Issues in using predictive gradient bousting models to predict complications in patients with surgical peritonitis
- Authors: Polidanov M.A.1,2, Volkov K.A.3, Sukhoy D.V.4, Maslyakov V.V.5,6, Barulina M.A.4,7
-
Affiliations:
- Saratov State Medical University named after V. I. Razumovsky
- Saratov Medical University "Reaviz"
- Saratov State Medical University named after V.I. Razumovsky
- Perm State National Research University
- Saratov State Medical University named after V. I. Razumovsky of the Ministry of Health of the Russian Federation
- Saratov Medical University «Reaviz»
- Saratov Scientific Center of the Russian Academy of Sciences, Institute of Problems of Precision Mechanics and Control of the Russian Academy of Sciences
- Issue: Vol 13, No 1 (2024): Материалы XVII Всероссийского форума с международным участием «Инновационные технологии в хирургии»
- Pages: 37-39
- Section: Хирургические дисциплины
- URL: https://new.vestnik-surgery.com/index.php/2415-7805/article/view/9247
Cite item
Full Text
Abstract
With the development of technology, computational capabilities, and artificial intelligence, it has become possible to create a medical decision aid system for predicting the occurrence of complications in peritonitis. The aim of the study was to use predictive models of gradient bousting to predict complications in patients with surgical peritonitis. Materials and Methods. The study involved data from 1192 patients hospitalized with the diagnosis of peritonitis. The methods of classical statistics - Pearson's pair correlation and machine learning methods - gradient bousting on decision trees - were used to investigate the importance of signs (indicators of clinical-diagnostic and instrumental investigations) on the occurrence of complications. Precision, recall and F1-metrics were used to assess the quality of the resulting model. Results. One of the ways to improve the results of abscess treatment is the possibility to detect and predict the occurrence of complications at an early stage, which, undoubtedly, can greatly simplify the construction and strategy of treatment. Conclusions. The paper shows the principal possibility of using artificial intelligence methods to predict the probability of complications occurrence in peritonitis.
Full Text
Introduction. There is no doubt that due to the development of technology, computing capabilities, and artificial intelligence, it has become possible to create a system to assist medical decision-making in predicting the occurrence of complications in various diseases, including peritonitis [1-3].
The aim of the work was to use predictive models of gradient bousting for predicting complications in patients with surgical peritonitis.
Materials and methods of the study. The data of 1192 patients hospitalized with the diagnosis of peritonitis were studied. The methods of classical statistics - Pearson's pair correlation and machine learning methods - gradient bousting on decision trees - were used to investigate the importance of signs (indicators of clinical-diagnostic and instrumental investigations) on the occurrence of complications. Precision, recall and F1-metrics were used to assess the quality of the resulting model.
Study results. After preliminary statistical analysis, the following parameters showed the highest (more than 0.4) correlation with the outcome: blood urea biochemistry, abdominal cavity pressure, renal ultrasound, OAC leukocytes, infusion volume, OAM sugar, OAM ketone bodies, patient's age. The most significant parameters obtained by GBDT: abdominal pressure, nature of peritonitis, BX blood urea, blood cultures. After training the predictive model, the following metrics for predicting the absence of complications were obtained: Precision (Precision) - 0.92, Completeness (Recall) - 0.99, Area under curve (AUC) - 0.86.
Conclusion. Thus, the constructed model showed a good predictive ability for the absence of complications. That is, the potential possibility of predicting the occurrence or non-occurrence of complications in patients after surgery on the basis of the data collected at the patient's admission to the emergency room has been shown. But its implementation in clinical practice is premature because the model produces many false negatives when predicting that patients may have complications, which is a serious problem, especially in medicine.
