Introduction. Healthcare is one of the priority sectors for the deployment of artificial intelligence (AI) technologies worldwide, including Russia. A key area of AI deployment is the integration of AI-base software as a medical device (AI SaMD) into the Unified digital systems of the healthcare sector of the Russian Federation.Aim. Research of the results of the deployment of AI SaMD in healthcare of the Russian Federation in 2023.Materials and methods. The State Register of Medical Devices and Organizations (individual entrepreneurs) engaged in the production and manufacture of medical devices was used as information about AI SaMD registered in Russia. As information on the deployment of AI SaMD, data from monitoring to the federal project “Creating a single digital system in healthcare” was used, including reports from constituent entities of the Russian Federation upon these activities. The results of the implementation of AI SaMD in Moscow were obtained according to data from the Moscow Department of Health as part of an experiment on the use of innovative technologies in the field of computer vision for the analysis of medical images.Results. As of January 1, 2024, Roszdravnadzor registered 26 AI SaMD, 77 % of them were developed by 13 Russian companies. At the end of 2023, 84 (94 %) constituent entities of the Russian Federation met the minimum established target for the purchase of AI SaMD. Within the framework of public procurement procedures provided by law, 106 government contracts were signed for the purchase and deployment of AI SaMD for a total amount of 448 million 430 thousand rubles.Conclusion. In 2023, the Russian healthcare system made a significant breakthrough in terms of the practical deployment of AI SaMD. Completed procurement and deployment projects are the basis for subsequent industry development.

Aim. To develop and validate a machine learning model designed to identify suspected pulmonary embolism (PE) based on various clinical features from electronic health records (EHRs) of out- and inpatients.Material and methods. Data from 19730 patients from 7 Russian regions were taken for analysis. EHR data were analyzed for the period from March 21, 2007 to February 4, 2022. Complaints, clinical and laboratory data, and concomitant diseases were used as diagnostic signs. PE was diagnosed in 1379 patients. Diagnosis of PE was based on ICD-10 codes. Seven machine learning algorithms were applied to diagnose pulmonary embolism: XGBoost, LightGBM, CatBoost, Logistic Regression, MLP Classifier, Random Forest Classifier, Gradient Boosting Classifier.Results. The Gradient Boosting Classifier-based model was selected for further prospective testing with the sensitivity of 0,899 (95% confidence interval (CI), 0,864-0,932), specificity of 0,875 (95% CI, 0,863-0,86), area under the ROC curve of 0,952 (95% CI, 0,938-0,964). The following signs had the greatest prediction value: cough, respiratory disorders, blood creatinine, body temperature, general weakness, heart rate, respiratory rate, edema, antihypertensive therapy, saturation and age.Conclusion. The model is designed for the initial encounter of patients with complaints and suspected PE, regardless of the type of care.

Introduction. The introduction of artificial intelligence (AI) technologies in Russian healthcare is an important step to improve the efficiency and quality of medical care. The development of AI technologies helps automate data processing, support physician decision-making and improve predictive analytics. Aim. Analysis of the results of creation and implementation of software solutions using AI technologies in Russian healthcare. Materials and Methods: A systematic search for data in the State Register of Medical Devices and on the official website of the Unified Information System in the field of procurement was conducted. The main research methods were analysis of registered AI-based medical devices and monitoring of their use in medical institutions. Results. Quantitative indicators of implementation of solutions using AI technologies in Russian healthcare have been determined. The factors facilitating and hindering the introduction of innovations were formulated. The list of components of the methodology of implementation and operation of medical solutions based on AI technologies, related changes in the organization of medical care, personnel training, and patients' involvement in the development of their health has been defined. Conclusions. Significant progress in the use of AI in Russian healthcare requires further disclosure of methodological, organizational, technological and economic issues. Continued sharing of regional practices and knowledge will be key to building trust in AI technologies.

