Risk of Clinically Significant Prostate Cancer after a Nonsuspicious Prostate MRI—A Comparison with the General Population

The benefits of multiparametric MRI of the prostate (mpMRI) with Prostate Imaging-Reporting and Data System (PI-RADS) (1) scoring have improved the specificity for the detection of clinically significant prostate cancer [csPCa; i.e., Gleason ISUP grade group (GG) 2–5] with little to no loss of sensitivity (2–8). MRI-targeted biopsies detect more csPCa and less clinically insignificant prostate cancer (isPCa; i.e., GG 1) than transrectal ultrasound-guided systematic biopsies (SB; refs. 3, 6, 8). Hence, the European Association of Urology (EAU; ref. 7) and National Institute for Health and Care Excellence (NICE; ref. 9) guidelines on the diagnosis of prostate cancer recommend mpMRI before biopsy.

However, how to proceed after a nonsuspicious mpMRI (nMRI) with a PI-RADS score of 1–2 or no lesions is controversial. The pooled estimate for the negative predictive value (NPV) of nMRI for csPCa was 91% in a recent meta-analysis (10). The NPV could be enhanced with additional SBs, but this would also increase the rate of overdiagnosis. Another recent meta-analysis (6) showed that for 100 biopsy-naïve men with nMRI, SBs would result in an average of 8 and 18 additional diagnoses of csPCa and isPCa, respectively. As the net benefit of supplemental SBs is equivocal, the PI-RADS Steering Committee (11) recommends monitoring (such as PSA monitoring and follow-up imaging) instead of immediate SB after nMRI if clinical variables such as PSA density (PSAD) indicate a low likelihood of csPCa. However, each patient and their clinician need to individually weigh the potential benefits (detection of csPCa) and disadvantages (unnecessary biopsies and burden of surveillance, as well as overdiagnosis of isPCa) of such approach.

Because the benefits (possible small reduction in prostate cancer mortality but not in overall mortality) of prostate cancer screening using PSA (and subsequent SBs) have not been shown to surpass the harms (biopsy and treatment-related complications and overdiagnosis; ref. 12), the general population presents a suitable reference for assessing the risk of prostate cancer in men with nMRI. Therefore, the primary aim of this study was to estimate the incidence rate (IR) of csPCa in men after nMRI in comparison with the general population. The secondary aim was to assess the degree of overdiagnosis (isPCa) and the impact of PSAD.

In this retrospective population-based cohort study, we used the radiology and pathology databases of the Helsinki University Hospital District (HUS) and the National Population Registry Data of Digital and Population Data Services Agency provided by Statistics Finland (StatFin database, RRID:SCR_024843) to form two cohorts. First, a general population cohort consisting of all men residing in the county of Uusimaa (population of 1.7 million) who had not undergone mpMRI and had not been diagnosed with prostate cancer prior to January 1, 2016. Second, an nMRI cohort consisting of all men without prior diagnosis of prostate cancer who underwent a primary mpMRI at our institution with a nonsuspicious (PI-RADS 1–2 or no lesions) result during the inclusion period between January 1, 2016 and May 29, 2019 (Fig. 1). Participation was allowed in both cohorts. The follow-up started at January 1, 2016 for general population cohort and at the time of mpMRI for nMRI cohort. Only men aged 50–79 years at the start of follow-up were included. We used pathology and population registries to classify cases as csPCa (GG ≥ 2) and isPCa (GG 1) and estimated their incidence in the two cohorts. We also utilized an institutional laboratory database to calculate PSAD in the nMRI cohort and obtained mortality data for all included men from the Causes of Death Registry of Statistics Finland.

For the general population, the onset of follow-up was January 1, 2016 (the day after the first update of the population registry after PI-RADS version 2 was introduced locally). For the nMRI cohort, follow-up started on the date of primary mpMRI. Follow-up ended at death, diagnosis of csPCa, emigration (general population cohort only), or at the end of the study period on the 31^st of December 2021 (date of the latest mortality data update), whichever occurred first. The process for calculating the follow-up time for the general population is shown in Supplementary Fig. S1.

