Development of a Web Application based on Machine Learning for screening esophageal varices in cirrhosis.

INTRODUCTION

Esophageal varices (EV) are a common manifestation of portal hypertension in cirrhotic patients. Upper gastrointestinal endoscopy (UGE) is the gold standard for diagnosing EV. However, it is an invasive examination with a relatively high cost.

AIM: To develop a machine learning model for the prediction of EV in cirrhotic patients.

METHODS

This is a cross-sectional observational study including all cirrhotic patients, for whom an UGE was performed, between January 2010 and December 2019. We adopted a structured methodical approach with reference to CRISP-DM (Cross-Industry Standard Process for Data Mining). The different steps carried out were: data collection and preparation, modelization, and deployment of the predictive models in a web application.

RESULTS

We included 166 patients, 92 women (55.4%) and 74 men (44.6%). The mean age was 57.2 years. In UGE, 16 patients (9.6%) did not have EV. Other patients had EV grade 1 in 41 cases (24.7%), grade 2 in 81 cases (24.7%) and grade 3 in 28 cases (16.9%). After the selection phase, among the 36 initial variables, 19 were retained. Three machine learning models have been developed with a performance of 90%.

CONCLUSIONS

We developed a machine learning model combining several clinical and para-clinical variables for the prediction of EV in cirrhotic patients.