated with having moderate anxiety rather than no anxiety on the EQ-5D) to patients with undiagnosed cancer, to reflect potential disutility resulting from raised anxiety associated with having a high PSA but no diagnosis. Further, we tested a multiplicative utility model whereby utility levels for diagnosed cancer states were set equal to the product of the cancer state utility and the utility of any treatment complications experienced. Finally, we assessed the impact of applying both of these modifications simultaneously. These deterministic sensitivity analyses were carried out only for the 60-year-old cohort, owing to there being a higher likelihood of changes to base-case assumptions affecting the cost-effectiveness of strategies in this group.

Table 32 shows the cost per QALY findings to be highly sensitive to these alterations. Under the first adjustment (utility decrement associated with undiagnosed cancer), the cost-effectiveness of MRS compared with T2-MRI improves substantially. However, applying the multiplicative model to further adjust utility for complications associated with radical treatment results in all of the MRS/MRI sequences having unfavourable incremental cost per QALY ratios in comparison with the less-sensitive systematic TRUS.

Applying both changes simultaneously (see Table 37, scenario 3) results in a pattern of findings more in keeping with the base-case analysis.

Further deterministic scenario analyses

The process of populating the model required a number of parameter and structural assumptions. To further assess the influence of these assumptions on findings, the following deterministic sensitivity analyses were undertaken:

1. Costs adjusted such that pathology costs for TRUS/Bx are increased by £86 to reflect a 25-minute increase in pathologist time for biopsies involving more than 10 cores, and MRI costs are adjusted to the NHS reference costs.

2. Sensitivity of MRS adjusted so that it misses only low-risk localised disease.

3. Comparator for MRS/MRI assumed to be a standard TRUS/Bx (10–12 cores) with sensitivity 60% (the lowest estimated sensitivity value obtained for systematic TRUS/Bx from studies assessed for inclusion in the systematic review) (see Appendix 11).

4. Application of the sensitivity/specificity estimates obtained from the indirect comparison (see Appendix 13.10).

5. Assumed a 14-core TRUS/Bx is £86 more costly than a MRI-/MRS-directed biopsy, and £112 more costly than obtaining a MRS scan.

6. Assumed that MRI-/MRS-directed biopsy reduces the risk of biopsy complications by 50% because fewer cores are obtained per patient.

7. Subsequent repeat biopsies (i.e. following a first repeat biopsy) have 95% sensitivity with 80% uptake (repeat offered every 12 months for those remaining with undiagnosed cancer).

8. Application of lower discount rate for health benefits (1.5% for QALYs vs 3.5% for costs).

9. Application of lower baseline risks of progression and less significant diagnosis treatment effects, based on new trial evidence on the effect of radical prostatectomy compared with watchful waiting. 149

10. Assumed all patients who are negative on MRI proceed with an extended 14-core TRUS biopsy.

Table 38 presents the results of these scenarios using QALYs as the unit of outcome. Appendix 17, Table 43, presents the results for these same scenarios using LYs as the unit of outcome.

These analyses demonstrate that T2-MRI-directed biopsy dominates systematic TRUS/Bx under several scenarios; specifically, when the sensitivity and specificity of T2-MRI are set equal to the values obtained from the indirect comparison (see Table 38, scenario 3), and when it is assumed that T2-MRI direction reduces the cost of the subsequent biopsy procedure relative to the cost of an extended-cores biopsy (scenario 4). The ICER for MRS compared with systematic TRUS also remains at < £30,000 for most scenarios. The cost-effectiveness of MRS compared with T2-MRI appears less sensitive to scenarios presented in Table 38. The ICER for MRS-directed biopsy (relative to T2-MRI) falls below £30,000 when (1) the cost of all biopsies is increased (scenario 1); (2) when it is assumed that MRS misses only low-risk cancer (scenario 2); and (3) when the discount rate applied to health benefits is reduced to 1.5% (scenario 7). Although increasing the costs of systematic extended-cores TRUS/Bx (but not MRI-/MRS-directed TRUS/Bx) results in both T2-MRI and MRS being more effective and less costly than systematic TRUS/Bx, this specific scenario has little impact on the comparison between MRS and T2-MRI. Application of the sensitivity/specificity estimates obtained from the indirect comparison also undermines the cost-effectiveness of MRS in relation to T2-MRI, owing to a substantial decline in the specificity of MRS.