All included mpMRIs were obtained and assessed following the PI-RADS version 2.0 (ref. 1; until April 2019) or 2.1 (ref. 13; after April 2019) criteria. The MRI protocol and interobserver agreement for radiologists have been described in detail previously (14, 15). Briefly, MRIs consisted of T2-weighted imaging, DWI with ADC maps, and dynamic contrast enhancement (DCE) obtained using 3-T magnets with a phased-array coil and no endorectal coil (Supplementary Table S1). Images were visually assessed and analyzed using DynaCad software (Invivo Corporation) by institutional radiologists, according to routine clinical procedures. The institutional database was searched for all reports of mpMRIs performed during the inclusion period. Semistructured radiology reports were categorized using custom Python scripts (J. Hoffström), and data on PI-RADS score, prostate volume, technical quality of the study, previous PSMA-PET study, and previous diagnosis or treatment of prostate cancer were collected. Ambiguous reports were reviewed and resolved manually by a radiologist (J. Pylväläinen) with six years of experience in radiology.

As depicted in Fig. 1, men with a previous diagnosis of any-grade prostate cancer, previous institutional PSMA-PET, or mpMRI were excluded. mpMRIs of nondiagnostic quality were excluded. Finally, men with nonsuspicious mpMRI findings were included in the nMRI cohort.

All institutional pathology reports related to the prostate from January 2000 to December 2021 were categorized using custom R scripts (to J. Pylväläinen.). Systematized Nomenclature of Medicine (SNOMED) topography coding and organ-related keywords were used to verify their association with the prostate. All prostate tissue samples (e.g., TURP, autopsy, prostate biopsy, and radical prostatectomy) were included to achieve maximal coverage.

Although the recommendation during the study period was to use mpMRI as a triage test before prostate biopsy in all men with suspected PCa, the decision on the diagnostic work-up was at the discretion of the treating urologists. Patients with a positive mpMRI underwent MRI-targeted biopsy using a transrectal ultrasound-MRI fusion biopsy system (UroNav, Philips Healthcare) with 2 to 4 cores per lesion. Concomitant systematic biopsies were not routinely performed to avoid overdiagnosis. All prostate tissue samples were reviewed according to the International Society of Urological Pathology 2014 consensus (16) by institutional pathologists following standard clinical practice.

The dichotomous any-grade prostate cancer status (absence vs. presence) and tumor features (GG) of each pathology sample were extracted from pathology reports. The prostate cancer status at the end of follow-up was based on the highest grade of cancer diagnosed during follow-up. GG ≥ 2 prostate cancer was considered csPCa and GG 1 as isPCa. Positive samples without assigned Gleason scores or grades were excluded from further analyses (Table 1).

All laboratory samples obtained in the public health care within the hospital district are analyzed in HUS hospital laboratory services (HUSLAB). The PSA values were collected from the HUSLAB database between November 1998 and May 2020. PSAD was calculated by dividing the latest PSA less than one year prior to or 30 days after mpMRI by the prostate volume at the primary mpMRI.

We calculated the age-standardized IRs and incidence rate ratios (IRR) of csPCa and isPCa. For age standardization of the IRs, the age distribution of the nMRI cohort was used as the standard. For the IRR calculations, the general population was used as the reference population (i.e., IR in the general population was used as the denominator). Analyses were conducted for the entire population (50–79 years) and for 10-year age groups (50–59, 60–69, and 70–79 years). A two-sided 5% statistical significance level was used for all analyses. 95% confidence intervals (95% CI) of age-standardized IRs and IRRs were calculated using the formula presented by Tiwari and colleagues (17) For subgroup analysis, a commonly (18–20) used PSAD threshold of 0.15 ng/mL/cm³ was used. The age-adjusted ORs with 95% CIs of a prostate cancer diagnosis being clinically insignificant were calculated using logistic regression (with isPCa as the cases and csPCa as controls), and χ² test of association was used to assess statistical significance. We conducted a sensitivity analysis to investigate how categorizing all lesions with missing Gleason grades as csPCa or isPCa would impact the crude incidence rates of prostate cancer. All analyses were performed using Stata Statistical Software Release 16 (Stata, RRID:SCR_012763) or R Statistical Software 2022.07.1 Build 554 for MacOS (R Project for Statistical Computing, RRID:SCR_001905).