Background: Improved survival of patients after acute coronary syndromes, population growth, and overall life expectancy rise have led to a significant increase in the proportion of patients with stable coronary artery disease (CAD), creating a significant load on the entire healthcare system. The disease often progresses with the development of many complications while significantly increasing the likelihood of hospitalization. Developing and applying a machine learning model for predicting hospitalizations of patients with CAD to an inpatient medical facility will allow for close monitoring of high-risk patients, early preventive interventions, and optimized medical care. Aims: Development and external validation of personalized models for predicting the preventable hospitalizations of patients with stable CAD and its complications using ML algorithms and data of real-world clinical practice. Methods: 135,873 depersonalized electronic health records of 49,103 patients with stable CAD were included in the study. Anthropometric measurements, physical examination results, laboratory, instrumental, anamnestic, and socio-demographic data, widely used in routine medical practice, were considered as potential predictors, a total of 73 features. Logistic regression, decision tree-based methods including gradient boosting (AdaBoost, LightGBM, XGBoost, CatBoost) and bagging (RandomForest and ExtraTrees), discriminant analysis (LinearDiscriminant, QuadraticDiscriminant), and naive Bayes classifier were compared. External validation was performed on the data of a separate region. Results: The best results and stability to external validation data were shown by the CatBoost model with an AUC of 0.875 (95% CI 0.865 -0.885) for the internal testing and 0.872 (95% CI 0.856 -0.886) for the external validation. The best model showed good performance evaluated through AUROC, Brier score and standardized net benefit (for the target NPV threshold) for the validation dataset that was only slightly similar to the train data. Conclusion: The metrics of the best model were superior to previously published studies. The results of external validation demonstrated the relative stability of the model to new data from another region that confirms the possibility of the model 's application in real clinical practice.

BACKGROUND: The incidence of diabetes mellitus (DM) both in the Russian Federation and in the world has been steadily increasing for several decades. Stable population growth and current epidemiological characteristics of DM lead to enormous economic costs and significant social losses throughout the world. The disease often progresses with the development of specific complications, while significantly increasing the likelihood of hospitalization. The creation and inference of a machine learning model for predicting hospitalizations of patients with DM to an inpatient medical facility will make it possible to personalize the provision of medical care and optimize the load on the entire healthcare system.AIM: Development and validation of models for predicting unplanned hospitalizations of patients with diabetes due to the disease itself and its complications using machine learning algorithms and data from real clinical practice.MATERIALS AND METHODS: 170,141 depersonalized electronic health records of 23,742 diabetic patients were included in the study. Anamnestic, constitutional, clinical, instrumental and laboratory data, widely used in routine medical practice, were considered as potential predictors, a total of 33 signs. Logistic regression (LR), gradient boosting methods (LightGBM, XGBoost, CatBoost), decision tree-based methods (RandomForest and ExtraTrees), and a neural network-based algorithm (Multi-layer Perceptron) were compared. External validation was performed on the data of the separate region of Russian Federation.RESULTS: The best results and stability to external validation data were shown by the LightGBM model with an AUC of 0.818 (95% CI 0.802–0.834) in internal testing and 0.802 (95% CI 0.773–0.832) in external validation.CONCLUSION: The metrics of the best model were superior to previously published studies. The results of external validation showed the relative stability of the model to new data from another region, that reflects the possibility of the model’s application in real clinical practice.

Aim. To estimate the prevalence of ischemic stroke (IS) and the appointment of anticoagulant therapy in patients with atrial fibrillation (AF) depending on body mass index (BMI) as part of a retrospective analysis of big data from certain subjects of the Russian Federation using artificial intelligence technologies.Material and methods. The information was obtained from the Webiomed predictive analytics platform, which includes depersonalized data from electronic health records of patients in 6 Russian constituent, extracted using artificial intelligence technologies. Individuals with AF aged ≥18 years were selected with available data on BMI of 18,5-60,0 kg/m2 inclusive (n=56003; men, 41,0%; age, 67,4±14,5 years, CHA2DS2-VASc score, 3,4±1,8). The following BMI ranges were identified: 18,5-21,9 kg/m2, 22,0-24,9 kg/m2 (taken as a reference), 25,0-29,9 kg/m2, 30,0-34,9 kg/m2, 35,0-39,9 kg/m2 and 40,0-60,0 kg/m2. The indicators were analyzed in age ranges (≤64 years, 65-74 years, ≥75 years) separately among men and women.Results. Among men ≤64 years of age, patients with overweight and class 2 obesity were characterized by a significantly higher incidence of IS. Among women ≤64 years, a significantly higher frequency of IS was found in subgroups with overweight and class 1-3 obesity, while among women aged 65-74 years — in a subgroup with a BMI of 18,5-21,9 kg/m2. Patients aged ≥75 years showed an insignificant trend towards the maximum frequency of IS with a BMI of 18,5-21,9 kg/m2. A higher frequency of anticoagulant therapy prescription was found in subgroups with overweight and class 1-3 obesity; in most age and sex subgroups, the differences are significant. A significantly lower frequency of anticoagulant therapy prescription to persons ≥75 years of age with a BMI of 18,5-21,9 kg/m2 was noted.Conclusion. The study showed a significant BMI paradox in the context of the relationship between the frequency of IS and BMI value in patients with AF. A higher incidence of IS in persons ≤64 years of age with a BMI ≥25 kg/m2 compared with pa­tients with normal weight may be an additional argument for establishing an inde­pen­-dent prognostic role of obesity in the development of thromboembolic events in AF.