The application of lower relative diagnosis/treatment effects and lower baseline cancer progression rates (scenario 8) undermine the cost-effectiveness of MRS compared with T2-MRI and systematic TRUS/Bx. The incremental cost per QALY gained with the most sensitive imaging strategy (positive on either T2-MRI or MRS) does not fall below £30,000 for any of the scenarios assessed. Further, none of the MRI strategies compares very favourably in terms of cost-effectiveness when it is assumed that all patients who are negative on MRI proceed to an extended-cores TRUS/Bx regardless (scenario 9).

In addition to the scenarios presented in Table 38, threshold analysis would suggests that MRS would dominate systematic extended-cores TRUS/Bx in contexts where the cost saving per biopsy averted is ≥ ∼£115 more than the cost of obtaining the MRS sequences to determine whether or not a biopsy should proceed.

Probabilistic sensitivity analysis

To assess the impact of joint uncertainty surrounding all model parameters and inputs, appropriate distributions were assigned and randomly sampled for each of 1000 iterations of the base-case analysis. The results were used to estimate the probability of each diagnostic strategy being preferred on grounds of cost-effectiveness for different values of decision-makers' willingness to pay for a QALY. The resultant CEAC is displayed in Figure 27. The results indicate that under base-case parameter values and assumptions, none of the strategies demonstrates a high probability of being the preferred option at the threshold value of £30,000 per QALY gained. The cost-effectiveness acceptability frontier (Figure 28) displays the probability of the strategy with the highest NMB, at different values of decision-makers’ willingness to pay per QALY gained, being cost-effective. At a threshold ratio of £30,000 per QALY, T2-MRI provides the option with the highest NMB but with only a 34% probability. The expected value of perfect information at this threshold (i.e. of eliminating uncertainty) is £27.30 per patient.

FIGURE 27.

Cost-effectiveness acceptability curves under base-case assumptions.

FIGURE 28.

Cost-effectiveness acceptability frontier under base-case assumptions.

In order to assess the sensitivity of the probabilistic analysis findings to several base-case assumptions, the analysis was repeated with the biopsy costs inflated to account for the possibility of higher pathology time requirements and the MRS/MRI costs were adjusted to the reference cost for direct access DCE-MRI (see Table 38, scenario 1). Further, DW-MRI was incorporated under the assumption that it could achieve sensitivity equal to that of MRS and specificity equal to that of T2-MRI. In addition, it was assumed that MRS and DW-MRI would miss only low-risk cancer (see Table 38, scenario 2). Under this specification, there is 74% probability of either DW-MRI or MRS being cost-effective at a willingness-to-pay threshold of £30,000 per QALY gained (Figure 29). MRS in fact retains the higher probability of being cost-effective in this scenario because, with biopsy costs set at a higher level, the superior specificity of MRS (over the assumed specificity of DW-MRI, which was set at 55% purely for illustrative purposes) outweighs the additional cost of running the sequence.

FIGURE 29.

Cost-effectiveness acceptability curves, assuming DW-MRI has sensitivity equal to that of MRS and specificity equal to that of T2-MRI, and that MRS and DW-MRI miss only low-risk cancer.

Summary of key results

The results of the deterministic economic modelling suggest that, when considering LYs as the unit of outcome, the use of T2-MRI, to determine and direct biopsies, may be cost-effective in comparison with systematic TRUS-guided extended-cores biopsy. This results from its modest implementation cost and slightly improved sensitivity over systematic extended-cores TRUS/Bx (14–16 cores). At the same time its specificity would suggest that it could avert the need for 55% of patients without cancer having to undergo a biopsy. The base-case incremental cost per QALY estimates for the more sensitive enhanced MRS/MRI techniques are somewhat less favourable (i.e. are > £30,000 per QALY gained in comparison with the next less costly option). However, the ICER for MRS compared with T2-MRI does fall below £30,000 in the high-prevalence (50%) 60-year-old cohort. In the lower-prevalence (10%) cohorts, T2-MRI was found to dominate systematic TRUS/Bx (i.e. be less costly and more effective) owing to its specificity, resulting in more biopsies being averted in this group.

