In an era of personalized medicine, individualized risk assessment using easily available tools on the internet and the literature are appealing. However, uninformed use by clinicians and the public raises potential problems. Herein, we assess the performance of published models to predict insignificant prostate cancer (PCa), using a multi-national low-risk population that may be considered for active surveillance (AS) based on contemporary practice.METHODS:
Data on men suitable for AS but undergoing upfront radical prostatectomy were pooled from three international academic institutions in Cambridge (UK), Toronto (Canada) and Melbourne (Australia). Four predictive models identified from literature review were assessed for their ability to predict the presence of four definitions of insignificant PCa. Evaluation was performed using area under the curve (AUC) of receiver operating characteristic curves and Brier scores for discrimination, calibration curves and decision curve analysis.RESULTS:
A cohort of 460 men meeting the inclusion criteria of all four nomograms was identified. The highest AUCs calculated for any of the four models ranged from 0.618 to 0.664, suggesting weak positive discrimination at best. Models had best discriminative ability for a definition of insignificant disease characterized by organ-confined Gleason score ≤6 with a total volume ≤0.5 ml or 1.3 ml. Calibration plots showed moderate range of predictive ability for the Kattan model though this model did not perform well at decision curve analysis.CONCLUSIONS:
External assessment of models predicting insignificant PCa showed moderate performance at best. Uninformed interpretation may cause undue anxiety or false reassurance and they should be used with caution.