Evaluating Drug Risk Using GAN and SMOTE Based on CFDA's Spontaneous Reporting Data

Area of Science:

• Pharmacovigilance and Drug Safety
• Machine Learning in Healthcare
• Computational Toxicology

Background:

• Adverse drug reactions (ADRs) are a significant public health concern, necessitating robust pharmacovigilance systems.
• Current methods for re-evaluating post-marketing drug risk levels and enabling status switches (e.g., prescription to over-the-counter) lack automation.
• China categorizes drugs into prescription (Rx), OTC-A, and OTC-B based on risk, highlighting the need for accurate classification.

Purpose of the Study:

• To develop an automated classification model for predicting drug risk levels and enabling status switches.
• To address the challenge of class imbalance in drug risk categorization using advanced machine learning techniques.
• To improve the efficiency and accuracy of pharmacovigilance by providing automated risk assessment tools.

Main Methods:

• Utilized a large dataset of 985,960 spontaneous ADR reports from China's CSRD (2011-2018).
• Employed feature enhancement using Generative Adversarial Networks (GAN) and Synthetic Minority Over-Sampling Technique (SMOTE) to address data imbalance and expand feature space.
• Integrated feature selection (FS) for identifying important ADR symptoms and Random Forest (RF) for final classification.

Main Results:

• The developed model achieved a high accuracy of 98% in classifying drug risk levels.
• Feature enhancement and sample balancing techniques effectively mitigated classification deviations caused by data imbalance.
• The model successfully expanded ADR data in both feature and sample spaces for improved prediction.

Conclusions:

• The proposed classification model offers an effective automated solution for evaluating drug risk levels and facilitating status switches.
• This approach significantly enhances pharmacovigilance by providing reliable automated risk assessment for post-marketing drugs.
• The integration of GAN and SMOTE with RF demonstrates a powerful strategy for handling imbalanced data in drug safety research.