Moreover, the deterministic sensitivity analysis shows the cost-effectiveness of MRS compared with T2-MRI (and systematic TRUS/Bx) to be particularly sensitive to two key parameters. The ICER for MRS falls below £30,000 when (1) the cost of biopsies is increased to £298 and (2) when MRS is modelled to detect all moderate- and high-risk cancer (only missing low-risk disease). The latter assumption is in keeping with data from case series, which suggest high levels of correlation between MRS positivity and tumour Gleason scores. 52,54 The cost-effectiveness of MRS also improves considerably when the accuracy of and compliance with subsequent repeat biopsies decreases. Thus, our findings would suggest that the use of MRS may well be cost-effective in certain contexts, for example in settings where the cost of TRUS/Bx exceeds the cost of obtaining a MRS sequence by ∼£115. It is likely that practice and costs will vary substantially locally.

Unfortunately, there were insufficient data on the diagnostic accuracy of DW-MRI in the population of interest, and as such we did not include it as a comparator in the base-case analysis. With its lower cost in comparison with MRS, our modelling suggests it could represent a cost-effective approach if it could be shown to have sensitivity similar to that of MRS and specificity similar to that of T2-MRI. At these levels of sensitivity/specificity, however, it might also need to be able to discriminate between low-risk and moderate/high-risk disease (so that false-negatives would be concentrated in the low-risk cases) to be cost-effective in comparison with T2-MRI alone. Evidence from other case series suggest it may be able to do this. 169,170

By the same token, changes to some of the model assumptions also undermine the cost-effectiveness of strategies that improve cancer detection rates at increased costs over standard practice and T2-MRI. These include reductions in the baseline risk of disease progression and reductions in the relative risk reductions associated with diagnosis and treatment.

Considering the probabilistic sensitivity analysis, when parameter distributions are centred on the base-case point estimates and all of the strategies are compared simultaneously, none of the strategies achieves a high probability of being preferred on grounds of cost-effectiveness (see Figure 27) at the threshold ceiling ratio of £30,000 per QALY gained. Although the point estimate of the ICER of T2-MRI compared with systematic TRUS/Bx is favourable, its probability of being cost-effective at £30,000 per QALY is only ∼60% in comparison with TRUS/Bx and only 34% considering all strategies simultaneously (see Figure 28).

However, Figure 29 demonstrates how the uncertainty surrounding the relative cost-effectiveness of strategies reduces if the average cost of biopsies is increased to £300 (the upper quartile reference cost reported for biopsies carried out in an outpatient setting) and MRS and DW-MRI are modelled to miss low-risk cancer only. Under this alternative model specification, the choice between strategies becomes one between MRS and DW-MRI above a willingness-to-pay threshold of ∼£20,000 per QALY gained. Note this still assumes disease progression rates and relative diagnosis/treatment effects in line with Bill-Axelson et al. 148

Generalisability of results

In developing the model we have attempted to use data applicable to the UK setting as far as possible. However, many of the data on disease progression and relative treatment effects were derived from a European cohort identified in the pre-PSA era. 148 Although we have tried as far as possible to model the risk of progression by clinical grade (low, intermediate, high), it is possible that the progression rates observed for low-risk patients in this pre-PSA cohort are higher than would be observed for low-risk patients identified in clinical practice in the UK today. This is potentially important as reducing progression rates and treatment effects for low-risk patients reduces the cost-effectiveness of more sensitive and more costly diagnostic strategies, and further emphasises the potential importance of further research to assess the sensitivity of alternative imaging sequences by tumour grade in cohorts of patients undergoing biopsy. The ongoing PROMIS trial may help address this question. 171

The modelling also relied on health-state utility data from a number of sources outwith the UK. Attempts were made to identify EQ-5D data from UK cohorts for the modelled states and treatment complications of interest, but limited data were identified. A decision was made to use the EQ-5D utilities observed for a European cohort post diagnosis and at varying time points post treatment with radical prostatectomy or EBRT. As these average utilities were obtained from cohorts with proportions experiencing complications, we chose not to further adjust utility for modelled treatment complications in the base-case analysis. Given a lack of EQ-5D-based utility estimates for metastatic and castration resistant metastatic disease, for men modelled to progress to metastatic disease we applied the average time trade-off values elicited for these states from a sample of US men. 166

Although there is uncertainty surrounding the wider applicability of some of the progression rates, treatment effects and utility values applied in the model, the cost inputs were based on the estimation of resource use in UK settings according to current guidelines. National average unit costs were applied to resource-use estimates wherever possible, making individual cost inputs generalisable across the UK. However, there is uncertainty as to how hospitals are reimbursed for TRUS-guided biopsies at the local level and also the true opportunity cost of TRUS/Bx compared with that of running MRS/MRI sequences. It would be useful if future prospective studies could estimate these more accurately using patient-level data.

