Development of Predictive models for Personalized medicine based on Electronic health records

This study explores the application of predictive machine learning models in personalized medicine using data from electronic health records (EHR). The purpose of this research was to evaluate different algorithms in predicting clinical outcomes and identifying at-risk patients. Retrospective analysis of EHR data was conducted on a sample of 1000 patients. Logistic regression algorithms, Random Forest and XGBoost were used. Model performance was assessed using AUC-ROC metrics, precision, sensitivity and F1-score. XGBoost model showed the highest accuracy (AUC-ROC = 0.88), while the logistic regression had the lowest predictive power (AUC-ROC = 0.78). The implementation of the Explainable AI methods (SHAP analysis) allowed for a better understanding of key risk factors. The results confirmed the potential use of machine learning in personalized medicine, but also indicated the challenges of model interpretation and the need for external validation. The limitations include the retrospective nature of the data as well as the ethical aspects of using AI in healthcare. This study confirms the usefulness of EHR-based predictive models for identifying at-risk patients and optimizing therapeutic strategies. Further research is needed for their full integration into clinical practice.