Diagnostic Prediction of Von Willebrand Disease using Multiple Bleeding Phenomics Datasets

Abstract

The rigorous assessment of bleeding symptoms is a key component in establishing a diagnosis in patients suspected of having von Willebrand disease (VWD) and other inherited bleeding disorders. Multiple bleeding questionnaires have been developed and validated to capture bleeding history phenotypes for assessing patients with bleeding disorders. In this study we developed a prediction model based on Naïve Bayes decision tree classifier by analyzing various phenotypic attributes derived from multiple bleeding questionnaires. We evaluated the classification effectiveness derived from the top 25 attributes with highest discriminations on prediction of VWD. We used data from 952 patients and the classifier achieved a precision of 95%, recall of 94%, and a Receiving Operating Curve (ROC) area under the curve of 0.97.