Strengths and limitations of the analysis

Dichotomising test results where studies reported an ‘equivocal’ category

Some studies reported suspicious results (test-positives), normal results (test-negatives) and a third category, neither positive nor negative, such as ‘equivocal’. In order to incorporate these results into a 2 × 2 table and for inclusion in meta-analyses they had to be dichotomised as either positive or negative. Our position was to class equivocal results along with positive results rather than with negative results. This approach increased sensitivity and decreased specificity, whereas the reverse would have been the case if equivocal results had been classed along with negative results. In Appendix 8, when studies have reported an ‘equivocal’ category, where possible, we have shown sensitivity and specificity both for equivocal cases classed with positive results and for equivocal cases classed with negative results. For example, in the study by Yuen et al. ,112 for MRS, when equivocal was classed as ‘suspicious’ sensitivity was 71.4% and specificity was 52.9%, whereas when equivocal was classed as ‘normal’ sensitivity was 57.1% and specificity was 82.4%.

False-positives

Eleven studies57,74,76,78,79,101,106,108,109,133,137 provided some additional information on the nature of their false-positive results (i.e. the test detected an abnormal area but the histopathological assessment of the biopsy cores taken from that area was negative for cancer). The false-positive rate for patient-level analysis (six studies74,76,79,101,108,137) ranged from 2.4% to 100% and for biopsy-level analysis (five studies57,78,106,109,133) ranged from 13.0% to 46.2%. High-grade PIN and prostatitis accounted for a substantial proportion of false-positives. One study presented this information separately for MRS and T2-MRI. 79 For MRS (11 false-positives), PIN accounted for six (54.5%), fibrosis for four (36.4%) and normal prostatic tissue for one (9.1%). For T2-MRI (13 false-positives) PIN accounted for three (23.0%), fibrosis for five (38.5%) and normal prostatic tissue for five (38.5%).

True-negatives

An extended TRUS/Bx procedure may miss cancers, as the transrectal approach renders sampling of apex tumours and anterior TZ tumours difficult. Using this as the approach to provide the biopsies for the reference standard has its limitations, especially if not combined with long-term follow-up. The number of true-negatives in these types of study could therefore potentially be lower than reported.

Detection of clinically significant disease

Using comprehensive (saturation) biopsy protocols based on TRUS/Bx may reduce the likelihood of missing cancers. However, although saturation biopsies may improve the detection rate of PC, solely increasing the number of biopsy cores may also lead to an increase in the detection of clinically insignificant disease. In addition, saturation biopsies have the disadvantage of possibly requiring anaesthesia and increasing the risk of adverse events. 105 On the other hand Scattoni et al. 173 reported that the detection rates of protocols including 20–38 cores ranged from 14% to 41% without significantly increasing the likelihood of detecting clinically insignificant cancers compared with initial or repeat biopsy.

One of the suggested advantages of MRS and other MRI techniques is the ability to detect clinically significant disease (Gleason score of ≥ 7). An explanation put forward for this in relation to MRS is that it is unable to detect the lowest grade of PC due to partial voluming of healthy surrounding tissue included in a spectroscopic voxel of spatial resolution 0.32 cm. 106

Twenty-nine studies74,76,78–84,86,87,90,91,95–98,102–106,108–113,136 reported the Gleason score based on the biopsy results of patients diagnosed with PC. Most studies reported a median Gleason score of ≥ 6 and in the one study96 reporting DW-MRI the median Gleason score was 7. Across six MRI studies reporting a median Gleason score of > 6,74,90,96,104,108,109 the percentage of patients with a Gleason score of ≥ 7 ranged from 50% to 66.7%. The limited evidence suggests that, potentially, a substantial number of cancers detected by MRS and other MRI techniques in patients with raised PSA levels and previous negative biopsy may be clinically significant (Gleason score of ≥ 7 and/or volume > 0.5 ml).

