Informatics in Medicine Unlocked (Jan 2020)
Early diagnosis of esophageal varices using Boosted-Naïve Bayes Tree: A multicenter cross-sectional study on chronic hepatitis C patients
Abstract
The standard method for diagnosing varices by upper endoscopy is invasive, costly, and has many drawbacks. To overcome these drawbacks, this study aims to build a predictive Esophageal Varices diagnosing model that uses a minimum number of the most significant variables, trying to avoid unneeded endoscopy procedures. A dataset of a prospective cohort of 5013 chronic hepatitis C Egyptian patients collected from 2006 to 2017 was used in this study. The dataset included more than forty individual clinical laboratory variables, only ten of them were found to be significant, which was achieved via merging the correlation coefficient and p-value filtering methods to acquire improved results. All the 5013 patients of the sample dataset underwent endoscopic assessment, a costly procedure that could be avoided in most cases; hence, a highly accurate non-invasive diagnosing model is justifiably mandatory. To improve the overall performance of the predictive diagnosis model, this research introduced a novel algorithm that improves the traditional Naïve Bayes Tree by adding a boosting technique, dubbed “Boosted-Naïve Bayes Tree” (B-NBT). Applying B-NBT on our dataset revealed an improved performance in both AUROC (Area Under Receiver Operating Characteristic Curve) of 0.865 and Accuracy of 79%. In conclusion, this study revealed that there are only ten most significant variables that are sufficient for the noninvasive diagnosing model to preemptively predict EV with acceptable performance. This could advise physicians an efficient choice to save time and money— a medical contribution to this research. Additionally, an engineering contribution to the research was adding a proposed feature selection method to the boosting technique which improves the predictive performance.
