Patient data analysis at the input stage: neural network models for supporting medical decisions

Predictive analytics plays a key role in modern medical research and clinical practice, providing the ability to study and predict individual and population outcomes. Its methods are used to assess patient survival across various diseases, taking into account the dynamics of clinical data and comorbid conditions. Additionally, it is widely applied to determine the risk of developing pathological complications, allowing for the personalization of preventive and treatment approaches. An important direction is the prediction of the effectiveness of therapeutic interventions based on mathematical modeling and big data analysis, which helps optimize the selection of treatment methods. Predictive analytics underpins the development of classification systems and patient stratification by risk, which is crucial for enhancing the accuracy of diagnosis, treatment, and long-term monitoring. Within such studies, specialized datasets are formed that contain both the outcomes of medical events and a set of predictive variables, allowing for a certain degree of accuracy in predicting the occurrence of these events. In this context, the quality of the source data, the representativeness of the sample, and the adequacy of the statistical methods used are of paramount importance, as they collectively determine the reliability of the predictive models obtained. The aim of the study is to develop a deep learning model for predicting clinical outcomes of hospitalization based on electronic medical record data, as well as to assess its effectiveness in conditions of strong class imbalance and to develop methods for its improvement for application in clinical practice. The implementation of such a model in medical information systems, including electronic health records, can automate the assessment of the risk of adverse outcomes upon patient admission and provide timely alerts to medical staff. This contributes to increased accuracy and efficiency in clinical decision-making, as well as improved quality of medical care. However, the successful implementation of such models requires further research, validation on real clinical data, and strict adherence to ethical norms and standards for data use in healthcare.