The experience of clinical experts (Anwar Padhani, Mount Vernon Cancer Centre, July 2012, personal communication) suggests that DW-MRI is capable of detecting more clinically significant disease than TRUS/Bx, and that cancers missed on DW-MRI that are detected on TRUS/Bx are generally not clinically significant. However, there is a lack of evidence on the detection of clinically significant disease by DW-MRI in the population of men with a previously negative biopsy who still have raised PSA levels and a continuing suspicion for cancer. There is some evidence on the ability of DW-MRI to detect clinically significant disease in the wider population of men with PC; for example, Hambrock et al. 64 undertook DW-MRI of 51 patients before prostatectomy and found that the median ADC in the tumours was negatively correlated with Gleason score in the PZ of the prostate. However, further research is needed to assess the extent to which this finding applies to men with suspected PC but a previously negative biopsy. In the TZ, although there is a small significant difference in ADC between Gleason 3 + 3 and Gleason 4 + 4 cancers, the overlap in ADC between the two cancer groups is so large that discrimination on an individual level is not possible (Tom Scheenen, Radboud University Nijmegen Medical Centre, July 2012, personal communication).

Cancer detection in the transition zone by magnetic resonance spectroscopy and other magnetic resonance imaging techniques

Hoeks et al. 87 commented that MRS had problems imaging the TZ and also noted that different choline–creatine ratios were needed for cut-offs for a positive test result for the PZ and TZ. The authors stated that their relatively high detection rates of PC in the TZ in men with one or more negative TRUS-guided biopsies agreed with the results of Hambrock et al. 86 (57% in the TZ). However, they noted that, in other reports by Roethke et al. 104 and Franiel et al. ,84 TZ cancer detection rates (47% and 35%, respectively) were lower than PZ cancer detection rates (53% and 64%, respectively). Testa et al. ,106 in a region-based analysis, also reported sensitivity and specificity separately for the PZ and TZ for MRS and T2-MRI. For the PZ, MRS sensitivity and specificity were 64.9% and 85.8%, respectively, compared with 72.2% and 93.3% for the TZ. T2-MRI sensitivity and specificity for the PZ were 27.0% and 95.8% respectively, compared with 61.1% and 98.9% for the TZ.

Heterogeneity across the studies

Across the studies, the prevalence of PC ranged from 9.5%76 to 100%. 117 The original biopsy scheme used will influence the prevalence of PC in patients with raised PSA and previously negative biopsy; the more cores taken during the original biopsy scheme(s), the more cancers are likely to be detected at this stage and consequently the prevalence of cancer in subsequent biopsies will be lower. The previous biopsy schemes reported by the studies ranged from four or six core57 to 12 core. 74,78,93,116 The number of previous biopsy sessions reported by the studies for their patient populations ranged from 1 to 9, with different numbers of sessions occurring within studies as well as across studies.

Transrectal ultrasonography: imaging compared with obtaining biopsies

Transrectal ultrasonography is used to either visualise the prostate in order to obtain a systematic, predefined biopsy (TRUS/Bx) or inspect the prostate for evidence of cancer and biopsy highly suspicious areas. Based on advice from clinical experts, the most common use for TRUS in the NHS is to obtain a systematic, predefined biopsy. Therefore, TRUS sensitivity could mean:

• The proportion of patients with prostate cancer correctly identified on systematic predefined biopsy This is complicated by the fact that TRUS is not independent from the reference standard, as discussed elsewhere. Therefore, to obtain the false-negatives necessary to calculate sensitivity, at least one repeat biopsy is needed. In studies that use this method, low-risk patients do not usually undergo repeat biopsies. This can lead to a falsely high sensitivity, as it assumes that all patients who did not undergo a biopsy did not have cancer. Furthermore, if there is a considerable interval between biopsies, there is the potential that an individual may have developed cancer in the intervening period.

Alternatively, TRUS sensitivity could mean:

• The proportion of patients with prostate cancer correctly identified when TRUS is used to identify suspicious lesions This scenario is complicated because often both systematic and targeted biopsies are taken and in reported studies is it unclear if the sensitivity refers to the combination or just the targeted lesions.