Aim. To assess the effect of febrile neutropenia (FN) prophylaxis with granulocyte colony-stimulating factors (G-CSF) in real-world cancer patients. Materials and methods. We conducted a statistical analysis of anonymized medical records collected in the Webiomed platform. Before analysis, the cards were validated by clinical experts. Electronic records were extracted according to two principles: mentioning D70 in the diagnosis or mentioning a chemotherapy regimen associated with a high risk of FN (20%), requiring the primary prevention of neutropenia. Thus, we obtained two datasets comprising 47.085 (590 patients) and 30.523 (398 patients) records, respectively. Results. Based on the analysis results, the most common risk factors for FN development were highly hematologically toxic chemotherapy regimens and elderly age about 50% in the adult population. In both datasets, the number of female patients prevailed (63.7% in dataset 1, 91.2% in dataset 2), so the most common was breast cancer. Less common were cervical cancer, digestive cancer, and lung cancer. Despite the indications for primary prevention of FN, for safety and importance of achieving the planned dose intensity, it was administered in 18.3% of patients in dataset 1 and 2.3% in dataset 2. No FN or G-CSF-related adverse events were reported in patients who received adequate primary prevention. Conclusion. Some issues related to G-CSF administration in cancer patients were identified. We identified the insufficient provision of patients with primary prevention of FN, which negatively affects survival rates and reduces adherence to antitumor therapy. Real-world data demonstrate the efficacy and safety of FN prevention and planned dose intensity maintance in cytotoxic therapy regimens.

Aim . To compare clinical characteristics of patients with atrial fibrillation (AF) depending on renal filtration function based on a retrospective analysis of data in individual subjects of the Russian Federation (RF). Material and methods . The information was taken from the Webiomed predictive analytics platform, including 80775 patients with AF (men, 42,5%, mean age, 70,0±14,3 years) who underwent outpatient and/or inpatient treatment in medical organizations in 6 Russian subjects in 2016-2019 with data on blood creatinine levels. For comparative analysis, the ranges of estimated glomerular filtration rate (eGFR) were selected: ≥60 ml/min/1,73 m 2 , 30–59 ml/min/1,73 m 2 , and <30 ml/min/1,73 m 2 . Results . The analysis showed that 45128 (55,9%) patients were characterized by eGFR <60 ml/min/1,73 m 2 , of which in 35212 (78%) patients eGFR was in the range of 30-59 ml/min/1,73 m 2 , in 9916 (22%) — <30 ml/min/1,73 m 2 . Patients with eGFR <60 ml/min/1,73 m 2 compared with those with eGFR ≥60 ml/min/1,73 m 2 were older (75,4±10,9 vs 63,0±15,2 years, p<0,001), had higher incidence of ischemic stroke (IS) (10,9 vs 6,5%, p<0,001), myocardial infarction (MI) (11,5 vs 7,7%, p<0,001) and intracranial hemorrhage (ICH) (1,0 vs 0,7%, p<0,01), as well as higher rate of anticoagulant therapy (ACT) (47,0 vs 33,2%, p<0,001). Men and women with eGFR of 30-59 and <30 ml/min/1,73 m 2 in the age ranges ≤64 years and 65-74 years had a higher incidence of IS and MI compared with patients with eGFR ≥60 ml/min/173 m 2 . The frequency of ICH on warfarin compared with direct oral anticoagulants was significantly higher in the subgroup with eGFR of 30-59 ml/min/1,73 m 2 (1,1 vs 0,7%, p<0,01). Conclusion . Patients with AF and eGFR <60 ml/min/1,73 m 2 are charac­terized by greater comorbidity, a higher incidence of IS, MI and ICH compared with patients with AF and eGFR ≥60 ml/min/1,73 m 2 , while ACT prescription rate as of 2016-2019 in some Russian subjects was unsatisfactory. This emphasizes the need to optimize risk stratification, ACT and algorithms for the prevention of athero­thrombotic events, as well as the development of nephro­protective strategies to reduce the rate of progression of renal dys­function in this cohort of patients.

