The utility of magnetic resonance imaging in prostate cancer diagnosis in the Australian setting

Abstract Objectives To investigate the utility of Magnetic Resonance Imaging (MRI) for prostate cancer diagnosis in the Australian setting. Patients and methods All consecutive men who underwent a prostate biopsy (transperineal or transrectal) at Royal Melbourne Hospital between July 2017 to June 2019 were included, totalling 332 patients. Data were retrospectively collected from patient records. For each individual patient, the risk of prostate cancer diagnosis at biopsy based on clinical findings was determined using the European Randomized study of Screening for Prostate Cancer (ERSPC) risk calculator, with and without incorporation of MRI findings. Results MRI has good diagnostic accuracy for clinically significant prostate cancer. A PI‐RADS 2 or lower finding has a negative predictive value of 96% for clinically significant cancer, and a PI‐RADS 3, 4 or 5 MRI scan has a sensitivity of 93%. However, MRI has a false negative rate of 6.5% overall for clinically significant prostate cancers. Pre‐ biopsy MRI may reduce the number of unnecessary biopsies, as up to 50.0% of negative or ISUP1 biopsies have MRI PI‐RADS 2 or lower. Incorporation of MRI findings into the ERSPC calculator improved predictive performance for all prostate cancer diagnoses (AUC 0.77 vs 0.71, P = .04), but not for clinically significant cancer (AUC 0.89 vs 0.87, P = .37). Conclusion MRI has good sensitivity and negative predictive value for clinically significant prostate cancers. It is useful as a pre‐biopsy tool and can be used to significantly reduce the number of unnecessary prostate biopsies. However, MRI does not significantly improve risk predictions for clinically significant cancers when incorporated into the ERSPC risk calculator.


| INTRODUC TI ON
Prostate cancer is a prevalent disease, and case finding for clinically significant prostate cancer is an important population health focus. In Australia, this involves the use of a clinical evaluation including family history and digital rectal examination, followed by Prostate Specific Antigen (PSA) testing in the community setting. 1 However, PSA lacks specificity as it is commonly elevated in many benign conditions and may lead to unnecessary, invasive biopsies.
This leads to cancer overdiagnosis particularly clinically indolent ones. There is an estimated overdiagnosis of 41% in Australia, 43% in Europe and 42% in the United States. 2-4 Well differentiated cancers with low metastatic potential carry minimal risk of symptomatic progression or mortality, but their detection may lead to overtreatment with radical surgery or radiation, which in turns comes with risks such as urinary incontinence and erectile dysfunction. 5 The challenge hence lies in improving our current evaluation strategies to be able to distinguish clinically significant from indolent cancer.
Magnetic Resonance Imaging (MRI) was first used in 1982 to study prostate cancer. As its accuracy improved with technical enhancements and the addition of functional parameters, multiparametric MRI has been increasingly used in the assessment of prostate cancer risk. Prostate Imaging Reporting and Data System (PI-RADS) V2 was established to grade MRI findings from 1 to 5, correlating to the likelihood of clinically significant cancer. 6 Several studies have validated the use of MRI in prostate cancer diagnosis, with good sensitivity and negative predictive value for clinically significant disease. [7][8][9] The lack of specificity of PSA as well as the risk of overdiagnosis of indolent cancers, has led to the development of personalized clinical prostate cancer risk calculators, which increasingly incorporate the findings of pre-biopsy prostate MRI. The European Risk Surveillance for Prostate Cancer (ERSPC) is one such risk calculator that generates two risk predictions, one for all prostate cancers and one for clinically significant cancers, based on information such as age, DRE, PSA, prostate volume, and previous negative biopsies.
Clinically significant cancer is defined as having a Gleason score equal or greater than 7. 10 Based on the RCT by Roobol et al, the risk calculator recommends that a risk prediction less than 12.5% does not require a biopsy, and a risk of more than 20% warrants a definite biopsy. 11 For percentages in between, the decision to biopsy depends on co-morbidities and if the risk for clinically significant cancer was more than 4%. In an international study involving ten cohorts in Europe and Australia, ERSPC was shown to have the highest predictive accuracy, in comparison to other risk calculators such as Sunnybrook and Prostate Cancer Prevention Trial (PCPT) risk calculator. 12 Furthermore, there have been improvements with a newer edition that allows the incorporation of MRI results in recent years.
Australia has the highest incidence of prostate cancer diagnoses globally, and as such it is plausible that findings from international studies may be less applicable in our local context. Limited data from a small number of studies suggest that the performance of the test is similar to that observed internationally, with the presence of an MRI abnormality being significantly associated with the presence of clinically significant disease. [13][14][15] However, these studies were performed prior to MRI being universally funded, and so estimates of performance may be biased by patient selection. Since the new Medicare Benefit Schedule was rolled out in Australia 2018, prostate MRI has become rebatable if the patient met eligibility criteria, such as having positive DRE, two serial elevated PSA and a free-tototal PSA ratio of less than 25%, or undergoing active surveillance. 16 With the increasing use of MRI in the Australian community, it is vital to study and understand the utility of MRI. This will impact how we can appropriately incorporate these findings into the risk stratification of our prostate cancer patients. In this study we evaluated the performance of prostate MRI in the pre-diagnostic setting in patients meeting funding current criteria, and asses its incremental diagnostic utility when incorporated into the established ERSPC clinical risk calculator. These data were de-identified and entered into a secure electronic database. This study was approved by the Melbourne Health Research Ethics Committee (approval no. QA2020011).