The data generated in this study are available upon request from the corresponding author.

At baseline, 787,495 men resided in the Uusimaa region, of whom 239,039 were aged 50 to 79 years. After excluding men with a previous prostate cancer diagnosis or mpMRI, 230,458 men with a median age of 61 years [interquartile range (IQR) 55–68] formed the general population cohort. A total of 6,621 mpMRIs, of which 5,709 were primary mpMRIs, were performed during the study period. Finally, 1,682 men aged 50 to 79 years [median age 66 years (IQR 60–71 years)] with their first MRI being an nMRI were included in the analysis (Fig. 1). At the time of the nMRI, 1,066 (63.4%) men were biopsy-naïve, and 616 (36.6%) had at least one previous negative biopsy. PSA values preceding nMRI were available for 1,459 (86.7%) men. Prostate size was recorded in 1,663 (98.9%) reports. Consequently, we were able to calculate PSAD for 1,442 (85.7%) men, and of them, 920 (63.4%) had a PSAD < 0.15 ng/mL/cm³. Demographic details are presented in Supplementary Table S2.

The median follow-up time in the general population cohort was 6.0 years (IQR, 6.0–6.0 years), among all men with nMRI 3.9 years (IQR, 3.1–4.7 years), and in the nMRI men with low PSAD 3.9 years (IQR, 3.2–4.7 years). The cumulative incidence of csPCa for the 3,264 men that underwent mpMRI with any result was 1,180 (36.2%). The cumulative incidence of csPCa after mpMRI with any result increased with age (27.1% for 50–59, 33.7% for 60–69, and 45.1% for 70–79-year-old). The detailed follow-up results for the general population and nMRI cohorts are presented in Tables 1 and 2. The age-standardized IRs of csPCa after nMRI were 552 (95% CI, 537–567), 1,879 (95% CI, 1,556–2,250), and 752 (95% CI 497–1,092) per 100,000 person-years, respectively, and the age-adjusted IRs of isPCa were 74 (95% CI, 68–80), 606 (95% CI, 431–828), and 528 (95% CI, 316–829) per 100,000 person-years. Compared with the general population, the absolute risk increases of csPCa and isPCa were 1,327 and 532 per 100,000 person-years for the nMRI cohort and 200 and 454 per 100,000 person-years for men with nMRI and low PSAD, respectively.

Figure 2 shows the differences in age-standardized IRs of csPCa among all cohorts. The age-standardized IRR of csPCa in the nMRI cohort relative to the general population cohort was 3.4 (95% CI, 2.8–4.1). The age-standardized IRR was substantially higher for the age group 50–59 years [10.4 (95% CI, 7.3–14.5)] than for those aged 60–69 and 70–79 years [IRR, 2.7 (95% CI, 2.0–3.7) and 2.5 (95% CI, 1.8–3.5); Table 2]. The relative risk of csPCa after nMRI was substantially lower among men with a low PSAD [IRR, 1.4 (95% CI, 0.9–2.0)].

In a sensitivity analysis we found that categorizing all lesions with missing Gleason grade as csPCa would increase the unadjusted IRR by approximately 5.6% (4.5 vs. 4.34), while categorizing them as isPCa would decrease the IRR of isPCa by approximately 0.4% (10.48 vs. 10.52).

The age-standardized IRR of isPCa for the nMRI cohort relative to the general population was 8.2 (95% CI, 5.8–11.3). Compared with the IRRs of csPCa, the IRRs of isPCa were substantially higher across all age groups (Table 2).

In the general population cohort, 0.14 (95% CI, 0.13–0.15) isPCas were diagnosed per each csPCa case, whereas the odds of isPCa (vs. csPCa) were 0.33 (95% CI, 0.23–0.47) in the nMRI cohort, and 0.68 (95% CI, 0.38–1.22) if the patient also had a PSAD < 0.15 ng/mL/cm³. The ORs were 2.4 (95% CI, 1.7–3.5) and 5.0 (95% CI, 2.8–9.1), respectively, when general population cohort was used as the reference (Table 3). We found no evidence of an association between the OR and advancing age (P = 0.163).

