Andrey D. Ermak
Research Subjects
Andrey D. Ermak has a total of 14 co-authors. Here are the top co-authors by number of shared publications:
- Aleksander V. Gusev
Federal Research Institute for Health Organization and Informatics
6 shared publications - Денис Владимирович Гаврилов
K-SkAI LLC
6 shared publications - Anna Andreychenko
ITMO University, St. Petersburg, Russia
5 shared publications - Roman E. Novitskiy
K-SkAI LLC
5 shared publications - A. N. Kaftanov
Burdenko Neurosurgical Institute, Moscow, Russia
2 shared publications A. E. Andreichenko
K-Sky
1 shared publication
Latest Publications
- Open Access
- 16 January 2026
- Published by Sechenov University in National Health Care (Russia)
- Vol. 6 (4) , 55-63
Disease diagnosis from unstructured medical texts using machine learning techniques
Modern machine learning methods open new opportunities for analyzing medical texts. The use of unstructured data enables improved clinical decision support and the development of personalized patient treatment approaches. The aim of the study: to develop an optimal algorithm for disease prediction using multi-label classification based on medical texts from selected patient treatment cases. Materials and methods. The study utilized anonymized electronic medical records of 387 590 patients. Textual data were processed using lemmatization and vectorization based on a pretrained FastText model. A multi-label classification model was developed to predict 156 diagnostic categories grouped by major disease classes. Neural network architectures and decision tree ensembles were applied for model building. Results. The proposed models demonstrated high effectiveness. The use of various text vector aggregation methods improved prediction quality. The model showed stability and clinical interpretability, supporting its applicability in real-world medical practice. Conclusion. The developed approach to analyzing unstructured medical texts using machine learning methods is a promising tool for disease diagnosis support. Further research will focus on improving model interpretability and adapting models to diverse clinical data sources.
- 16 January 2026
- Published by Sechenov University in National Health Care (Russia)
- Vol. 6 (4) , 55-63
Open AccessDisease diagnosis from unstructured medical texts using machine learning techniques
Modern machine learning methods open new opportunities for analyzing medical texts. The use of unstructured data enables improved clinical decision support and the development of personalized patient treatment approaches. The aim of the study: to develop an optimal algorithm for disease prediction using multi-label classification based on medical texts from selected patient treatment cases. Materials and methods. The study utilized anonymized electronic medical records of 387 590 patients. Textual data were processed using lemmatization and vectorization based on a pretrained FastText model. A multi-label classification model was developed to predict 156 diagnostic categories grouped by major disease classes. Neural network architectures and decision tree ensembles were applied for model building. Results. The proposed models demonstrated high effectiveness. The use of various text vector aggregation methods improved prediction quality. The model showed stability and clinical interpretability, supporting its applicability in real-world medical practice. Conclusion. The developed approach to analyzing unstructured medical texts using machine learning methods is a promising tool for disease diagnosis support. Further research will focus on improving model interpretability and adapting models to diverse clinical data sources.
- Open Access
- 11 September 2025
- Published by Autonomous non-profit scientific and medical organization - Questions of Oncology in Voprosy onkologii
- Vol. 71 (4)
Machine Learning for Cancer Risk Prediction Using Electronic Health Records: An Automated Screening Framework
Введение. Своевременная диагностика онкологических заболеваний повышает выживаемость пациентов и снижает затраты на здравоохранение за счет сокращения числа госпитализаций и повышения шансов на ремиссию. Сохраняется необходимость в практичных и интерпретируемых инструментах скрининга, которые могут эффективно способствовать раннему выявлению пациентов с онкологическими заболеваниями, для своевременного вмешательства.Цель. Разработка и внешняя валидация моделей машинного обучения для прогнозирования вероятности развития онкологических заболеваний в течение 18 мес. на основе данных реальной клинической практики.Материалы и методы. В исследовании использовались анонимизированные данные электронных медицинских карт 1,3 млн пациентов 36 регионов Российской Федерации. В качестве предикторов рассмотрены пол, возраст, среднее изменение массы тела за месяц, скорость оседания эритроцитов, гемоглобин крови, индекс массы тела и история клинически значимых сопутствующих заболеваний. Целевое событие представлено любым онкологическим заболеванием, определенным по кодам группы С МКБ-10 у 177 384 пациентов. Для сравнения использовались модели Logistic Regression, LGBMClassifier, Random Forest, Linear Discriminant Analysis и Naive Bayes. Внешняя валидация проводилась на данных из регионов с различным географическим происхождением (29 681 и 25 145 пациентов). Результаты. Модель на основе LGBMClassifier продемонстрировала лучшие результаты с AUROC 0,807 (95 % ДИ 0,798–0,815) при внутреннем тестировании, а также на внешних данных, взятых из отдельного региона и отдельного временного промежутка (0,794 (95 % ДИ 0,786–0,800) и 0,790 (95 % ДИ 0,782–0,798) соответственно).Заключение. Новый подход с использованием модели машинного обучения, подготовленной на простых и распространенных клинических, лабораторных и анамнестических признаках, продемонстрировал эффективность и практичность применения как на внешних данных, так и по сравнению с предыдущими исследованиями.
