An artificial network model improves diagnostic ability of nuclear matrix protein 22 in the screening for bladder urothelial carcinoma: a prospective analysis of 800 consecutive cases

Abstract

4581 Background: Although urine nuclear matrix protein 22 (NMP22) is one of useful molecular markers for the diagnosis of urothelial cancer, sole use of NMP22 would not be superior to urine cytology. It is mainly because NMP22 is affected by leukocyturia, erythrocyturia, and urinary tract instrumentation. We established an artificial neural network model to improve the diagnostic ability of NMP22.. Methods: We prospectively assessed urinalysis, urine cytology, and NMP22 in 800 consecutive patients (men; 48%, mean age ± standard deviation; 61.5 ± 14.8 years) who had no history of urological malignancy. Microscopic hematuria (57.5%) was the major reason for reference. Definitive diagnosis of urothelial tumor was obtained by cystoscopy, ultrasonography, excretory urography, and contrast enhanced computerized tomography. Receiver operating characteristic (ROC) analysis was performed for NMP22, urine cytology, and output of the neural network model incorporating patient age, gender, microscopic count of erythrocyte, leukocyte, and epithelial cell, and proteinuria grade on dipstick test. Results: There were 41 patients diagnosed to have urothelial carcinoma including 32 bladder cancers and 9 upper urinary tract cancers. ROC analysis failed to show a significant difference in areas under curve (AUC) for sole use of NMP22 (AUC; 0.862) and cytology (AUC; 0.827, p = 0.472). AUC for the neural network was 0.945 which is significantly superior to sole use of NMP22 (p = 0.005) and cytology (p = 0.007). Positive and negative predictive values (PPV and NPV) for our model were superior to sole NMP22 and cytology (see table). Conclusions: Our neural network model enables to create a screening test superior to conventional modalities by combining NMP22 with urinalysis results in patients without history of urothelial cancer. No significant financial relationships to disclose.