Abstract
To validate and compare the accuracy and performance of nomograms predicting insignificant prostate cancer and to analyze their performance in patients with different cancer locations.
Our cohort consisted of 370 radical prostatectomy patients with Gleason ≤6 prostate cancer diagnosed on transrectal biopsy with at least 10 cores. We quantified the performance of each nomogram with respect to discrimination, calibration, predictive accuracy at different cut points, and the clinical net benefit. We also evaluated these parameters in subgroups of patients with predominantly anterior-apical (AA) and posterior-basal (PB) tumor location.
Insignificant prostate cancer was present in 141 patients (38%). The Kattan and Steyerberg nomograms outperformed other studied models and demonstrated fair discrimination (areas under the receiver operating characteristics curve 0.768 and 0.770, respectively), good calibration, balanced predictive accuracy, and the highest net benefit. All nomograms were less accurate at higher levels of predicted probability. The performance of the nomograms was better in patients with PB tumors than in those with AA tumors. The loss of correlation with the actual prevalence of insignificant prostate cancer at higher levels of predicted probability was not seen in the PB subgroup but was particularly noticeable in the AA subgroup.
The Kattan and Steyerberg nomograms demonstrated the best performance in predicting the probability of insignificant prostate cancer in a contemporary cohort of patients with Gleason ≤6 cancer diagnosed on specimens from an extended transrectal biopsy. However, all studied nomograms were more accurate in identifying significant rather than insignificant disease, particularly for tumors located in the apical and anterior prostate.
