Publications

Pre-biopsy prostate cancer nomograms based upon serial PSA screening data in the PLCO Trial With Verification Bias Correction

Author(s): Izmirlian I,  Grubb III RL,  Black A,  Prorok PC,  Berg C,  Crawford D,  Andriole Jr G,  Kibel A

Journal: J Urology

Date: 2009 Apr

Major Program(s) or Research Group(s): EDRG

PubMed ID:

PMC ID: not available

Abstract: INTRODUCTION AND OBJECTIVES: Prostate cancer risk calculators have proven to be very useful in counseling patients as to risk of being diagnosed with prostate carcinoma (PCa). Unfortunately, for a given set of parameters, there is a wide discrepancy as to risk of disease among different prostate cancer risk calculators. This discrepancy may in part be failure to account for verification bias (ie not all patients underwent a biopsy). Herein we developed nomograms for 1 year incidence of PCa based upon models which controlled for verification bias.METHODS: From 1993-2001, PLCO investigators randomized 38,343 men aged 55-74 to receive annual PSA and DRE screening. 3,510 men underwent biopsy and 1,844 PCa were diagnosed with PCa(519 were Gleason ≥7 (Gl ≥7)). We performed Cox regression analysis of the endpoints of all PCa and Gl ≥7 using data on men who underwent a biopsy. These analyses incorporated baseline patient characteristics, PSA, PSA velocity (PSAV), DRE, and current number of screens, withthe latter 4 variables treated as time-dependent covariates. Adjustment for verification bias was done by weighting the main analyses. Subject weights were inversely proportional to probabilities of biopsy, predicted via a Cox regression of the biopsy endpoint on all screened men. Nomograms for the calculation of 1 year incidence probabilities were produced based upon the results of these models.RESULTS: Model selection was done a-priori based upon strong risk factors known in the literature. Upon analysis of the endpoints PCa and Gl ≥7, only the risk factors age, PSA, and PSAV retained statistical significance (p<0.05). Even so, all predictors were retained in the models as they enhanced the overall fit (p<0.0001). The nomograms had AUCs of 70.6% and 60.6% relative to the PCa and Gl ≥7 endpoints, respectively.CONCLUSIONS: We have developed nomograms, which control for verification bias, to predict the one year incidence of both all prostate cancer and high grade PCa based on serial PSA screening data from a large randomized prostate cancer screening trial. If verified, thesenomograms could provide more accurate prediction of true prostate cancer risk.