Methodological considerations in assessing the utility of imaging in early prostate cancer

INTRODUCTION

The accuracy of a transrectal ultrasound (TRUS) 10–12 cores biopsy in risk-stratifying men with clinical suspicious prostate cancer is limited due to its inherent random deployment of biopsy needles. This has led to the criticism of the entire pathway in early prostate cancer, from screening, diagnosis and treatment. Standard TRUS biopsy may missclassify men in up to 50% of the cases ^[1]; in other words, the management of men diagnosed with early disease has been driven by a test that has almost the same reliability as flipping a coin. In addition, TRUS biopsy is not a harmless test, the risk of infection and sepsis is continuously increasing over time at a rate of 1% to 2–4% in large cities ^[2]. A curious status quo from the poor diagnostic performance has arisen with criticism of both the consequent overdiagnosis and overtreatment of clinically insignificant disease, as well as underdiagnosis and undertreatment of clinically significant disease.

Much has been written about the potential value of switching to an image-driven pathway to overcome these shortcomings. This novel strategy incorporates two tests. First, multiparametric MRI (mpMRI) to detect or rule out a target, second, targeted biopsy for tissue sampling in those that have a suspicious area on imaging. mpMRI is a noninvasive test which might allow men to avoid a biopsy altogether if the probability of clinically significant disease conferred by a negative scan is low. When suspicious, a limited number of biopsies may significantly improve detection rates and risk stratification. A number of recent systematic reviews and meta-analyses have consistently shown that targeted biopsy strategies guiding needles to magnetic resonance (MR)-identified lesions have better accuracy, efficiency and utility when compared with standard TRUS biopsy protocols ^{[3▪▪,4–6]}.

In this article, we explore the methodological considerations which should be addressed to robustly assess the MR-based diagnostic pathway.

KEY POINTS IN EVALUATING A NOVEL DIAGNOSTIC TEST

Robust evaluation of a novel diagnostic test should be carried out with respect to the following key points: technical accuracy, place in the clinical pathway, diagnostic accuracy, impact on patient outcome and cost-effectiveness (Fig. 1) ^[7]. We assess the current state of the art with respect to each of these key points, except for cost-effectiveness which is discussed in a complementary article by Willis et al. (pp. 483–489) in this issue of Current Opinions in Urology.

Technical accuracy

Technical accuracy refers to the ability of a diagnostic test to provide valid information in a preclinical setting. The established mpMRI protocol includes T2-weighted, diffusion-weighted imaging (DWI) and dynamic contrast-enhanced (DCE) sequences as well as spectroscopy in some centres, based on preclinical studies suggesting the complementary and added value of these sequences ^[8–10]. The technical accuracy of mpMRI depends on the threshold for disease significance, which has not yet been established for animal models. mpMRI has been shown to be accurate in detecting clinically significant prostate cancer but not very accurate in identifying low-grade low-volume disease. Therefore, in many preclinical studies, the technical accuracy of mpMRI has been probably underestimated, at least for clinical purposes.

Place in the clinical pathway

Determining the place of a novel diagnostic test in a clinical pathway is a crucial step with three possible options available: a replacement, a triage, or an add-on test to the standard one (Fig. 2) ^[11]. In the first instance, a replacement test, the standard test is abandoned in favour of the novel one in light of better accuracy and/or decreased invasiveness and/or equal diagnostic ability with decreased overall cost. In the second instance, a triage test, the novel diagnostic test is performed before the existing one to select those patients who should undergo further assessment. Typical advantages of triage tests are simplicity, interpretability and cost. In the third instance, an add-on test, the novel diagnostic test is performed after the existing for a subgroup of patients who need further assessment. Add-on tests often help identify other characteristics of disease not diagnosed by the standard test, for example metastatic spread.

The role of mpMRI has evolved over time with an initial use as an add-on test to more recently, as a potential triage test after an individual has an elevated Prostate Specific Antigen (PSA). mpMRI was initially used in subgroups of men with known prostate cancer in which knowledge of local extension may have influenced management. In other words, after an initial positive TRUS biopsy, some men underwent mpMRI to plan surgery and/or to switch to radiation therapy in case of locally advanced disease.