Whether or not, and which men should undergo SB after nMRI remains debatable. In this study, we attempted to provide a broader perspective by comparing men with nMRI with men in the general population. We showed that the risk of csPCa is not negligible after nMRI. However, for men with nMRI who also had a low PSAD, the risk of csPCa approached that of the general population, while the risk for isPCa was substantially increased. Altogether, our findings do not support routine systematic biopsy of all men with nMRI but instead suggest the use of risk stratification with PSAD to reduce overdiagnosis.

The age-standardized IRR of csPCa after nMRI was much higher for the 50- to 59-year-old men than for older age groups. This is largely explained by the differences in cumulative incidence of csPCa in the general population between the age groups, whereas the differences in csPCa incidence after nMRI were marginal in our study (Table 2). Men selected to undergo mpMRI are suspected of harboring csPCa based on a clinical evaluation, such as PSA testing or digital rectal examination, which already reduces the heterogeneity of subsequent csPCa risk among the age groups. However, discrepancies in the cumulative incidence of csPCa after mpMRI and nMRI among the age groups (27% and 8.4% for ages 50 to 59 years vs. 34% and 5.9% for ages 60 to 69 years, accordingly) lead us to perform a post hoc analysis, that revealed lower sensitivity of mpMRI for 50- to 59–year-old men (81%) than for the ages 60 to 69 (91%) and 70 to 79 years (93%). Similar findings have been reported in previous literature (21) and it has been suggested that age-dependent changes in the architecture of prostate may hinder interpretation of mpMRI in young (21, 22). Nevertheless, the age-related variation of the age-standardized IRR of csPCa after nMRI was significantly reduced for men with low PSAD. Altogether, these results suggest that the implications and requirements of mpMRI are not the same for the young and old, while risk stratification using PSAD seems to perform well for men of all ages.

Ultimately, the acceptable level and cost of csPCa risk reduction are determined individually. However, the definition of csPCa is vague and there is no consensus on its histopathological characteristics. The threshold of GG ≥ 2 for csPCa is commonly used but has been criticized as MRI-targeted biopsy may result in grade inflation (23). Recently, a post hoc analysis (24) of the PROMIS trial was performed, and the authors established the attributes of cancers missed by prostate mpMRI. Although the authors used template mapping biopsy as the reference standard and applied various criteria for clinical significance, they concluded that tumor visibility on prostate mpMRI may confer prognostic information that cannot be captured by the histopathologic grade of the lesion alone. Instead, a long-term follow-up is necessary to determine the ultimate clinical outcomes. As the PI-RADS recommendation was only published in 2012, such long-term follow-up studies with clinically relevant mortality endpoints after nMRI are unlikely to be available soon. Therefore, studies, such as ours, with surrogate endpoints are needed to help clinicians make informed decisions. The long-term follow-up data gathered from PSA screening studies indicate that routine testing of low-risk populations may not reduce all-cause mortality and may cause more harm than good (12). The current consensus is not to perform PSA- and SB-based prostate cancer screening due to its limited benefits and increased rate of overdiagnosis. In our study, men with low PSAD who underwent nMRI had an IRR of 1.4 (absolute risk increase of 200 per 100,000 person-years) for csPCa and 7.1 (absolute risk increase of 454 per 100,000 person-years) for isPCa compared with the general population cohort. In addition, none of the 920 men in the cohort died of prostate cancer, whereas 2.9% of the men died from other causes between 2016–2020. As the men with nMRI and low PSAD carry a significantly increased risk of overdiagnosis and only a moderately increased risk of csPCa compared with general population, it's doubtful that the benefits of routine biopsies could overcome the harm. However, this needs to be assessed more carefully when long-term mortality data is available.