| Imaging
Prostate multiparametric MRI scans were performed using 3-Tesla magnets Siemens Prisma and Skyra MRIs. T1-and T2-weighted imaging, diffusion-weighted imaging and apparent-diffusion coefficient functional sequences were acquired and processed with Fujifilm Synapse PACS. Images were reported using the PI-RADS v2 criteria by | radiologists with expertise in MRI and body imaging. In our study, data were collected from 2017, a year after PI-RADS v2 was adopted locally in Royal Melbourne Hospital to minimize the initial learning curve.

| ERSPC risk calculator
The 3rd and 4th versions of the ERSPC calculator allow for the incorporation of MRI results. 10

| Statistical analysis
All data were analyzed and presented with simple descriptive statistics. Categorical data were compared using chi-square test and continuous variables using the student t-test. The MedCalc statistical software was used to calculate and generate the Receiver Operator Curve (ROC). Thereafter, Area Under Curve was calculated using the Hanley and McNeil algorithm. All statistical tests were 2-sided with P < .05 considered statistically significant.

| RE SULTS
In total, 332 patients were identified, the clinical, radiological and pathological characteristics of whom are summarized in Table 1. The average patient in our study was 64.6 years old at the date of biopsy, with a PSA of 6.6 ng/mL and a prostate volume of 40 cc. The majority of patients had a normal DRE, no family history of prostate cancer, and no previous biopsy. Sixty-two percent of the patients had an MRI scan of their prostate, of which 40.3% of the scans were PI-RADS 2 or lower and 59.7% were PI-RADS 3 and above. The majority of patients underwent TP biopsy. There were almost equal proportions of benign and malignant biopsy results; 51.0% negative and 49.0% cancers. Of the cancers diagnosed, 41.1% were ISUP grade group 1, 25.8% were ISUP grade group 2 and 33.1% were ISUP grade group 3 and above.
Metastatic disease was detected in 6.7% of the cohort. In terms of management, 41.1% had active surveillance or watchful waiting and 7.4% had palliative hormonal and/or chemotherapy. The remainder were actively treated, of which 62.7% underwent radical prostatectomy and 37.3% received radiotherapy and/or hormonal therapy.
Of the patients who underwent radical prostatectomy, 3.9%, 53.9% and 42.2% were found to have ISUP1, ISUP2 and ISUP 3 or above, respectively, on final pathological review. Patients without MRI were significantly older, with higher PSAs and higher rates of high grade and metastatic disease, likely reflecting the patient factors that contribute to investigating with an MRI, rather than any actual clinically significant difference between the 2 groups.
MRI PI-RADS grading corresponded well with the biopsy results, especially with clinically significant cancers. The proportions of patients with each PI-RADS and ISUP grade are summarized in had PI-RADS 3 and above lesions, and 6.5% of them had PI-RADS 2 or lower. These 6.5% of clinically significant cancers were all ISUP 2 cancers, with less than 5% pattern 4 disease and an average maximum cancer core length of 6 mm. Figure 1 depicts the distribution of ERSPC prediction risk scores in our cohort, where significant skewing to lower risk scores was observed. We found that 29.5% of our patients had a risk of less than 12.5% for all prostate cancers, where a prostate biopsy is not recommended by the studies that validated the risk calculator. Overall, 51.9% of prostate biopsy results had the same ISUP grades as the final prostatectomy histopathology, as seen in Table 3.
Of the remaining discordant results, 72.0% had upgrading. From our study, the use of pre-biopsy MRI does not appear to increase the concordance of the biopsy results and final histopathology (P = .19).
However, the sample is small and larger numbers will be needed to reliably interpret this finding.

