Automated algorithm for diagnosing gastrointestinal bleeding

BACKGROUND: Gastrointestinal bleeding (GIB) is a complication of many diseases of the gastrointestinal tract (GIT), including erosive and ulcerative lesions, vascular malformations, diverticula, and tumors. In developed countries, the GIB mortality rate ranges from 5% to 15%, reaching 30%–40% in the group of patients with severe recurrent bleeding.

AIM: The study aimed to develop an automated diagnostic algorithm for patients with GIB.

METHODS: Knowledge engineering is used to extract terms and their relationships from the scientific literature related to the GIT. After agreement with the experts, information on the diagnosis and treatment of patients with GIB was arranged using a MS Excel spreadsheet editor. For building GIB localization rules, the study included data from histories of 280 patients aged 20–94 years (61 [44; 74]); of these, 47.5% were women, while all others were men. The patients were diagnosed and treated at the Municipal Clinical Hospital No. 31 between 2008 and 2021. For testing the algorithm, data from histories of 514 patients aged 20–96 years (62 [46; 74]) were used; of these, 57% were men, while the rest were women. The patients under study were diagnosed and treated at the Municipal Clinical Hospital No. 17 and the Municipal Clinical Hospital No. 31 between 2008 and 2022. For each study subject, data were available on 37 signs, including 19 clinical, 3 laboratory, and 15 endoscopic signs. Statistical data analysis was performed using the Statistica 13 software package, R Project programming language, and GraphPad online calculator. The software implementation of the algorithm was performed using the JavaScript programming language.

RESULTS: Using polynomial logistic regression, an algorithm for differential diagnosis of GIB according to the preliminary localization of the bleeding source was developed. Having the data from 8 clinical and laboratory parameters with probability, a preliminary localization of the bleeding source may be determined. Thus, the probability of bleeding localization in the upper, middle, and lower GIT accounts for 84% (95% CI [78%; 89%]), 84% (95% CI [74%; 91%]), and 75% (95% CI [69%; 80%]), respectively. A final algorithm to support clinical decision-making in the management of patients with GIB (sergisa.smrtp.ru/medical/edit.html) was developed and implemented as a web-service, working with 92.2% efficiency.

The sequence of operations of the algorithm for diagnosing GIB is as follows:

1. Obtaining clinical and laboratory signs of a patient with suspected GIB.
2. Determining the nature of bleeding (overt/occult) using literature data, assessing the severity of bleeding (mild/medium/severe) based on Gorbashko classification, and detecting the preliminary localization of bleeding (upper/middle/lower GIT) by regression equation.
3. Providing recommendations for selecting a research method based on production rules and expert opinions.
4. Obtaining endoscopic signs.
5. Determining the localization and source of bleeding using the production rules.
6. Providing recommendations for stopping/preventing GIB based on literature and expert opinions.

CONCLUSIONS: An algorithm to support clinical decision-making for the management of patients with GIB, considering the nature of clinical manifestations and the severity and the cause of bleeding based on expert opinions, production rules, and polynomial logistic regression, which allows to assume a preliminary localization of the source of bleeding, was developed for the first time in the Russian Federation. The developed algorithm is implemented as a web-service and may be integrated into the medical information system at the automated workstation of a surgeon, an endoscopist, and a gastroenterologist to support clinical decision-making in the management of patients with GIB.