Ricardo de Matos Simoes1, F Emmert-Streib2, Brian Duggan3, Mark W Ruddock4*, Declan O’Rourke5, Hugh F O’Kane6, Funso Abogunrin6, Cherith N Reid4, John V Lamont4, Peter Fitzgerald4 and Kate E Williamson7
1Dana Farber Cancer Institute, Brookline Avenue, Boston, USA
2Institute of Biosciences and Medical Technology, Finland
3Department of Urology, Ulster Hospital Dundonald, BT16 1RH, Northern Ireland, UK
4Randox Laboratories Ltd, Crumlin, BT29 4QY, Northern Ireland, UK
5Department of Pathology, Belfast City Hospital, BT9 7AB, Northern Ireland, UK
6Department of Urology, Belfast City Hospital, BT9 7AB, Northern Ireland, UK
7Centre for Cancer Research and Cell Biology, Queen's University Belfast, BT7 1NN, Northern Ireland, UK
*Corresponding Author: Mark W Ruddock, Randox Laboratories Ltd, Crumlin, BT29 4QY, Northern Ireland, UK.
Received: June 05, 2021; Published: July 14, 2021
Background: There is an urgent clinical need for evidence-based risk stratification modalities because of the increased prevalence of haematuria in the aging population. Currently, urothelial cancer (UC) classifiers have insufficient diagnostic accuracy to inform clinical decisions. Therefore, there is an urgent need for new tests which can at least stratify and if possible, be diagnostic.
Methods: To study patient characteristics associated with misclassification by UC diagnostic classifiers we analysed data collected from 156 patients recruited to a case control study between November 2006 and October 2008. First, we undertook a random forest classification based on measurements of 29 protein biomarkers measured in urine, serum and plasma and urinary creatinine, osmolality and protein. Second, we used random subsampling to generate 1000 training and test patient datasets and then estimated probabilities of correct, incorrect and inconsistent classification for each patient as either control or UC based on their hypergeometric distribution. Third, we identified clinical variables associated with incorrect classification using Fisher’s exact test.
Results: One hundred patients were classifiable, 46 non-classifiable and 10 inconsistently classifiable. Common confounders included smoking, age, grade, stage, AH medication, dipstick analyses, history of BPE, cytology diagnosis and presence/absence of inflammatory cells in cytology. Five patients with newly diagnosed prostate or kidney cancer and seven with “no diagnosis” were misclassified as UC. Sixteen ≥ pT2 and 24/47 pTa stage tumours were classifiable; 21 pTa tumours were non-classifiable and two were inconsistently classified.
Conclusion: Patients classified as UC require urgent referral; patients classified as controls > 65 years or smokers should also be referred because of their risk of early stage UC. Patients classified as controls who were non-smokers and ≤ 65 years were low risk. Our novel classification approach could increase understanding about the application of diagnostic classifiers in many complex diseases.
Keywords: Age; Biomarkers; Classification; Inflammatory Cells; Logistic Regression; Random Forest; Predictive Classifiers; Smoking
Citation: Mark W Ruddock., et al. “Confounding Effects of Heterogeneity and the Impact on the Accuracy of Urothelial Cancer Classifiers for Haematuria Patients”.Acta Scientific Medical Sciences 5.8 (2021): 33-46.
Copyright: © 2021 Mark W Ruddock., et al. This is an open-access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.