Ability of a urine gene expression classifier to reduce the number of follow up cystoscopies in bladder cancer patients

Abstract

This study aimed to improve our previous urine gene expression classifiers focusing on the detection of non-high-risk non-muscle-invasive bladder cancer (NMIBC), and develop a new classifier able to decrease the frequency of cystoscopies during bladder cancer (BC) patients' surveillance. A total of 597 urines from BC patients, controls and patients in follow-up for BC (PFBC) were included. The study has 3 phases. In the urinary biomarker discovery phase, 84 urines from BC and control patients were retrospectively included and analyzed by Ribonucleic Acid (RNA) sequencing. In the classifier development phase, a total of 132 selected genes from previous phase were evaluated by nCounter in 214 prospectively collected urines from PFBC (98 with tumor). A diagnostic classifier was generated by logistic regression. Finally, in the classifier validation phase, a multicentric and international cohort of 248 urines (134 BC and 114 nonrecurrent PFBC) was used to validate classifier performance. A total of 521 genes were found differentially expressed between non-high-risk NMIBC samples and all other groups (P < 0.05). An 8-gene diagnostic classifier with an area under curve (AUC) of 0.893 was developed. Validation of this classifier in a cohort of PFBC achieved an overall sensitivity (SN) and a negative predictive value (NPV) of 96% and 97%, respectively (AUC = 0.823). Notably, this accuracy was maintained in non-high-risk NMIBC group (SN = 94%; NPV = 98%). In conclusion, this 8-gene expression classifier has high SN and NPV in a real clinical scenario. The use of this classifier can reduce the number of follow-up cystoscopies in PFBC, although assessing its final place in clinical setting is necessary.