The role of systematic biopsies distant to the MRI visible lesion remains uncertain. We assessed whether baseline clinical parameters from large secondary care diagnostic cohort might predict which patients could avoid non-targeted systematic biopsies.
Random forest models were developed from the Rapid Assessment for Prostate Imaging and Diagnosis (RAPID) Online Registry. Patients with mpMRI disease in only one half of the gland (either laterality or anterior / posterior) were included. Possible predictor variables included age, ethnicity, family history, 5-ARI use, prior biopsy status, DRE findings, PSA, PSA density, PIRADS and Likert score and the number and location of lesions. Primary outcome was the presence of csPCa within the opposite half of the prostate. Four thresholds for csPCa were chosen. These included Gleason grade (GG) >/=3+4, GG >/=4+3, PROMIS 1 (GG >/=4+3 and/or maximum cancer core length (MCCL) >/=6mm of any grade) and PROMIS 2 (GG >/=3+4 and/or maximum cancer core length (MCCL) >/=4mm of any grade). Clinical utility of the models was tested by decision curve analysis.
Between 04/2017 and 03/2021, 3853 patients were included in RAPIDOnline between 3 centres. 974/3853 (25.3%) of these satisfied the inclusion criteria. Median [IQR] age was 67 [60-73], PSA 7.1 [5.2-10.7] and PSA density 0.16 [0.10-0.26]. Decision curve analysis showed that except at very low patient or clinician tolerance for missing csPCa of any definition, the models all have utility in deciding whether to perform systematic biopsies in patients with an mpMRI visible lesion. At a tolerance threshold of 10%, the GG >/=3+4 model offers a net benefit increase of 2.5% over performing SBx on all patients. A similar net benefit increase of 2.5% is seen with the GG >/=4+3 model at a tolerance threshold of 5%. The PROMIS 1 and 2 models offer net benefit increases of 2.5% and 5% for tolerances of 5% and 10%, respectively. This may also be visualised by the biopsy avoidance rate. i.e., the number of patients that could have avoided a biopsy at a given risk threshold. For a risk tolerance of 10% for GG >/=3+4 and PROMIS 2, or 5% for GG 4+3 and PROMIS 1 around 1 in 3 patients undergoing MRI-TB could avoid SBx if the model were used clinically.
This model can make clinically useful predictions regarding the presence of csPCa outside of regions of interest on mpMRI. In its current iteration around a third of men undergoing MRI-TB could avoid non-targeted systematic biopsy at acceptable levels of risk tolerance.