For about three decades, active surveillance (AS) has been the preferred management for favorable-risk prostate cancer (PC) and has traditionally included confirmatory biopsy to address any under sampling at diagnosis.

Recent guidelines, however, have suggested that magnetic resonance imaging (MRI) might replace confirmatory biopsy to do that. A new study led by the San Francisco VAMC raised concerns that not enough data exists to confirm MRI’s accuracy, however.

In a recent research letter published in JAMA Oncology, a study team that also included VA researchers in Boston; Los Angeles; Durham, NC; San Diego; New York and Washington, DC, as well as related academic institutions, questioned whether existing research supported that change.¹

The researchers cited a 2020 study from the University of California San Francisco which found that, at confirmatory biopsy, the negative predictive value (NPV) of MRI for grade group (GG) 2 or higher PC was 74% and only 57% given PSA density (PSAD) greater than 0.15 ng/mL. To test those findings, they used the records of a larger cohort of U.S. veterans, taking into account clinical variation in radiology, biopsy technique and pathologic findings.

For the study, clinical variables were extracted from MRI and biopsy reports using natural language processing (NLP). Included were patients with GG 1 to 2 at diagnostic biopsy between 2013 and 2023 undergoing MRI (with Prostate Imaging-Reporting and Data System [PI-RADS] score) within 180 days prior to confirmatory biopsy and/or subsequent surveillance biopsy.

Defined as the primary outcome was GG 2 or higher on confirmatory biopsy or surveillance biopsy. PI-RADS of 3 or higher was considered positive. The study sought to determine the NPV of MRI (PI-RADS ≤2) in ruling out GG 2 or higher, while also considering sensitivity, specificity and positive predictive value.

The authors wrote that 1,901 patients had GG 1 to 2 disease at diagnosis and underwent confirmatory biopsy and/or surveillance biopsy with prior MRI. In fact, MRI preceded more biopsies over time, increasing from 0.7% in 2013 to 25.9% in 2023.

In terms of MRI accuracy in predicting GG 2 or higher, NPV was 75 overall (68-81) and for confirmatory biopsy (95% CI, 66%-82%), and 77 (95% CI, 62%-87%) for surveillance biopsy. For the 80% who had GG 1 at diagnostic biopsy, NPV was 79% (95% CI, 70%-86%) for confirmatory biopsy and 79% (95% CI, 64%-89%) for surveillance biopsy. For those with GG 2 at diagnostic biopsy, NPV was lower at an overall average of 44.

“For patients with PSAD lower than 0.15 ng/mL2, NPV was 78% (95% CI, 67%-87%) at confirmatory biopsy and 83% (95% CI, 64%-93%) at surveillance biopsy, higher than any other subgroup. NPV was lower among Black patients across subgroups,” the researchers pointed out. “In a sensitivity analysis reclassifying PI-RADS 1 to 3 as negative and PI-RADS 4 to 5 as positive, the proportion of negative scans increased, but NPVs were lower: 68% (95% CI, 65%-72%) and 72% (95% CI, 66%-78%) at confirmatory biopsy and surveillance biopsy, respectively.”

They noted that, when seeking to rule out GG 3 or higher, multiparametric MRI performed better. NPVs across many subgroups exceeded 95% when considering PI-RADS 3 to 5 to be positive and remained greater than 90% when considering PI-RADS 4 to 5 to be positive. Positive predictive values (PPVs) ranged from 20% to 30% under these definitions, they wrote.

“In this large, diverse cohort, NPV for negative MRI in ruling out GG 2 or higher PC at confirmatory biopsy was only 75%, lower than reported previously,” the authors advised. “Results at surveillance biopsy were marginally better. MRI was more reliable if diagnostic biopsy was GG 1, but in no subgroup was NPV greater than 80% except surveillance biopsy with PSAD lower than 0.15 ng/mL.”

They suggested that considering PI-RADS 3 negative would avoid more biopsies but miss more GG 2 or higher disease, adding that MRI was better at ruling out GG 3 or higher.

“Caveats should be considered,” the researchers stated. They explained that MRI and biopsy decisions were made per routine clinical decision-making, and that patients with negative MRI results might not have received a biopsy. In addition, the study could not account for MRI quality or inter-reader variability in PI-RADS assignment; the researchers posited that NPV of MRI would likely improve if MRI were performed more consistently before diagnostic biopsy, and if expert radiologists read more of the scans.

“While quality multiparametric MRI clearly can help guide prostate biopsies and improve diagnosis, our data confirm that relying on MRI as a surrogate for biopsy in AS would result in underdiagnosis of clinically significant cancers—though we acknowledge few of these would likely be imminently life-threatening. Improvements over PI-RADS clearly are in the pipeline; until they can be implemented broadly, routine confirmatory testing in AS for PC should still include biopsy,” the authors recommended.

The findings are especially significant because prostate cancer is the most common oncological diagnosis among veterans. The VA treats approximately 15,000 new cases of prostate cancer each year, and more than 200,000 veterans are currently survivors of the disease.

Table. Predictive Performance of PI-RADS 3 to 5 on Multiparametric MRI in Predicting GG 2 or Higher at Confirmatory Biopsy and Surveillance Biopsy, for All Patients and for Subgroups Stratified by PSAD at Time of Biopsy and Original GG at Diagnosis

Source: MRI Instead of Biopsy in Prostate Cancer Under Active Surveillance Appears to Result in Underdiagnosis – U.S. Medicine

References

1. Cooperberg MR, Bihn JR, Culnan JM, et al. Magnetic Resonance Imaging or Confirmatory Biopsy for Patients With Prostate Cancer Receiving Active Surveillance. JAMA Oncol. Published online November 20, 2025. doi:10.1001/jamaoncol.2025.4167
2. Chu CE, Lonergan PE, Washington SL, et al. Multiparametric magnetic resonance imaging alone is insufficient to detect grade reclassification in active surveillance for prostate cancer.  Eur Urol. 2020;78(4):515-517. doi:10.1016/j.eururo.2020.06.030