Aim. To develop a model for predicting the subclinical carotid atherosclerosis (SCA) in order to refine cardiovascular risk (CVR) using machine learning methods in overweight and obese patients without hypertension, diabetes and/or cardiovascular disease (CVD).Material and methods. Anonymized database (DB) Webiomed (2.9 million patients) was used. There were following inclusion criteria: age ≥18 years, body mass index ≥25 kg/m2, availability of data on ultrasound of extracranial arteries. Patients with hypertension, diabetes and/or CVD were excluded from the analysis. Data on 5750 patients were selected, of which atherosclerotic plaques were detected in 385 people. The final data set contained information on 447 patients, 197 (44,1%) of which had SCA. Quantitative and categorical traits for model training were taken with 40% occupancy in the database. The number of final traits for machine learning was 28. When creating the model, 3 Random Forest algorithms, AdaBoostClassifier, KNeighborsClassifier and the Scikit-learn library were used. To improve the model performance, the fill missing function was used. The target parameters of the model were given a predictive ability (accuracy) of at least 75%, while the area under the ROC curve was at least 0,75.Results. The resulting dataset was divided into training and test parts in a ratio of 80:20. Depending on the applied algorithms, the learned model was characterized by a predictive ability of 75-97%, sensitivity of 77-92%, specificity of 80-98%, and area under the ROC-curve of 0,88-0,97. Taking into account the accuracy metrics, the best results were obtained for the model learned by the Random Forest algorithm (95%, 92%, 98% and 0,95, respectively).Conclusion. The developed model can help a physician make a decision to refer an overweight and obese patient without cardiovascular diseases for ultrasound of extracranial arteries, which contributes to a more accurate CVR stratification. The introduction of such risk stratification algorithms into practice will increase the accuracy and quality of CVR prediction and optimize the system of preventive measures.

Aim. Comparative analysis of mathematical models obtained using multivariate logistic regression (MLR) with stepwise inclusion of predictors and machine learning (ML) for assessing the probability of subclinical carotid atherosclerosis in normotensive overweight and obese patients without cardiovascular diseases and/or diabetes.Material and methods. We received data on patients from the Webiomed platform database. The inclusion criteria were age ≥18 years, body mass index ≥25 kg/m2, extracranial artery ultrasound results, while the exclusion criteria included diabetes and/or cardiovascular disease. MLR analysis was carried out with stepwise inclusion of predictors. ML algorithms were used to create an alternative model.Results. The overall percentage of true results for MLR model was 73,2%, while the proportion of true negative and positive predictions was 80,1% and 63,4%, respectively. Mathematical models created using ML methods are characterized by a predictive value from 75 to 97% with a sensitivity of 77 to 92% and a specificity of 80 to 98%.Conclusion. A significant superiority of ML models was revealed in the study of available clinical and paraclinical parameters. Integration of ML mathematical models into a diagnostic algorithm for making a decision to refer a low-risk patient for extracranial artery ultrasound will significantly improve its accuracy and cost efficiency.

The use of artificial intelligence technologies in Russian healthcare is a priority area for implementing a national strategy for the development of artificial intelligence in the country. The introduction of digital solutions based on artificial intelligence in healthcare facilities should improve the standard of living of the population and the quality of medical care, including areas of preventive examinations, diagnostics based on image analysis, prediction of disease development, selection of optimal drug dosages, reducing the threat of pandemics, and automating and increasing the accuracy of surgical interventions. Policy management and technical regulation are under development in the field of artificial intelligence in healthcare. The domestic market for relevant solutions has been created, and some products have been certified as medical devices from Roszdravnadzor (Federal Service for Surveillance in Healthcare). Various teams of scientists are conducting research. However, Russia is still behind the leading countries in the field of artificial intelligence, such as the United States and China. Investments in healthcare products based on artificial intelligence decreased significantly in 2021. The major reasons for the lag, at least in terms of market indicators, are low demand and the inability of state medical organizations to fund artificial intelligence projects. There are also other issues related to trust in the safety and effectiveness of such solutions.

The first studies and publications devoted to the use of information technologies in the health care of the USSR appeared in the 60s of the last century. Since that time, our country has formed its own scientific school and a market for a wide variety of software products intended for use in medicine and healthcare. Since 2011, a number of major federal projects in the field of healthcare informatization have been implemented, which, in general, made it possible to provide infrastructure and basic equipment, communication channels and software for the lion’s share of medical organizations, connect them into a single secure information network and ensure the exchange and accumulation of data on the work of the health care system of the Russian Federation. The article presents a description of the history of the development of these processes, as well as the current results of informatization of the healthcare industry in Russia.

