A system for supporting medical decision-making in autoimmune gastritis: Clinical and morphological patterns and digital model architecture

Abstract

Introduction . Autoimmune gastritis (AIG) remains underdiagnosed and is most often discovered only at the stage of advanced atrophy, nutritional deficiencies, or neuroendocrine hyperplasia. Therefore, the development of clinical decision support systems (CDSS) is particularly relevant to the improvement of quality and standardization of the disease's diagnosis. Aim . To summarize current understanding of CDSS in the context of chronic gastritis diagnosis and to present our own CDSS integrating clinico-laboratory, endoscopic, and immunomorphological features to improve the diagnosis of AIG and patient risk stratification. Materials and methods . To systematize data on CDSS used in the diagnosis of chronic gastritis, a systematic literature search was conducted in the PubMed/MEDLINE, Embase, and Google Scholar databases using keywords and their combinations such as autoimmune gastritis, atrophic gastritis, clinical decision support system, and neural network. Our own CDSS was built by using probability based on the Bayesian theorem. This method helped verify a diagnostic model involving the selection of the most significant predictors, followed by ROC curve analysis providing a quantitative measure of the model's discriminative ability. Results. Currently, there are no algorithms for gastric cancer risk stratification or for selecting patient management strategies for autoimmune gastritis. Our own architecture of a prototype CDSS is presented. The key AIG diagnosis patterns within this system include postprandial distress syndrome, anemia, iron and vitamin B12 deficiencies, autoimmune comorbidities, atrophy predominantly from the gastric body to the fundus, a combination of intestinal and pseudopyloric metaplasia, and gastrindependent hyperplasia of gastrin-producing (G cell) and enterochromaffin-like cells (ECL). The CDSS is designed to evaluate quantitatively the probability of developing AIG, identify its mixed forms, and stratify the risk of neuroendocrine neoplasms, while generating personalized recommendations on frequency of endoscopic surveillance and laboratory monitoring. Conclusions. Further multicenter studies are needed to carry out external validation and integrate the proposed CDSSs into clinical practice.