AB048. A urinary-metabolomics-based panel for non-invasive detection of bladder cancer

Objective: Bladder cancer (BCa) is a common malignancy worldwide and has a high probability of recurrence. Early detection is vital to improve the overall survival rate. The common diagnostic modalities, such as cystoscopy and urinary cytology, have their limitations. In this study, potential metabolic biomarkers have been discovered through gas chromatography-mass spectrometry. Based on distinct metabolomics of urine between BCa patients and healthy people, we forged a non-invasive BCa diagnostic model and investigated its performance.

Methods: This study includes Training Phase, Modeling Phase and Test Phase. During the Training Phase, urine samples were collected from 32 patients diagnosed of bladder cancer and 21 healthy controls. We applied unsupervised principal component analysis (PCA) and orthogonal partial least-squares discriminant analysis (OPLS-DA) model used as a diagnostic model to distinguish two groups. We further constructed logistic regression model using combinations of the metabolites to improve the sensitivity and specificity for early BCa determination. In addition, we screened metabolites which AUC was more than 0.75 for establishing the model of diagnostic panel using logistic regressive analysis. In Test Phase, urine samples from 79 BCa patients and 51 non-BCa controls were subjected to test the diagnostic model. Moreover, by subgroup analysis of BCa, some metabolites were indentified to associate with tumor grade and stage.

Results: In Training phase, a set of 22 candidate differential metabolites was based on statistical significance and fold difference. Logistic diagnostic model has been established as below: Y=1.3333-8.891X(Glycine)×10-8-4.811X(3-Phosphoglycericacid)×10-5-5.625X(Cytosine)×10-5, with Area Under ROC Curve (AUC) =0.88, sensitivity =78.1% and specificity =95.2%. In Test phase, the efficiency of our diagnostic model shown AUC =0.705, sensitivity =62.0% and specificity =72.5%, better than that of urinary cytology. Besides, our non-invasive model can distinguish different BCa grade with high efficiency.

Conclusions: Our diagnostic model results suggest that urine metabolic profiling may have potential for clinical diagnosis of bladder cancer.

Keywords: Bladder cancer (BCa); a non-invasive BCa diagnostic; metabonomics; cancer markers