At present, mpMRI is rather proposed as a triage test followed by a targeted biopsy as an add-on test or eventually as a replacement test to standard TRUS biopsy. There are various reasons for this shift (Fig. 3). First, the high performance of mpMRI in ruling out clinically significant disease makes it suitable as a triage test to select those men who need to undergo biopsy ^[12]. This would substantially decrease the number of biopsies and may therefore increase the overall diagnostic ratio of prostate biopsy protocols. However, the high negative predictive value of mpMRI with regard to clinically significant disease is not a consistent finding in the literature. A recent systematic review has shown that the negative predictive value varies from 63–98% across different series ^[13▪]. This discrepancy in the literature could be due to variable expertise of different centres but could also depend on the way the negative predictive values are calculated: for example, considering whole-gland vs. zonal level, low threshold vs. high threshold of significance, or standard vs. reference test used for comparison. At present, TRUS biopsy is still proposed in case of nonsuspicious mpMRI in most centres. Second, there seems to be little value in performing TRUS biopsy in addition to targeted biopsy for men with positive MR findings. In the latest study considering a high histological threshold of significance (Gleason ≥ 4 + 3), around 200 men would need to undergo 10–12 cores additional biopsies to detect one clinically significant disease missed by MR-targeted biopsy ^[14▪▪]. However, the overall utility of performing additional TRUS biopsy in men with positive mpMRI findings is higher with up to 1 patient over 14 missed by targeted biopsy in studies setting the histological threshold of significance at a lower level ^[3▪▪]. Third, mpMRI with subsequent histological verification of disease by accurate targeted biopsy offers an opportunity to monitor disease for those men undergoing tissue-preserving approaches; something which is difficult to achieve when using a ‘blind’ test, such as TRUS biopsy ^[15].

Diagnostic accuracy

Diagnostic accuracy relates to the ability of a test to correctly identify or exclude a given condition ^[7]. Diagnostic accuracy and main test characteristics (sensitivity, specificity, positive and negative predictive values) are usually calculated using a 2 × 2 table, in which the findings of the index test are compared with those of a reference test (Table 1). When performing diagnostic studies assessing accuracy, systematic errors and bias need to be avoided to obtain valid results.

To avoid spectrum bias, the novel test should be assessed in a relevant study population in which the test will actually be performed in the ‘real world’. Initial diagnostic studies evaluating mpMRI in men undergoing radical prostatectomy were typically limited by a spectrum bias, as they were carried out in men in whom clinically significant disease has been diagnosed or is suspected. However, recent studies have overcome this limitation by comparing the results of a targeted approach to a valid reference test – template prostate mapping biopsy – in consecutive men with suspicious disease, regardless of the ultimate treatment allocation ^[16–18]. This design has also addressed verification bias, which is present when only a subgroup of the study population undergoes the reference test.

Observer bias, in which the test is performed and/or interpreted by someone who has access to other test results, is another common error leading to systematically overestimating the performance of a novel diagnostic test. Blinding and objective interpretation of findings are ways to address this bias which have been used in some studies comparing targeted biopsy to standard protocols ^[3▪▪].

Incorporation bias, in which the results of the tests that are being compared are not independent, might have occurred in some studies, but this would have been in favour of the standard approach. Indeed, in most studies comparing MR-targeted biopsy to TRUS biopsy, the design was a paired cohort study in which both tests were performed in the same biopsy session. The operator was aware of the MR findings, and this knowledge is likely to have influenced the way in which TRUS biopsy was performed. However, some studies overcame this bias by sampling the same prostate by two different operators, one performing standard TRUS biopsy was blinded to MR findings and another operator performing targeted biopsy aware of MR results ^[3▪▪].

Impact on patient outcome

Test findings do not have a direct consequence on patient outcome, rather the test result influences subsequent management. In the area of diagnostic test for cancer diagnosis, mortality, disease-free survival, morbidity and quality of life are often used as relevant outcomes to evaluate. Due to the prolonged natural history of early prostate cancer, an imaging-driven pathway has not been compared with a standard one in terms of mortality or disease-free survival. A study designed to provide results for such an outcome would need at least 10–15 years of follow-up.

