Optimising the Diagnosis of Prostate Cancer in the Era of Multiparametric Magnetic Resonance Imaging: A Cost-effectiveness Analysis Based on the Prostate MR Imaging Study (PROMIS)

Background: The current recommendation of using transrectal ultrasound-guided bi- opsy (TRUSB) to diagnose prostate cancer misses clinically significant (CS) cancers. More sensitive biopsies (eg, template prostate mapping biopsy [TPMB]) are too resource intensive for routine use, and there is little evidence on multiparametric magnetic resonance imaging (MPMRI). Objective: To identify the most effective and cost-effective way of using these tests to detect CS prostate cancer. Design, setting, and participants: Cost-effectiveness modelling of health outcomes and costs of men referred to secondary care with a suspicion of prostate cancer prior to any biopsy in the UK National Health Service using information from the diagnostic Prostate MR Imaging Study (PROMIS). Intervention: Combinations of MPMRI, TRUSB,


Introduction
Multiparametric magnetic resonance imaging (MPMRI) is increasingly being recommended for the diagnosis of clinically significant (CS) prostate cancer, if the initial biopsy proves negative [1,2]. An alternative approach is to begin with MPMRI imaging to inform who needs a biopsy and, in those who need it, how it might be best conducted [3]. Recent studies have reported encouraging results on the performance of MPMRI in detecting CS prostate cancer [3][4][5]. The Prostate MR Imaging Study (PROMIS) was the largest accuracy study on the use of MPMRI and transrectal ultrasound-guided biopsy (TRUSB) in the diagnosis of prostate cancer [4]. Using template mapping biopsy (TPMB) as the reference standard, it was found that MPMRI had better sensitivity for CS prostate cancer compared with TRUSB but worse specificity [4]. It is therefore necessary to explore how best to combine these tests and the consequences of incorrect diagnosis on health outcomes. This study aims to identify the combinations of testsdiagnostic strategies-that detect the most CS cancers per pound spent in testing and achieve the maximum health given their cost to the healthcare service.

Patients and methods
The target population was men at risk of prostate cancer referred to secondary care for further investigation [4,6]. The perspective was the UK National Health Service (NHS). Costs were expressed in pound sterling from a 2015 price base. The time horizon is the population's predicted lifetime. Costs incurred and health outcomes attained in the future were discounted to present values at 3.5% per annum [7].

Diagnostic strategies
The diagnostic strategies consisted of clinically feasible combinations of MPMRI, TRUSB, and TPMB, in addition to the use of TRUSB and TPMB in isolation (Table 1; details in the Supplementary material, section 1.1).
These included strategies using MPMRI to decide whether a TRUSB or TPMB is necessary and target the TRUSB, and strategies starting with TRUSB and using MPMRI to decide whether a repeat biopsy is warranted.
A diagnosis of CS cancer requires a biopsy, hence strategies were defined to always end with a confirmatory biopsy. Within each test combination, there are alternative ways each test can be used, following the definitions used in PROMIS (see Tables 2 and 3). Each of the 32 test combinations were tested for the alternative classifications and cut-offs, returning a total of 383 strategies.

Model structure
The model had a diagnosis and a long-term component ( Supplementary   Fig. 1). For diagnosis, a decision tree combined the information on diagnostic accuracy of the tests to determine the accuracy of the test combinations (Fig. 1)

Health-related quality of life and costs
For health-related quality of life (HRQoL), the model considers the direct impact of TPMB, obtained from the patient-reported EQ-5D collected in the PROMIS [4]. TRUSB is assumed to have no impact on HRQoL given that no effect was found in a large European screening study [10]. Regarding costs, the model included the direct cost of the tests and the costs associated with managing their related complications [11,12].
In the long term, the model considers the reduction in HRQoL from any metastatic disease [13] and ageing [14]. The model included the direct cost of radical prostatectomy and surveillance, the costs of their complications, and the costs of metastatic disease [8]. Details are The diagnostic performance of MPMRI was obtained from the individual patient data of the PROMIS [4]. For interpretation of MPMRI, the definitions for CS cancer were a radiologist estimation of (1) lesion volume 0.5 cc and/or Gleason score 4 + 3, and (2) lesion volume 0.2 cc and/or Gleason score 3 + 4. Suspicion of a lesion meeting these definitions was scored on a Likert scale (1-5, 1 being highly likely benign and 5 being highly likely malignant). This scale was also used to score the image for whether any cancer (whether considered CS or not) is present.  (1) dominant Gleason pattern 4 and/or any Gleason pattern 5 and/or cancer core length 6 mm (histology definition 1) and (2) any Gleason pattern 4 and/or cancer core length 4 mm (histology definition 2). Since the PROMIS collected information on blind first TRUSB, external evidence was used on the sensitivity of repeat TRUSB and MRI-targeted TRUSB, as either first or second TRUSB [16,22,23].

