Prediction of Prescription Drug–Drug Interaction Scores Using an Artificial Neural Network

Prediction of Prescription Drug–Drug Interaction Scores Using an Artificial Neural Network

Parvaneh Ahmadi1 sayed rashid khalifeh soltani2

1) M.Sc. Graduate in Industrial Engineering, Systems Optimization, Faculty of Engineering, University of Kurdistan, Sanandaj, Iran Email:
2) Ph.D. Student in Industrial Engineering, Systems Optimization, Department of Industrial Engineering, Faculty of Engineering, Yazd University, Yazd, Iran Email:

IOI Code : ABEB-ACZGZA
Publication : 4th International Conference on Recent Advances in Engineering, Innovation and Technology - Belgium(eitconf.com)
Abstract :
    The increasing diversity of medications, the growing prevalence of polypharmacy, and the complexity of medical treatments have made the assessment of drug–drug interactions an important issue in patient safety. Traditional approaches to interaction assessment generally rely on case-by-case examination of drug pairs and human expertise, which limits their ability to rapidly analyze a large volume of prescriptions. This study aims to develop a data-driven model for predicting the drug–drug interaction scores of medical prescriptions using indicators that can be extracted from prescription data. To this end, after data preprocessing and filtering, 96,093 medical prescriptions were analyzed. Features including patient gender, age group, physician specialty, medication diversity, and drug group were extracted and standardized. Subsequently, a drug interaction database was constructed for 500 generic drugs and their pairwise combinations. Interaction severity was converted into a quantitative index, with unknown, minor, moderate, and major interactions assigned coefficients of 0, 1, 2.5, and 5, respectively. The interaction score of each prescription was then calculated and normalized. Subsequently, an artificial neural network with 50 input features, one hidden layer with 64 neurons, a ReLU activation function, and the Adam optimization algorithm was trained. The results showed that the proposed model achieved a coefficient of determination (R²) of 0.9995 on the test data and 0.9994 on the validation data. These findings indicate that prescription-based indicators can provide a suitable basis for intelligent prescribing systems, including alert and clinical decision support systems.

Keywords : drug drug interaction intelligent prescribing artificial neural network health data mining patient safety decision support