Surrogate endpoints that are likely to have a substantial impact on patient outcome have instead been used across different studies. Most studies have compared a targeted approach to a standard one with regard to the detection of clinically significant disease, assuming that only the treatment of clinically significant disease has an impact on patient survival. Consequently, a diagnostic test able to detect clinically significant prostate cancer is considered to provide patient benefit in the long-term. Despite an ongoing debate on what constitutes clinically significant disease, an improved detection rate of the targeted approach over the standard is a consistent finding of comparative diagnostic studies in the field, regardless of the threshold used to define significant disease ^[3▪▪].

In addition, most studies show a reduced detection of clinically insignificant disease of targeted biopsy ^[5]. Indeed, a majority of low-grade low-volume lesions are not seen on mpMRI, and in the centres in which biopsies are not performed in case of negative mpMRI, this leads to a decrease in the number of patients diagnosed with low-risk disease. Also, when lesions are visible, the imaging phenotype seems to well correlate with the aggressiveness of the disease. Therefore, setting a threshold for performing targeted biopsy is likely to have an important impact on the detection rate of insignificant disease ^[19]. The underdetection of clinically insignificant disease represents a beneficial outcome because avoiding the diagnosis of ‘cancer’ that has no consequence on the expected quality or duration of life leads to an obvious cascade of advantages for patients, from reducing anxiety to avoiding treatment-related harms.

Another useful endpoint used has been the efficiency of the novel test – the number of needles deployed to detect one man with clinically significant prostate cancer. Also with regard to efficiency, the results have been consistently in favour of a targeted approach ^[3▪▪]. Fewer needles deployed might result in less discomfort for patients and a reduced risk of test-related complications, although this is yet to be proven.

DISCUSSION

MR-targeted biopsy has been successfully assessed as a novel diagnostic test with regard to almost all key points from technical accuracy to impact on patient outcome. Reduced invasiveness combined with increased accuracy for ruling in and ruling out clinically significant disease provide mpMRI with the ideal attributes as a triage test if followed by targeted biopsy as a replacement test or at least an add-on test to standard TRUS biopsy.

Nevertheless, there are limitations in the literature that need to be addressed. The accuracy of a targeted approach has been reported by many expert centres, but reproducibility in less experienced and low-volume centres are unproven. An imaging-based strategy is a complex intervention requiring image acquisition, reporting and the use of imaging phenotype to guide targeted biopsy. The learning curve for each of these steps needs to be determined, and as many opportunities as possible for training need to be offered to disseminate this complex procedure. This is nothing new. The development, refinement and dissemination of the Breast Imaging Reporting and Data Systems (BI-RADS) over the last 30 years is a good example ^[20]. After the first version of the BI-RADS in the early 1980s, between 1987 and 1988, the American College of Radiology supported the development of the BI-RADS committee to standardize imaging acquisition and reporting, and at the same time launched the Mammography Accreditation Program to ensure quality across the institutions offering breast imaging ^[20]. The development of multiple opportunities for training within national and international societies, the standardization and implementation over time of the BI-RADS report, the availability of direct and distant resources for self-assessment, and the possibility of objective comparison of own results against established benchmarks of quality have led to the success of the BI-RADS program.

Standardization of targeted biopsy will help to disseminate the procedure, although at present a complete standard operating procedure is lacking. Crucial aspects such as image acquisition protocols and the systematic use of validated scores have been met with success. The five-scale Likert score and the development of the Prostate Imaging Reporting and Data Systems (PI-RADS) mimic the efforts in breast imaging ^[21,22]. However, there is wide discrepancy in the literature with respect to the radiological score warranting a biopsy, with some studies reporting targeted biopsy also for men with low likelihood of diseases. At present, there is consensus that targeted biopsy should be offered to men with imaging score 4 (likely) or 5 (extremely likely) for the presence of clinically significant disease, and avoided in men with score 1 (extremely unlikely) or 2 (unlikely), as these ranges of score have a reliable high positive and negative predictive value, respectively ^[23]. Using targeted biopsy in men with MR score 3 (equivocal) is a matter of debate, although during adoption of the technique it may be warranted to offer sampling to this subgroup of men.

