Predicting Serious Adverse Events, Medication Abuse, Misuse, and Risk of Dependence for Medications with High...

Area of Science:

• Pharmacovigilance
• Machine Learning in Healthcare
• Drug Safety

Background:

• Adverse drug events (ADEs) linked to medication abuse, misuse, and dependence pose significant public health challenges.
• Identifying predictors and developing robust detection methods for these serious adverse events (SAEs) is crucial for patient safety.

Purpose of the Study:

• To determine the frequency and predictors of ADEs associated with medication abuse, misuse, and dependence.
• To develop and evaluate machine learning models for detecting serious cases of medication abuse, misuse, and dependence.

Main Methods:

• Analysis of 455,415 ADE reports from the Korea Adverse Event Reporting System (KAERS DB) between 2013-2022.
• Multivariate logistic regression to identify predictors of abuse-, misuse-, and dependence-related ADEs.
• Development and comparison of random forest (RF), support vector machine, and eXtreme Gradient Boosting models.

Main Results:

• Concomitant use of acetaminophen, antidepressants, antipsychotics, and anticonvulsants increased ADE reporting likelihood.
• Reports from the general public were more likely to indicate abuse, misuse, and dependence compared to healthcare professionals.
• Ketamine and bromazepam had the highest likelihood of being SAEs; cardiovascular and respiratory disorders were the most reported SAE types.
• The RF model achieved the highest predictive performance (AUC-ROC 0.928, accuracy 94.4%), with reporter type as a key feature.

Conclusions:

• The random forest model demonstrates superior performance in detecting serious medication abuse, misuse, and dependence cases.
• Reporter type is a critical feature for identifying high-risk individuals.
• Integrating patient-reported data and polypharmacy surveillance is vital for early detection of serious ADEs.