Main outcomes and measures
The main outcomes were cost effectiveness of diagnosis, defined as the strategies that detect the most CS cancers for a given pound spent in testing, and long-term cost effectiveness, defined as the strategies that achieve the most health outcomes given their costs, for alternative cost-  Figure 2B shows the expected lifetime health outcomes and costs achieved by each strategy per man referred for testing  Each bubble represents one of the 383 diagnostic strategies evaluated; their size is directly related to the probability that the strategy is cost effective and therefore forms the frontier (ie, forms the red line). The red bubbles represent the 14 diagnostic strategies that form the frontier at expected values. This means that, on average, these are the best strategies per pound spent. The black bubbles represent the strategies that do not form the frontier at expected values, but that have some probability of being in the frontier given their distribution of costs and outcomes. The grey bubbles represent the strategies that do not form the efficiency frontier at any simulation. Given the distribution of parameter inputs, these strategies are never efficient or cost effective. CS = clinically significant; NHS = National Health Service. (see the Supplementary material, section 9, for details, including costs in euro). The line linking the cost-effective strategies (in red) is the cost-effectiveness frontier, and its slope corresponds to the incremental cost-effectiveness ratio (ICER) of a strategy versus the next best (to its left); the strategies on the frontier and their ICERs are shown in Table 4. The strategy attaining the greatest expected health outcomes was P4 2, and the next best strategy is M7 222. The ICER of P4 2 versus M7 222 was £30 084/QALY. Next best to M7 222 is T7 223, and the ICER of M7 222 versus T7 223 is £7076/QALY gained, making it a cost-effective strategy in the UK setting. These results are consistent with the cost-effectiveness acceptability frontier (Supplementary Fig. 1), in which M7 222 is the strategy most likely to be cost effective for cost-effectiveness thresholds between £7250 and £30 000/QALY.

Sensitivity analysis
The cost-effective strategy changed from M7 222 to T7 222, T9 222, or P4 2 in response to a reduction in the sensitivity of MRI-targeted TRUSB and an increase in the sensitivity of the MRI-targeted second TRUSB. The cost-effective strategy changes to P4 2 if the sensitivity of the MRI-targeted second TRUSB reduces, as this does not reduce its the CS cancer detection rates. Increases in the cost of MPMRI coupled with reductions in the cost of TRUSB result in strategies starting with TRUSB becoming cost effective, while reductions in the cost of TPMB favour strategies involving TPMB for all or a large proportion of men. The cost-effective strategy changed to less costly, less sensitive strategies (T7 223 and T6 222) if radical prostatectomy is less cost effective, for example, due to reduced effectiveness, higher HRQoL burden or greater costs. Conversely, the costeffective strategy changed to more sensitive strategies (P4 2) in men incorrectly classified as no cancer has worse health outcomes. For full results, see the Supplementary material, section 9.