- 11 September 2025
- Published by Autonomous non-profit scientific and medical organization - Questions of Oncology in Voprosy onkologii
- Vol. 71 (4)
Open AccessMachine Learning for Cancer Risk Prediction Using Electronic Health Records: An Automated Screening Framework
Введение. Своевременная диагностика онкологических заболеваний повышает выживаемость пациентов и снижает затраты на здравоохранение за счет сокращения числа госпитализаций и повышения шансов на ремиссию. Сохраняется необходимость в практичных и интерпретируемых инструментах скрининга, которые могут эффективно способствовать раннему выявлению пациентов с онкологическими заболеваниями, для своевременного вмешательства.Цель. Разработка и внешняя валидация моделей машинного обучения для прогнозирования вероятности развития онкологических заболеваний в течение 18 мес. на основе данных реальной клинической практики.Материалы и методы. В исследовании использовались анонимизированные данные электронных медицинских карт 1,3 млн пациентов 36 регионов Российской Федерации. В качестве предикторов рассмотрены пол, возраст, среднее изменение массы тела за месяц, скорость оседания эритроцитов, гемоглобин крови, индекс массы тела и история клинически значимых сопутствующих заболеваний. Целевое событие представлено любым онкологическим заболеванием, определенным по кодам группы С МКБ-10 у 177 384 пациентов. Для сравнения использовались модели Logistic Regression, LGBMClassifier, Random Forest, Linear Discriminant Analysis и Naive Bayes. Внешняя валидация проводилась на данных из регионов с различным географическим происхождением (29 681 и 25 145 пациентов). Результаты. Модель на основе LGBMClassifier продемонстрировала лучшие результаты с AUROC 0,807 (95 % ДИ 0,798–0,815) при внутреннем тестировании, а также на внешних данных, взятых из отдельного региона и отдельного временного промежутка (0,794 (95 % ДИ 0,786–0,800) и 0,790 (95 % ДИ 0,782–0,798) соответственно).Заключение. Новый подход с использованием модели машинного обучения, подготовленной на простых и распространенных клинических, лабораторных и анамнестических признаках, продемонстрировал эффективность и практичность применения как на внешних данных, так и по сравнению с предыдущими исследованиями.
- Open Access
- 20 July 2025
- Published by Federal Research Institute for Health Organization and Informatics in Public Health
- Vol. 5 (2) , 4-16
A model for predicting death in adult patients within 10 years
Introduction. The identification of risk factors and the prediction of mortality from various causes are important issues in medicine. From a preventive perspective, it is crucial to identify patients at high risk of death, as early detection and treatment of diseases effectively increase life expectancy. The purpose of the study: to develop a universal model for predicting death in adult patients within 10 years and to compare the predictive ability of predicting death in a large contemporary cohort of the machine learning model (decision trees) with a Cox regression. Materials and methods. The data source for the study was the database of the Webiomed predictive analytics platform. The study included 1,129,268 records of 201,985 patients aged 18 years and older. 177 predictive features were investigated, of which 12 were selected for modelling as a result of a multi-stage selection process. Two survival analysis algorithms, CoxPHFitter and RandomSurvivalForest, were used for modelling. The models were used to determine the probability of death within 1, 3, 5 and 10 years. Results. Both models performed well in predicting death, however, the best result was obtained by the RSF model. Metrics of the best model with 95% CI for predicting death within 10 years: AUC0.921 (0.914–0.929), Accuracy 0.849 (0.84–0.858), Sensitivity 0.813 (0.795–0.83), Specificity 0.871 (0.859–0.882), Concordance index 0.867 (0.861–0.874), Positive predictive value 0.791 (0.776–0.806), Negative Predictive Value 0.886 (0.876–0.895). Conclusion. Machine learning models predict mortality outcomes well, demonstrating high discrimination and classification accuracy. Their use may help to identify high-risk patients to inform decisions to prevent death.
