Jonathan Smith *;
Leeds Teaching Hospitals NHS Trust
Objective:
PI-RADS is an internationally recognized system to quantify the risk of prostate cancer on MRI. In the UK, national guidelines advise radiologists also provide Likert scores, which assess overall risk of cancer by including clinical factors. This can improve predictive accuracy, but such scores are subjective. Recently, several mathematical tools have been proposed that aim to help to improve consistency in prostate reporting. These are based on modeling of known clinical risk factors alongside the radiological findings. This work aimed to compare the diagnostic accuracy of PI-RADS with a subjective Likert score given by experienced reporters, and with an objective alternative termed the “calculated adjustment of PI-RADS equivocal score” (CAPES).
Materials and Methods:
Five experienced reporters in a quaternary referral unit in the UK used a standardized reporting template to prospectively collect PI-RADS and Likert scores for 1467 mpMRI scans performed for query prostate cancer between January 2021 and June 2022. The CAPES tool was retrospectively applied to the cases scoring PI-RADS 3, and the recommended score was recorded. To allow fair comparison between the three scoring systems, analysis assumed that all patients scoring above a theoretical cut off would be referred for biopsy going forward. In view of national guidance, a PI-RADS, Likert, or CAPES score = 3 was used as this threshold. 434/1467 (29.6%) of patients were biopsied in practice. The available histological data was used to calculate the sensitivity, specificity, and PPV of each scoring system, based on the standardized biopsy protocol (score =3). Clinically significant prostate cancer was defined as ISUP grade = 2.
Results:
Each reporter consistently allocated roughly 60% (56.0% - 61.3%) of patients a PI-RADS score = 3, the theoretical threshold for biopsy. In contrast, there was significant variation in the proportion of patients who would have been recommended biopsy using Likert system (36.6% - 58.1%). Across all reporters, significantly fewer equivocal “3” scores were given using Likert (15.7%) or CAPES (2.2%) compared to PI-RADS (24.1%). Likert had higher specificity (69.0% vs 54.4%), sensitivity (98.3% vs 97.7%), and PPV (49.9% vs 40.3%) than PI-RADS for identifying ISUP =2 cancer. The CAPES tool had even higher specificity (81.4%) and PPV (61.2%) with only slightly lower sensitivity (93.4%) resulting in 37.1% (<em>n</em> = 316) fewer biopsies than PI-RADS, and 22.4% (<em>n</em> = 155) fewer biopsies than Likert across 1467 patients. The cancers missed by The CAPES tool (<em>n</em> = 23) were predominantly low grade (87.0% ISUP grade 2).
Conclusion:
Compared to PI-RADS scoring, Likert scoring and CAPES would both result in fewer equivocal scores, greater PPV, and fewer unnecessary biopsies. CAPES has the advantage of being objective, consistent, and less reliant on having the expertise and clinical information necessary for Likert scoring. Although no substitute for radiological knowledge, the high specificity and PPV achieved by the tool suggest that it may offer useful insight in equivocal cases.