| D ISCUSS I ON
There is limited data on the use of MRI in the screening and diagnosis of prostate cancer in Australia. Our results have shown that MRI has good diagnostic accuracy for clinically significant prostate cancer. A PI-RADS 2 or lower finding has a high negative predictive value of 96% for clinically significant cancer and a PI-RADS 3 or above lesion has a sensitivity of 93%. This is similar to the findings from other international studies which ranges from 79%-100% and 76%-100%, respectively. [7][8][9] MRI may reduce the number of unnecessary biopsies. In our retrospective study, 50.0% of ISUP 1 or negative biopsies had PI-RADS 2 or below findings, which reflects 38.8% of the total biopsies. This is similarly reflected with the ERSPC risk predictions, where 29.5% of our cohort would be recommended not to proceed to biopsy. If MRI was used universally as a risk stratification tool, the reduction of around a third of biopsy procedures is of great significance in reducing healthcare spending in terms of the procedural cost, perioperative investigations and clinic reviews. It can also reduce the waiting lists to ensure better and more timely allocation of services and reduce side effects from biopsies and unwarranted patient anxiety over false positive results.
There remains concern that MRI can miss some clinically significant prostate cancers. In our study, 6.5% of clinically significant prostate cancers had PI-RADS 2 findings or lower, which is similar to the PROMIS study, where there were 11% MRI-negative clinically significant cancers found on TP biopsy. 9 However, in our study, their prediction accuracy in ten independent cohorts, of which nine are European and one is Australian. 12 No specific analysis was done for the independent cohorts, and hence the data for validation is not specific for Australia.
As seen in Figure 1, our cohort is skewed towards having low calculated risk scores, with almost a third of patients having risk scores less than 12.5%. This could be attributed to the use of freeto-total PSA ratio in the Australian context, as this is one of the pre-requisite tests to obtaining a Medicare-funded MRI. 16 A ratio of less than 25% was associated with an increased risk of prostate cancer, but the risk is still low at 16%-20% for a free-to-total ratio from 15% to 25%. 24 The regular monitoring of this ratio could have Our study has some limitations. Firstly, the retrospective nature of this study precludes a standardized and more robust follow-up.
There was also incomplete incorporation of MRI data into the algorithm as not every patient had MRI scans. A prospective and randomized trial would also be able to provide more control and more direct analysis without biases. Our study was based on screening assessment of a patient population from a single institution and the findings may not be consistent with other worldwide patient populations or institutions. As a referral centre, many of the patients also who attended RMH for biopsies ended up having their definitive treatment in other centres, hospitals or private practices, which contributed to the paucity of follow up data. A larger sample size will also be more statistically powered to give stronger recommendations. The utility of multiparametric MRI could also be evaluated further with the use of current fusion technology with other imaging modalities. Future studies could also analyze the utility of MRI in the different subgroups such as a post biopsy surveillance tool or for a biopsy-naive patient.