Prostate cancer detection before and after the introduction of prebiopsy MRI at a single centre: a retrospective comparison of 384 men undergoing systematic ultrasound-guided biopsy

Prebiopsy MRI may raise csPCa yield both by selecting men for biopsy and by influencing core placement; these channels are hard to separate when operators are not blinded to MRI. We compared detection before and after MRI was introduced at a centre using a systematic template throughout, and examined how far a change in recorded tumour-zone involvement reflected report coding. Retrospective two-cohort study of men biopsied in 2017 or 2022 at one Chinese hospital; all used a systematic template, with no targeted/fusion cores. After removing duplicates and episodes without an outcome, 384 men were analysed (2017, n = 271; 2022, n = 113). Endpoints were any cancer, csPCa (Gleason ≥ 7) and high-grade cancer (Gleason ≥ 8). Logistic regression adjusted for age and PSA (categorical and spline) estimated the era association; complete-case and imputed analyses were both conducted, with standardised absolute risk differences by marginal standardisation. Zone involvement in 160 positive biopsies, and the 2022 MRI reports, were each coded two ways and validated against a blinded second reader. MRI was performed in 111/113 men (98.2%) in 2022, absent in 2017. Crude detection did not differ, but 2022 patients presented at lower PSA ( P = 0.008). Standardised for age and PSA, detection was higher in 2022 (any cancer + 14.4 points, 95% CI 4.6–23.5; csPCa + 10.4, 0.9–19.7; high-grade + 12.2, 3.2–20.9), with similar spline/imputed estimates. Biopsy numbers fell by 59%. Second-reader MRI classification agreed with the automated pass in 77.5% (kappa 0.67), with a graded detection association (25.0%, 27.5%, 79.5%; P < 0.001). Transition-zone involvement appeared to rise from 27.5% to 84.3% by field coding but was 66.1% vs. 78.4% by text re-coding; a blinded reader agreed with the text rule (kappa 0.78) far more than the field rule (0.29; 0.09 in 2017), giving 70.6% vs. 74.5% ( P = 0.71). Detection was higher in 2022, an interval spanning MRI introduction. Because sampling stayed systematic but operators saw the MRI, selection and MRI-informed sampling cannot be separated; the difference is era-associated. The zone-involvement change was sensitive to coding and should not be read as biological.