The diagnostic accuracy of urine-based tests for bladder cancer varies greatly by patient

Background: Spectrum effects refer to the phenomenon that test performance varies across subgroups of a population. When spectrum effects occur during diagnostic testing for cancer, difficult patient misdiagnoses can occur. Our objective was to evaluate the effect of test indication, age, gender, race, and smoking status on the performance characteristics of two commonly used diagnostic tests for bladder cancer, urine cytology and fluorescence in situ hybridization (FISH). Methods: We assessed all subjects who underwent cystoscopy, cytology, and FISH at our institution from 2003 to 2012. The standard diagnostic test performance metrics were calculated using marginal models to account for clustered/repeated measures within subjects. We calculated test performance for the overall cohort by test indication as well as by key patient variables: age, gender, race, and smoking status. Results: A total of 4023 cystoscopy-cytology pairs and 1696 FISH-cystoscopy pairs were included in the analysis. In both FISH and cytology, increasing age, male gender, and history of smoking were associated with increased sensitivity and decreased specificity. FISH performance was most impacted by age, with an increase in sensitivity from 17 % at age 40 to 49 % at age 80. The same was true of cytology, with an increase in sensitivity from 50 % at age 40 to 67 % at age 80. Sensitivity of FISH was higher for a previous diagnosis of bladder cancer (46 %) than for hematuria (26 %). Test indication had no impact on the performance of cytology and race had no significant impact on the performance of either test. Conclusions: The diagnostic performance of urine cytology and FISH vary significantly according to the patient demographic in which they were tested. Hence, the reporting of spectrum effects in diagnostic tests should become part of standard practice. Patient-related factors must contextualize the clinicians' interpretation of test results and their decision-making.