Objective: to review domestic and foreign literature on the issue of machine learning methods applied in medical information systems (MIS), to analyze the accuracy and efficiency of the technologies under study, their advantages and disadvantages, the possibilities of implementation in clinical practice.Material and methods. The literature search was performed in the PubMed/MEDLINE databases covering the period from 2000 to 2020 (using groups of keyphrases: "machine learning", "laboratory data", "clinical events", "prediction diseases"), CyberLeninka ("machine learning", "laboratory data", "clinical events", "prediction diseases" Russian keyphrases combinations) and Papers With Code ("clinical events", "prediction diseases", "electronic health record"). After reviewing the full text of 30 literature sources that met the selection criteria, the 19 most relevant articles were selected.Results. An analysis of sources that describe the application of artificial intelligence techniques to obtain predictive analytics, taking into account information about patients, such as demographic, anamnestic, and laboratory data, the data of instrumental studies, information about existing and former diseases available in MIS, was performed. The existing ways of predicting adverse medical outcomes using machine learning methods were considered. Information about the significance of the used laboratory data for constructing high-precision predictive mathematical models is presented.Conclusion. Implementation of machine learning algorithms in MIS seems to be a promising tool for effective prediction of adverse medical events for wide application in real clinical practice. It corresponds to the global trend in the development of personalized medicine based on the calculation of individual risk. There is an increase in the activity of research in the field of predicting noncommunicable diseases using artificial intelligence technologies.

Currently, software products for use in medicine are actively developed.Among them, the dominant share belongs to clinical decision support systems (CDSS), which can be intelligent (based on mathematical models obtained by machine learning methods or other artificial intelligence technologies) or non-intelligent.For the state registration of CDSSs as software medical products, clinical trials are required, and the protocol of trial is developed jointly by the developer and an authorized medical organization.One of the mandatory components of the protocol is the calculation of the sample size.This article discusses the calculation of the sample size for the most common case, the binary outcome in diagnostic/screening and predictive systems.For diagnostic/screening models, cases of a non-comparative study, comparative study with testing of the superiority hypothesis, comparative study with testing of a hypothesis of non-inferiority in cross-sectional studies are considered.For predictive models, cases of randomized controlled trials of the complex intervention "prediction + prediction-dependent patient management" with testing of the hypothesis of superiority and non-inferiority are considered.It is emphasized that representativeness of the sample and other design components are no less important in clinical trials than sample size.They are even more important since systematic biases in clinical trials are primary, and even the most sophisticated statistical analysis cannot compensate for design defects.The reduction of clinical trials to external validation of models (i.e.evaluation of accuracy metrics on external data) seems completely unreasonable.It is recommended to perform clinical trials with the design adequate to the tasks, so that further clinical and economic analysis and comprehensive assessment of medical technologies are possible.The sample size calculation methods described in the article can potentially be applied to a wider range of medical devices.

Development of artificial intelligence methods in medicine requires large volumes of input data available. The source of this data is electronic health records. Extraction of data from health records is accompanied by a number of difficulties, mainly associated with their being filled out in any form and doctors using various abbreviations when putting down the information. The paper describes a method of processing electronic health records which allows extracting the necessary information from them - the one needed for building work algorithms of artificial intelligence software complexes and their learning. The method was developed and tested out on electronic health records filled out in Russian. For working with medical documents filled out in other languages, it needs no special adaptation. For this, it is sufficient to change teaching data and perform complete learning of all models.

ЦЕЛЬ ИССЛЕДОВАНИЯ Стандартизировать требования к описанию результатов создания и валидации моделей машинного обучения для обеспечения прозрачности и доверия со стороны практического здравоохранения и надзорных органов. МАТЕРИАЛ И МЕТОДЫ Выполнено аналитическое исследование данных на основе методологии Cross-Industry Standard Process for Data Mining (реализованы четыре первые фазы цикла исследования; фазы оценки и внедрения должны быть выполнены в формате внешней независимой валидации). Использованы аналитические методы научного познания: анализ, синтез, индукция. РЕЗУЛЬТАТЫ Разработаны практико-методические рекомендациии по подготовке отчета (рукописи научной статьи) о разработке и валидации модели машинного обучения. Рекомендации оформлены в виде: а) методического документа; б) чек-листа для ускоренной проверки структуры, полноты и качества содержания рукописи. Принципиальные отличия авторской разработки от аналогов: наличие стандартной структуры медицинской научной статьи; баланс представления клинической и технологической информации; включение специфических аспектов машинного обучения, наиболее релевантных с медицинской точки зрения; методическая информация для поддержки авторов; обеспечение возможности применения чек-листа разными заинтересованными специалистами. ЗАКЛЮЧЕНИЕ Оригинальный методический подход и чек-лист могут повысить качество, прозрачность и воспроизводимость научных отчетов в сфере машинного обучения для здравоохранения. Разработка отличается балансом технологических и клинических аспектов, образовательной значимостью и комплексными возможностями. Предложенный инструментарий может быть валидирован независимыми исследователями.