Establishment and Characterisation of Empirical Biomarker Predictive Value ROC Curves (PV-ROC) Using Cut-Off Distribution Curves from UBC® Rapid Test - An Urinary Point-of-Care (POC) Assay for Diagnosis of Bladder Cancer
23 Pages Posted: 12 Mar 2019More...
Objectives: UBC® Rapid Test measures antigens of soluble fragments of cytokeratins 8 and 18 in urine. Using data from a study applying the UBC® Rapid Test in bladder cancer patients, patients with urinary bladder cancer positive history, and healthy controls, predictive value cut-off distribution curves were constructed and transformed into predictive value ROC curves (PV-ROC), which can be compared to sensitivity-specificity ROC curves (SS-ROC) in a single SS/PV-ROC curves graph.
Material, Methods and Patients: In total 289 urine samples from 111 patients with tumours of the urinary bladder, 32 from patients with nonevidence of disease (NED) and from 146 healthy controls have been included in this study. Urine samples were visually qualitatively analysed by the UBC® Rapid point-of-care (POC) assay, and in addition quantitatively by the Concile Omega 100 POC Reader. Data for pairs of sensitivity/specificity as well as positive/negative predictive values were created by variation of test cut-off values for the whole patient cohort as well as for the healthy control group. Results: According to the distribution curves and their resulting ROC curves, elevated levels of UBC® Rapid Test in urine are generally higher in patients with bladder cancer in comparison to the healthy control group, showing clinically applicable optimal thresholds. With respect to visual evaluation of the qualitative assay, sensitivity was 58.5% and specificity was 88.2%, leading to an estimated cut-off at =12.0 µg/l. PPV was 87.8% and NPV 74.9%, leading to an estimated cut-off level at =22.5 µg/l. According to the optimal values of the respective ROC curves, both concentrations were in areas of optimal cut-off for clinical decisions, however, in case of predictive values, this only applied to the NPV. Evaluation of PV-ROC showed that two or more distinct values of PPV can correspond to the same value of NPV and conversely, indicating a complexity in PV-ROC curves which is not existing in SS-ROC curves. Moreover, in contrast to the SS-ROC curve, which (generally) contains an area under the curve (AUC) and a full range from 0% to 100%, the PV-ROC curve in our investigation in did have neither an AUC, nor a range from 0% to 100%.
Conclusions: According to our knowledge, predictive value distribution curves including PV-ROC curves for biomarkers have not yet been published by other authors. The presented results show that PV-ROC curve characteristics differ distinctly from those of SS-ROC curves in several aspects. In addition, this is the first concentration estimation of the decision limit for predictive judgement for the visual UBC® Rapid Test at hand of distribution curves, and for detecting an optimal biomarker cutoff for clinical prediction for the quantitative as well for the qualitative POC evaluation. The new approaches described are applicable to any other biomarkers, and can be extended for predictive ROC curves in other fields of application, like evaluations in biology, biochemistry, or technology.
Funding Statement: The test systems were sponsored by concile GmbH, Freiburg/Breisgau, Germany and IDL Biotech AB, Bromma, Sweden.
Declaration of Interests: The authors report: "None."
Ethics Approval Statement: The study was approved by the local Institutional Review Board of Medical Association Brandenburg (AS 147(bB)/2013).
Keywords: ROC; biomarker; predictive value
Suggested Citation: Suggested Citation