Development of a prediction model for peptic ulcer disease using the fusion of a feature selection strategy and ensemble algorithms.

backgroundPeptic ulcer disease is a prevalent gastric and duodenal disorder associated with H. pylori infection, which is related to numerous complications and diseases. Early prediction of this disease plays a significant role in reducing and eliminating its complications. So far, machine learning has shown promise for efficiently predicting health conditions in healthcare settings. So, this study aimed to establish a prediction model by getting assistance from machine learning and risk factors for peptic ulcer disease to predict this disease efficiently and achieve preventive purposes.

methodsThis retrospective study was conducted from January 2018 to December 2023 and included 1803 individuals with suspected peptic ulcer disease. We used selected ensemble algorithms, including XG-Boost, Random Forest, Ada-Boost, LightGBM, and Cat-Boost, to build prediction models and choose the best-performing ones.

resultsXG-Boost with an AU-ROC of 0.936 and a 95% CI of AU-ROC = [0.909–0.956] obtained more predictive strength than others. Factors of the history of H. pylori, NSAIDs, family history of gastric ulcer, smoking, and fruit intake were considered the top-ranked risk factors for peptic ulcer disease.

conclusionsThis study revealed that XG-Boost, with satisfactory performance and clinical usability, can potentially identify high-risk individuals for peptic ulcer disease in clinical settings.