- 20 July 2025
- Published by Federal Research Institute for Health Organization and Informatics in Public Health
- Vol. 5 (2) , 4-16
Open AccessA model for predicting death in adult patients within 10 years
Introduction. The identification of risk factors and the prediction of mortality from various causes are important issues in medicine. From a preventive perspective, it is crucial to identify patients at high risk of death, as early detection and treatment of diseases effectively increase life expectancy. The purpose of the study: to develop a universal model for predicting death in adult patients within 10 years and to compare the predictive ability of predicting death in a large contemporary cohort of the machine learning model (decision trees) with a Cox regression. Materials and methods. The data source for the study was the database of the Webiomed predictive analytics platform. The study included 1,129,268 records of 201,985 patients aged 18 years and older. 177 predictive features were investigated, of which 12 were selected for modelling as a result of a multi-stage selection process. Two survival analysis algorithms, CoxPHFitter and RandomSurvivalForest, were used for modelling. The models were used to determine the probability of death within 1, 3, 5 and 10 years. Results. Both models performed well in predicting death, however, the best result was obtained by the RSF model. Metrics of the best model with 95% CI for predicting death within 10 years: AUC0.921 (0.914–0.929), Accuracy 0.849 (0.84–0.858), Sensitivity 0.813 (0.795–0.83), Specificity 0.871 (0.859–0.882), Concordance index 0.867 (0.861–0.874), Positive predictive value 0.791 (0.776–0.806), Negative Predictive Value 0.886 (0.876–0.895). Conclusion. Machine learning models predict mortality outcomes well, demonstrating high discrimination and classification accuracy. Their use may help to identify high-risk patients to inform decisions to prevent death.
- Open Access
- 1 October 2024
- Published by Silicea - Poligraf, LLC in Cardiovascular Therapy and Prevention
- Vol. 24 (1) , 4130
Development and validation of machine learning models predicting hospitalizations of hypertensive patients over 12 months
Разработка и валидация моделей машинного обучения, прогнозирующих госпитализации пациентов с артериальной гипертензией в течение 12 месяцев
- 1 October 2024
- Published by Silicea - Poligraf, LLC in Cardiovascular Therapy and Prevention
- Vol. 24 (1) , 4130
Open AccessDevelopment and validation of machine learning models predicting hospitalizations of hypertensive patients over 12 months
Разработка и валидация моделей машинного обучения, прогнозирующих госпитализации пациентов с артериальной гипертензией в течение 12 месяцев
- Open Access
- 6 May 2024
- Published by Endocrinology Research Centre in Diabetes mellitus
- Vol. 27 (2) , 142-157
Development and validation of machine learning models to predict unplanned hospitalizations of patients with diabetes within the next 12 months
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.
- 6 May 2024
- Published by Endocrinology Research Centre in Diabetes mellitus
- Vol. 27 (2) , 142-157
Open AccessDevelopment and validation of machine learning models to predict unplanned hospitalizations of patients with diabetes within the next 12 months
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.
- Open Access
- 27 March 2024
- Published by Silicea - Poligraf, LLC in Russian Journal of Cardiology
- Vol. 29 (4) , 5679
Assessment of pulmonary embolism probability using a machine learning model
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.
- 27 March 2024
- Published by Silicea - Poligraf, LLC in Russian Journal of Cardiology
- Vol. 29 (4) , 5679
Open AccessAssessment of pulmonary embolism probability using a machine learning model
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.