This study is the first to compare prostate cancer incidence in men after a nMRI with the general population. The observed 93% NPV of mpMRI at median follow-up of 3.9 years is well within the wide range (85%–97%) of previously reported NPVs of mpMRI to exclude GG ≥ 2 prostate cancer in studies (5, 18–20, 25) that have used SBs and template biopsies as a reference. However, the assessment of the true NPV or prevalence was not the objective of our study. Only 31.4% of patients with an nMRI and 5.0% of men in general population underwent a biopsy during the follow-up. More importantly, our results are in line with the NPV of mpMRI in a clinical follow-up study (26) at two years (95%–96%). On the basis of the Swedish population (Statistics Sweden, RRID:SCR_024842) and incidence statistics (National Prostate Cancer Register of Sweden, RRID:SCR_024844) of men aged 50 to 79, we estimate that the IR of csPCa in Sweden was 420 per 100,000 men per year between 2016 and 2021, which is almost identical to the IR of 426 in our study, further supporting the validity of our methodology. Since the Finnish Cancer Registry (FCR), with 96% coverage of solid tumors, does not provide information on prostate cancer grading, in-depth comparison of our findings with the national registry was not possible (27). However, during the study period (2016–2021), the number of cases of any prostate cancer among 50- to 79-year-old men in Uusimaa was 6% higher (5,015 vs. 4,712) in our study than reported by FCR (Finnish Cancer Registry, RRID:SCR_005881). The difference may be due to either imperfect coverage and delays in compiling the FCR data or overestimation of IR in the general population in this study (e.g., by inclusion of men residing outside the catchment area).

Men with nMRI had higher odds of isPCa than men in the general population, suggesting an increased risk of overdiagnosis after nMRI. While 14 isPCas were found per 100 csPCas in the general population, the ratio was 33 isPCas per 100 csPCas among men with nMRI (age-adjusted OR 2.4). These odds are somewhat lower than those reported in the literature. According to the Swedish population registry (Statistics Sweden, RRID:SCR_024842) and the National Prostate Cancer Registry (National Prostate Cancer Register of Sweden, RRID:SCR_024844), 33 isPCas were diagnosed per 100 csPCas in Sweden in 2016–2021. In the control arm of the Finnish Randomized Study of Screening (FinRSPC; ref. 28), 34 isPCas were diagnosed per 100 csPCas between 1996–2015. In recent studies (5, 18–20, 25, 26) evaluating the NPV of MRI, the ratio of isPCa to csPCa has varied between 0.17 and 6.3. Hence, the isPCa-to-csPCa ratio in men with a previous nMRI in our study lies at the lower end of the previously reported range. This likely reflects the fact that during the study period, the local recommendation for all men with prostate cancer suspicion was to use a prebiopsy MRI, to perform targeted biopsies of only suspicious lesions (PI-RADS 3–5), and not to biopsy men with a low likelihood of prostate cancer, typically low PSAD and normal DRE, minimizing the risk of overdiagnosis. We found no previous studies estimating the relative risk of overdiagnosis in men with a previous nMRI and the general population. However, the proportion of isPCa is generally (18–20) higher in patients with nMRI than in those with positive mpMRI findings, supporting our findings.

The current study has some limitations. This is a retrospective registry-based analysis making it susceptible to potential biases. Misclassification due to incorrect coding was possible. However, in a post hoc intrinsic validation, the rate of misclassification was marginal (0.2%–1.9%). Also, for the nMRI population, we were unable to ascertain loss to follow-up (attrition) because of emigration from the study region. Some men may have undergone MRI and biopsy in the private sector or in other hospital districts. Reflecting the real-life nature of the study, the criteria for mpMRI, biopsy, and surveillance were not rigorously defined. PSA values were not found for a small fraction of the patients, which may be related to the tendency to measure PSA in the private sector prior to referral to public health care. In addition, PSA values obtained up to one year before MRI may not represent the accurate level at the time of MRI. In addition, we were unable to account for the use of 5-alpha-reductase inhibitors. The follow-up time was relatively short, but no nMRI cohorts with long-term follow-up have been published to date and are likely unavailable for some time. No routine biopsies or repeat MRIs were performed after nMRI, but instead was at the discretion of treating urologists.

The major strengths of the study are the population-based analysis in a setting with comprehensive public health care and a single tertiary referral center, usage of contemporary PI-RADS version 2 classification, large sample size, and a comprehensive, real-life setting. As the estimates of all groups were derived from the same data sources, inaccuracies are likely to be nondifferential and hence dilute any differences, which suggests underestimation of the differences (and, in this sense, strengthening the observed contrasts). As routinely collected registries were used, selection and information biases were likely minimal, and the comparability of data across groups and completeness of coverage are high.