Discussion
A diagnostic strategy consisting of MPMRI first and up to two MRI-targeted TRUSB at the more sensitive definitions (definition 2) and cut-offs is more likely to be cost effective at cost-effectiveness thresholds at and below £30 000. For MPMRI, this is lesion volume 0.2 cc and/or Gleason score 3 + 4 (likely benign or above); for TRUSB this is any Gleason pattern 4 and/or cancer core length 4 mm. The most clinically effective strategy is testing all men with TRUSB at definition 2 and retesting men in whom CS cancer was not detected with TPMB; however, this is not cost effective at current cost-effectiveness thresholds and will not be clinically feasible to deliver across the board in any healthcare setting. These findings can directly inform UK policy, but they can also be generalised to similar, international, settings. The extent to which the cost effectiveness results can be generalised to other jurisdictions depends on the similarities of the population, outcomes, health systems, and pricing. The strategies in the cost-effectiveness frontier are shown, together with their ICERs versus the next best strategy. TRUSB after an MPMRI is assumed to be an MRI-targeted TRUSB, as information on the location of the lesion is provided by the MPMRI.
The sensitivity of MPMRI and TRUSB depends on their definitions and cut-offs. An MPMRI cut-off of 2 and above refers 96% of men to biopsy, but ensures that only 2% of men with intermediate-risk cancer and none of the men with high-risk cancer are missed. Furthermore, it means that most men receive a more sensitive TRUSB, since MRItargeted TRUSB is thought to be more sensitive than the standard [16]. The recent guidance based on the Prostate Imaging Reporting and Data System (PI-RADS) suggests that men with PI-RADS 1 or 2 should not be referred for biopsy given concerns about overdiagnosis [17]. This may not be equivalent to the cut-off recommended here, since the PROMIS diagnostic study did not use PI-RADS, which is a limitation. Nonetheless, higher MPMRI cut-offs, whilst reducing the proportion of men receiving biopsy, also reduce the proportion of CS cancers detected and treated. This is the first study comparing all possible ways of using MPMRI, TRUSB, and TPMB to diagnose CS prostate cancer, using data from PROMIS, the largest study on MPMRI and TRUSB [4]. A limitation of PROMIS, and of this study, is that it did not include other tests, such as transperineal biopsies, or the combination of additional clinical and genetic characteristics for diagnosis and risk stratification. This is an area for future research. Another area for future research is the sensitivity of the first and second MRI-targeted TRUSBs, since these parameters were key cost-effectiveness drivers. Previous cost-effectiveness studies compared up to two ways of using MPMRI, either as a first test to determine which men should receive MRI-targeted TRUSB [1,18], as MRI-targeted TRUSB for all men [1], or for men with previous negative biopsy [19]. For these reasons, this study is the most comprehensive cost-effectiveness analysis to date of alternative diagnostic strategies for prostate cancer.
The appropriate MPMRI cut-off, and ultimately the optimal diagnostic strategy, depends on the cost effectiveness of early diagnosis and treatment. Although this study did not include radiotherapy, it tested the impact of changes on the cost effectiveness of treatment. If cost effectiveness of radiotherapy is similar to or more favourable than that of radical prostatectomy, highly sensitive strategies such as M7 222 are cost effective. Highly sensitive diagnostic strategies may not be cost effective if radical treatment is not as cost effective in the manner modelled here. The cost effectiveness of treatment is less favourable if (1) treatment is less effective, (2) it impacts negatively on HRQoL, or (3) it is costlier than that assumed for this study.
Management of men classified as having no cancer or non-CS cancer also has an impact on the scope for investment in diagnosis. More sensitive monitoring protocols improve the cost effectiveness of less sensitive and less costly diagnostic strategies. There is a dearth of evidence on the effectiveness of repeated testing protocols, which constitutes an important limitation of the current evidence base in support of policy, and meant that these analyses could not formally evaluate the use of such protocols.
In order to evaluate the cost effectiveness of diagnostic tests, evidence is required on the long-term outcomes of patients who are correctly diagnosed and those who are misclassified, given their true disease status. The extensive literature searches conducted for this study did not identify evidence on the outcomes of patients and the effectiveness of treatments when true disease status is known (eg, using TPMB to identify and risk stratify patients). The existing studies used TRUSB to diagnose and risk stratify patients [8,20,21]; hence, some individuals may have been underdiagnosed. As a consequence, their long-term qualityadjusted survival may have been overestimated, and the cost effectiveness of treatment may have been underestimated. This issue can only be resolved with betterquality evidence on the outcomes of men with prostate cancer, based on a perfect test such as TPMB for their diagnosis and classification.

Conclusions
MPMRI is cost effective as the first test for the diagnosis of prostate cancer, when followed by an MRI-targeted TRUSB in men in whom the MPMRI suggests a suspicion for CS cancer, and a second TRUSB if no CS cancer is found, under the most sensitive CS cancer definitions and cut-offs. These findings are sensitive to the cost of each test, sensitivity of MRI-targeted TRUSB, and long-term outcomes of men with cancer, which warrant more empirical research.
Author contributions: Rita Faria had full access to all the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.
Study concept and design: All authors.
Acquisition of data: All authors.
Analysis and interpretation of data: All authors.
Drafting of the manuscript: Faria.
Critical revision of the manuscript for important intellectual content: All authors.
Other: None. E U R O P E A N U R O L O G Y X X X ( 2 0 17 ) X X X -X X X