The way in which targeted biopsy is performed is not standardized. The number of cores per lesion and its variation according to lesion volume is a matter for debate. An arbitrary number of cores per lesion have been determined in most series. Some advocate that the smaller the lesion, the greater the number of cores needed to provide good sampling. Others suggest that large lesions are more likely to harbour grade heterogeneity, and therefore additional cores are needed for risk attribution in this case. Although this issue will be difficult to be determined in the short term, adequate sampling with at least three cores per lesion should be mandatory, as mathematical algorithms have demonstrated there is great utility in deploying three needles per lesion ^[24].

Finally, the way in which needles are guided towards targets is another important issue. Targeted biopsy to MR-detected lesions may be carried out using three strategies: in the MR scan, by cognitive MR to TRUS registration, or by software-based MR to TRUS registration. Without direct comparative studies addressing this issue, it is not known whether one of these strategies is better than the others or whether they are all equal. Some speculate that ‘in-bore’ targeted biopsy might be more accurate as needles are directed under MR control, but this is yet to be proven, and MR biopsy is limited by cost and resource issues. Software-based MR to TRUS registration seems slightly superior to cognitive registration in initial studies, although these preliminary findings should be confirmed by larger studies powered to prove statistical significance ^[3▪▪].

Ongoing multicentre studies will provide the answers to most of these open questions. The Prostate MR Imaging Study (PROMIS), a NIHR-HTA/MRC trial evaluating mpMRI as a triage test to select those men who need further assessment is a key one. The place in the pathway, diagnostic accuracy, impact on patient outcome, as well as cost-effectiveness will be determined with high precision in this multicentre study including around 700 men with the suspicion of localized prostate cancer (abnormal DRE and/ or elevated PSA </ = 15 ng/ml) who undergo in a blinded fashion the index test (mpMRI), the standard test (TRUS biopsy) and the reference test (template prostate mapping biopsy) ^[25]. The PROMIS trial has already recruited over two-thirds of the overall sample size, and initial results will be available in 1 year. Another important study which is being set up at present is the international PRostate Evaluation for Clinically Important disease (PRECISION) study. This randomized trial will assess the findings of standard TRUS biopsy against MR-targeted biopsy in many European and North American centres. The trial setting combined with the heterogeneity in competence and technique of participating centres will further clarify the key issues of standardized techniques and reproducibility.

CONCLUSION

MR-targeted biopsy is being adopted in different centres after a stepwise evaluation starting with the definition of technical accuracy and evolving into impact on patient outcome. The findings of comparative diagnostic studies are consistently in favour of MR-targeted biopsy against standard TRUS protocols. Ongoing studies are robustly evaluating the final methodological aspects: reproducibility, place in the pathway and cost-effectiveness. In a few years, it is likely that the current ‘blind’ TRUS biopsy will be abandoned or at least implemented by mpMRI followed by targeted biopsy, in case of positive findings.

Acknowledgements

The SICPA foundation supports the ongoing fellowship and PhD programme of M. Valerio. M. Emberton and H.U. Ahmed would like to acknowledge funding from the Medical Research Council (UK), the Pelican Cancer Foundation charity, Prostate Cancer UK, St Peters Trust charity, Prostate Cancer Research Centre, the Wellcome Trust, National Institute of Health Research-Health Technology Assessment programme, and the US National Institute of Health-National Cancer Institute. M. Emberton holds Senior Investigator status with the UK National Institute of Health Research (NIHR). M. Emberton receives funding in part from the UK UCLH/UCL NIHR Comprehensive Biomedical Research Centre.

Financial support and sponsorship

None.

Conflicts of interest

M. Valerio has received funding for conference attendance from Geoscan Medical. M. Emberton and H.U. Ahmed receive funding from USHIFU, GSK, AngioDynamics and Advanced Medical Diagnostics for clinical trials. M. Emberton is a paid consultant to AngioDynamics, Steba Biotech and SonaCare Medical (previously called USHIFU). Both have previously received consultancy payments from Oncura/GE Healthcare and Steba Biotech. None of these sources had any input whatsoever into this article.

REFERENCES AND RECOMMENDED READING

Papers of particular interest, published within the annual period of review, have been highlighted as:

• ▪ of special interest
• ▪▪ of outstanding interest