Systematic biopsies used in conjunction with magnetic resonance spectroscopy and other magnetic resonance imaging techniques

In studies reporting the sensitivity and specificity of MRS or other MRI techniques, a number of cores were targeted for biopsy, based on suspicious areas identified by the imaging test and a systematic (generally 10- or 12-core) TRUS/Bx was also undertaken. In most studies it was unclear how the results from the systematic biopsy contributed to the sensitivity and specificity values reported for the imaging technique, i.e. it was unclear whether the sensitivity and specificity reported were for the imaging test alone or the imaging test plus the systematic biopsies.

Subsidiary questions from the protocol

There was insufficient information from the included studies to address the subsidiary questions of (1) identifying specific patient groups in which MRS and other MRI techniques are most clinically effective; (2) whether or not these techniques can identify cases in which PC is present but further procedures are unnecessary and (3) evidence of use of MRS and other MRI techniques leading to a change in patient management.

Ongoing studies

The search strategy identified a few ongoing studies, although none focused on our population of interest – men with suspected PC and elevated PSA level but previously negative biopsy. The largest of the ongoing studies identified was the UK multicentre study ‘PROstate MRI Imaging Study: evaluation of multiparametric magnetic imaging (MP-MRI) in the diagnosis and characterisation of PC (PROMIS)’ that was anticipated to start in March 2012 and end in April 2015. 171 The objectives of this study are to (1) determine the ability of MP-MRI to identify men who can safely avoid biopsy; (2) assess the ability of the MP-MRI-based diagnostic pathway to improve the rate of detection of clinically significant cancer compared with TRUS/Bx; and (3) estimate the cost-effectiveness of an MP-MRI-based diagnostic pathway. The study design is described as a prospective validating paired cohort study. Participant inclusion criteria are men who (1) are aged ≥ 18 years who are at risk of PC and have been advised to have a prostate biopsy; (2) have a PSA level value of ≤ 15 ng/ml in the last 3 months; (3) have suspected stage of ≤ T2 on rectal examination (organ confined); and (4) are fit for general/spinal anaesthesia and all study procedures. Exclusion criteria include a previous history of prostate biopsy. The intervention is MP-MRI scan and combined prostate biopsy procedure (template prostate mapping biopsy) followed by TRUS/Bx. The primary outcome measures are (1) proportion of men who could safely avoid biopsy; (2) proportion of men correctly identified by MP-MRI to have clinically significant PC; and (3) primary definition of cancer according to biopsy: dominant Gleason pattern of ≥ 4 and/or cancer core length of ≥ 6 mm.

Another ongoing study is the ‘Prostate Cancer Localization With a Multiparametric Magnetic Resonance (MR) Approach’. 174 This is a prospective, observational, international, multicentre study that started in June 2010. Its primary objective is to prove the diagnostic accuracy of in vivo 3-T multimodality MRI (high-resolution T2-MRI, DCE-MRI, MRS and DW-MRI techniques) in distinguishing carcinoma from other prostate tissue. Specific objectives include (1) determining the diagnostic accuracy of 3-T multimodality non-endorectal coil MR imaging in localising PC and (2) proving that multimodality MR data allow for predicting tumour grade. Inclusion criteria include men with biopsy-proven diagnosis of PC in whom radical prostatectomy and histopathological examination are planned.

Comparison of our results with other systematic reviews

Our searches identified four other systematic reviews70,175–177 that assessed MRI techniques for detecting PC in men, although only the review by Lawrentschuk and Fleshner176 focused on men with previous negative biopsies and elevated PSA levels.

Umbehr et al. 70 undertook a systematic review and meta-analysis of MRI combined with MRS in the diagnosis of PC. Thirty-one studies were included, seven of which recruited participants with a previous negative biopsy. Six74,76,79,100,101,112 of these seven studies74,76,79,100,101,112,178 were included in our review; the seventh178 was excluded as the participants did not have a previous negative biopsy or this was unclear. The authors performed a meta-analysis of seven studies57,74,76,79,100,101,112 examining patients with suspected PC and found a sensitivity of 82% (95% CI 59% to 95%) and specificity of 88% (95% CI 80% to 95%). However, in this meta-analysis only four74,76,79,112 of the seven studies57,74,76,79,100,101,112 involved men with a previously negative biopsy.

