Accurate prediction tools in prostate cancer require consistent assessment of included variables. (3rd July 2016)

• Record Type: Journal Article
• Authors: Jäderling, Fredrik; Nyberg, Tommy; Blomqvist, Lennart; Bjartell, Anders; Steineck, Gunnar; Carlsson, Stefan
• Is Part Of: Scandinavian journal of urology. Volume 50:Number 4(2016)
• Page Start: 260
• Page End: 266
• Publication Date: 2016-07-03
• Subjects: Clinically organ-confined disease -- extraprostatic extension -- multivariable logistic regression -- nomogram -- prediction model -- preoperative variables -- prostate cancer -- validation; Urology -- Periodicals; 616.6
• DOI: 10.3109/21681805.2016.1145736
• Languages: English
• ISSNs: 2168-1805

Abstract

Abstract: Objective : The aim of this study was to create a preoperative prediction model predicting extraprostatic tumour growth in men with clinically organ-confined disease from a prospectively collected Swedish cohort.Materials and methods : The study used data from 3386 men in the prospective multi-centre Laparoscopic Prostatectomy Robot Open (LAPPRO) trial, with 14 participating urological departments. External validation was performed using a cohort of 634 men from the largest study centre with patients who underwent surgery before and after the inclusion period of the LAPPRO study. External validation of the updated Partin table was used for comparison. The prediction models were created by multivariable logistic regression. Nomogram prediction performance, internal, internal–external and external validation are presented as the area under the receiver operating characteristic curve (AUC).Results : The nomogram reached a prediction performance with an AUC of 0.741, with internal and external validation of 0.738 and 0.698, respectively. Internal–external validation showed great divergence between centres, with AUCs ranging from 0.476 to 0.892, indicating inconsistencies in pathological staging or one or more of the included variables in the regression model. When including centre as a variable in the multivariable model it was significantly associated with the outcome of pT3 ( p < 0.001). AUC for external validation of the Partin table was 0.694.Conclusions : Accurate prediction tools in prostate cancer require consistent assessment of included variables, and local validation is needed before the use of such tools in clinical practice.