Our study should be interpreted in the light of the fact that our methodology was designed to maximize the accuracy of identifying men diagnosed with isPCa and csPCa in daily clinical routines. The histopathologic samples assessed were not limited to needle biopsies of the prostate, but also included alternative samples such as TURP and prostatectomy specimens. In addition, we are aware that our methodology inherently induces selection bias, as men who undergo nMRI remain in clinical monitoring, while men in the general population are scarcely examined. Hence, it is probable that the incidence of both csPCa and isPCa was underestimated more in the general population cohort than in the nMRI cohort. The population cohort represents the unexamined “healthy” individuals who are unaware of their prostate cancer status, who, as per the current consensus, are also deemed to possess prostate cancer risk that does not necessitate prostate cancer screening. The nMRI population on the other hand presents men who, after initial clinical assessment, were found to have an increased risk of prostate cancer, which was downgraded to a level of uncertain clinical significance after nMRI. These men tend to undergo surveillance and additional biopsies without proof of benefit. Our study aimed to compare the prostate cancer risk profiles of these two populations and highlight their similar trade-offs.

Finally, we believe that routine SB would cause more harm (overdiagnosis) than benefit (csPCa diagnosis). This hypothesis will be tested in our ongoing randomized population-based screening trial (ProScreen; ref. 15), the only MRI-based screening trial powered for PCa mortality. In the trial, 115,000 men are randomized to the screening (PSA – 4Kscore – MRI) and control groups. Importantly, men with a negative MRI result will not be biopsied and are instead systematically followed with repeated screening rounds.

In conclusion, this study provides valuable insights into prostate cancer risk after nMRI in relation to the general population. The incidence of csPCa in not negligible after nMRI. However, for men with nMRI and PSAD < 0.15 ng/mL/cm³, it is difficult to justify routine prostate biopsies as the benefit-to-harm ratio shifts toward harm. Thus, our findings do not support routine systematic biopsies of all men with nMRI. These findings should be considered in shared decision-making regarding follow-up approach after nMRI.

A. Rannikko reports grants from Jane and Aatos Erkko Foundation, Cancer Foundation Finland, and grants from Competitive State Research Funding during the conduct of the study; personal fees from Bayer, Janssen Pharmaceuticals, Orion Corporation; personal fees and other support from Ida Montin Foundation, and personal fees and other support from Orion Research Foundation outside the submitted work. No disclosures were reported by the other authors.

J. Pylväläinen: Conceptualization, resources, data curation, software, formal analysis, investigation, visualization, methodology, writing–original draft. J. Hoffström: Resources, data curation, software, writing–review and editing. A. Kenttämies: Resources, data curation, writing–review and editing. A. Auvinen: Formal analysis, validation, investigation, methodology, writing–review and editing. T. Mirtti: Conceptualization, resources, data curation, project administration, writing–review and editing. A. Rannikko: Conceptualization, resources, data curation, supervision, funding acquisition, validation, methodology, project administration, writing–review and editing.

This work was supported in part by a grant from the Cancer Foundation Finland (to A. Rannikko; Grant number 180141), Competitive State Research Funding (VTR) administered by HUS Helsinki University Hospital (to A. Rannikko; grant number TYH2021332) and the Jane and Aatos Erkko Foundation (to A. Rannikko on May 29, 2020). The study was conducted in compliance with the STROBE guidelines for observational studies. This study was performed under an institutional research permit (HUS/333/2019). As the study was registry-based, no ethical committee approval was required. This study was conducted in compliance with the good research practices of the World Medical Association Declaration of Helsinki. As the study was registry-based and no individual data were presented, consent to participate was not required. The data were handled in accordance with national laws and EU regulations. A related abstract with title “Incidence of clinically significant prostate cancer after negative prostate MRI - comparison to general population” has been previously presented at the AACR Annual Meeting 2021 and Nordisk Urologisk Forening (NUF) 2022.