Lawrentschuk and Fleshner176 undertook a review of studies of MRI or MRS which recruited participants with a previous negative biopsy and persistently elevated PSA. Six studies were included, five of which were included in our review;57,74,79,101,112 the sixth179 was excluded as the test used was outwith our inclusion criteria. The authors did not statistically pool the results, but rather narratively presented each study. For MRI or combined MRI and MRS, they reported a sensitivity of 57% to 100% and a specificity of 44% to 96%. The authors found that 54% of patients (34/63) were diagnosed with cancer solely on the basis of a MRI-targeted biopsy.

Wang et al. 177 undertook a meta-analysis of PC studies that used MRS as a diagnostic tool. The inclusion criteria were not limited to men with suspected PC and previously negative biopsy. Seven studies were included, of which two were included in our review;101,112 the remaining five18,51,53,180,181 were excluded as the types of participants were outwith our inclusion criteria. The authors reported the sensitivity and specificity of MRS using a CC/C ratio cut-off of 0.75 as 82% and 68% respectively, and with a cut-off of 0.86 as 64% and 86%, respectively.

Engelbrecht et al. 175 performed a systematic review of local staging of PC using MRI. The authors included 76 studies and calculated the area under the ROC curve using trapezium methodology. It was not reported how many of these studies included participants with a previous negative biopsy and raised PSA and the list of included studies was not included in the list of references. On the ROC curve the joint maximum sensitivity and specificity of MRI was at 71%; however, the authors found unexplained heterogeneity throughout the results.

Future technological developments

Magnetic resonance-guided biopsies are not usually carried out in the UK, resulting in challenges in ensuring the correspondence of TRUS/Bx spatial accuracies to suspicious areas identified by MRS/MRI. Hoeks et al. 87 stated that the clinical use of MR-guided biopsies was currently restricted by limited availability and long procedure times. They commented that the application of MRI–ultrasound fusion techniques, needle-guided tracking sequences, and implementation of robotics may improve these drawbacks in the near future and that, when these issues were resolved, multiparametric-MRI- and MR-guided biopsies could be applied on a larger scale for PC detection in patients with an elevated PSA level and one or more negative TRUS/Bx sessions.

Strengths

Attempts have been made to use the best available evidence to model both the diagnostic pathways and subsequent treatment pathways and outcomes. The model provides a flexible framework allowing the comparison of many different diagnostic strategies in the context of the patient's lifetime. It captures the trade-off between the increased upfront costs of imaging and the reduction in subsequent biopsies, and also the trade-off between earlier diagnosis and potential utility decrements associated with treatment side-effects. Further, the model is risk stratified to allow for comparison of strategies that vary in their ability to differentiate between tumours of different stage and grade. The model can easily be updated to incorporate new, more detailed evidence as it becomes available.

Limitations

The modelling was hampered by limited availability of data on the diagnostic accuracy of alternative diagnostic strategies, the natural history of cancer detected at repeat biopsy, and the impact of diagnosis and treatment on disease progression and health-related QoL.

The systematic review uncovered limited data on the relative sensitivity (cancer detection rates) of MRI-/MRS-targeted biopsy techniques and systematic TRUS-guided approaches. In particular, there were no identified studies providing head-to-head comparisons of MRI-/MRS-targeted approaches and systematic TRUS-guided sampling schemes based on different numbers of cores. Further, it was not possible to obtain pooled estimates for the sensitivity/specificity of alternative MRS/MRI sequences by grade of tumour, which is potentially an important parameter for informing cost-effectiveness. There is therefore a need for a large comparative study to prospectively assess the sensitivity of systematic approaches and MRS/MRI sequences for detecting cancer by D'Amico risk strata. It would also be beneficial for follow-up to be built into such a cohort study to ascertain how contemporary cohorts are treated, how they progress over time, and how their health-related QoL is affected by diagnosis and subsequent treatment with different modalities.

Although the model attempted to capture all the important clinical and cost events, it was not possible to capture and/or value all the important factors that might influence cost-effectiveness. For example, we were not able to ascertain and assign utility decrements for pain and short-lived complications associated with undergoing biopsy. Further, we did not have a good source of EQ-5D utility weights for a UK-based cohort of patients undergoing repeat biopsy and follow-up, making it necessary to draw on alternative sources. It would be beneficial to incorporate a measure of health-state utility into future cohort studies assessing the accuracy of alternative approaches to diagnosis. In addition, with the survival and QALY gains being so small, and of questionable clinical significance, the choice between strategies might be better informed by patient or public preferences for process of care factors to which the standard QALY model may be insensitive.