Abstract 282: Smoking and increased age in hematuric patients significantly confound the accuracy of classifiers for the diagnosis of urothelial cancer

Abstract

Abstract Smoking and increased age in hematuria patients significantly confound the accuracy of classifiers for the diagnosis of urothelial cancer Ricardo de Matos Simoes (a), Brian Duggan(b), Mark W. Ruddock (c), Frank Emmert-Streib (a), Declan O'Rourke (d), Hugh F. O'Kane (b), Funso Abogunrin (e), Cherith N. Reid (c), Neil H Anderson (f), David Curry (b), Kate E. Williamson (a). a Centre for Cancer Research and Cell Biology, Queen's University Belfast, 97 Lisburn Road, Belfast, UK. b Department of Urology, Belfast Health and Social Care Trust, Belfast, UK. c Randox Laboratories Ltd., County Antrim, UK. d Department of Pathology, Belfast City Hospital, Belfast, UK. e Department of Urology, Craigavon Area Hospital, Northern Ireland, f Department of Pathology, Royal Victoria Hospital, Belfast, UK. There is an urgent need for evidence-based risk stratification of hematuria patients. Currently, we have insufficient understanding of the characteristics which confound the accuracy of diagnostic classifiers for urothelial cancer (UC). Between November 2006 and October 2008 156 hematuric patients were recruited to a case control study. We defined UC diagnostic classifiers for 16 stratified subpopulations. Using a random forest classification based on ∼29 biomarkers measurements in urine, serum and plasma and random sub-sampling we generated 1000 training and test patient datasets to estimate probabilities of correct, incorrect and inconsistent classification for each patient. We defined classifiers for control and UC classified subpopulations. The distribution of characteristics across classifiable and non-classifiable patient subpopulations was explored using Fisher's exact test. The non-smoking stratified subpopulation achieved the greatest accuracy AUC = 0.95 (sd = 0•1). One hundred patients (64%) were classified correctly; 46/156 (30%) were non-classifiable, and 10/156 (6%) were inconsistently classified. The 23 non-classifiable UCs comprised 21 pTa, one pT1 and one CIS; two pTa and one pT1 were inconsistently classified. Five patients with newly diagnosed cancers other than UC, seven with “no diagnosis” and 11 with benign conditions were misclassified as UC. Consideration of smoking, increased age and antihypertensive medication in addition to biomarker profiles may achieve more accurate classification of UC in hematuria patients. This novel approach to diagnostic classifiers, once fine-tuned, may have the potential to correctly classify low risk bladder tumours and to reduce false positive cases. Citation Format: Kate E. Williamson. Smoking and increased age in hematuric patients significantly confound the accuracy of classifiers for the diagnosis of urothelial cancer. [abstract]. In: Proceedings of the 105th Annual Meeting of the American Association for Cancer Research; 2014 Apr 5-9; San Diego, CA. Philadelphia (PA): AACR; Cancer Res 2014;74(19 Suppl):Abstract nr 282. doi:10.1158/1538-7445.AM2014-282