About the authors
Maxim Andreevich Polidanov
Saratov State Medical University named after V. I. Razumovsky; Saratov Medical University "Reaviz"
Email: maksim.polidanoff@yandex.ru
ORCID iD: 0000-0001-7538-7412
SPIN-code: 2629-7545
laboratory assistant of the Department of Mobilization Training of Public Health and Disaster Medicine; Postgraduate student of the Department of Surgical Diseases, laboratory assistant of the Research Department
Russian Federation, 410012, Volga Federal District, Saratov Region, Saratov, 112 Bolshaya Kazachya St.; 410012, Volga Federal District, Saratov Region, Saratov, Verkhny rynok St., 10.Kirill Andreevich Volkov
Saratov State Medical University named after V.I. Razumovsky
Email: kvolee@yandex.ru
ORCID iD: 0000-0002-3803-2644
SPIN-code: 1127-3119
2nd year student of the Faculty of Medicine
Russian Federation, 410012, Volga Federal District, Saratov Region, Saratov, 112 Bolshaya Kazachya St.112Daniil Vladimirovich Sukhoy
Perm State National Research University
Email: kvolee@yandex.ru
ORCID iD: 0009-0000-6081-993X
3rd year student of the Faculty of Mechanics and Mathematics
Russian Federation, 614068, Perm, Bukireva str., 15Vladimir Vladimirovich Maslyakov
Saratov State Medical University named after V. I. Razumovsky of the Ministry of Health of the Russian Federation; Saratov Medical University «Reaviz»
Email: maslyakov@inbox.ru
ORCID iD: 0000-0001-6652-9140
SPIN-code: 4232-3811
Доктор медицинских наук, профессор, профессор кафедры мобилизационной подготовки общественного здравоохранения и медицины катастроф; Доктор медицинских наук, профессор, профессор кафедры хирургических болезней
Russian Federation, 410012, Volga Federal District, Saratov Region, Saratov, 112 Bolshaya Kazachya St.; 410012, Volga Federal District, Saratov Region, Saratov, Verkhny rynok St., 10.Marina Alexandrovna Barulina
Perm State National Research University; Saratov Scientific Center of the Russian Academy of Sciences, Institute of Problems of Precision Mechanics and Control of the Russian Academy of Sciences
Author for correspondence.
Email: marina@barulina.ru
ORCID iD: 0000-0003-3867-648X
SPIN-code: 1987-9965
Doctor of Physics and Mathematics, Professor, Advisor to the Rectorate, Interim Dean of the Faculty of Mechanics and Mathematics; Doctor of Physics and Mathematics, Professor, Head of the Laboratory «Analysis and synthesis of dynamic systems in precision mechanics», Chief Scientific Associate
Russian Federation, 15 Bukireva str., Perm, 614068References
- Барулина, М.А., Масляков, В.В., Полиданов, М.А., Романов, Р.А., Волков, К.А. Математические и алгоритмические методы исследования признаков возникновения осложнений при перитоните / М.А. Барулина, В.В. Масляков, МА. Полиданов, Р.А. Романов, К.А. Волков // Математическое моделирование, компьютерный и натурный эксперимент в естественных науках. – 2023. – № 2. – С. 39-44.
- Алипов, В.В., Тахмезов, А.Э., Полиданов, М.А., Мусаелян, А.Г., Кондрашкин, И.Е., Волков, К.А., Алипов, А.И. Улучшение результатов лечения и диагностики послеоперационных осложнений в абдоминальной хирургии с применением многофункционального устройства / В.В. Алипов, А.Э. Тахмезов, М.А. Полиданов, А.Г. Мусаелян, И.Е. Кондрашкин, К.А. Волков, А.И. Алипов // Медицинская наука и образование Урала. – 2023. – № 24 (1-113). – С. 67-71.
- Михайличенко, В.Ю., Воронков, Д.Е., Кисляков, В.В., Цап, А.А. Лечение тяжелых форм распространенного гнойного перитонита / В.Ю. Михайличенко, Д.Е. Воронков, В.В. Кисляков, А.А. Цап // Таврический медико-биологический вестник. – 2022. – № 25 (1). – С. 20-26.
Supplementary files
There are no supplementary files to display.
