Target Adverse Event Profiles for Predictive Safety in the Postmarket Setting

Area of Science:

• Pharmacovigilance
• Computational toxicology
• Drug safety science

Background:

• Predicting drug adverse events (AEs) at the time of drug approval is crucial for patient safety.
• Previous target adverse-event (TAE) profile models showed promise but required improvement.

Purpose of the Study:

• To enhance a TAE profile model for predicting AEs on US Food and Drug Administration (FDA) drug labels at drug approval.
• To improve the accuracy and scope of AE prediction using an updated model and algorithm.

Main Methods:

• An improved model incorporated more drugs and features, utilizing a novel algorithm.
• Comparator drugs with similar target activities were analyzed using aggregated AEs from FDA Adverse Event Reporting System (FAERS), FDA drug labels, and medical literature.
• An ensemble machine learning model evaluated FAERS case counts, disproportionality scores, and AE prevalence in comparator drug labels and literature.

Main Results:

• The enhanced model achieved a classifier performance of F1 score 0.71, area under the precision-recall curve (AUC-PR) 0.78, and area under the receiver operating characteristic curve (AUC-ROC) 0.87.
• The study demonstrated the effectiveness of the improved TAE analysis in predicting drug-related adverse events.

Conclusions:

• The enhanced TAE profile model shows significant promise for predicting adverse events at the time of drug approval.
• This approach can aid regulatory agencies and pharmaceutical companies in proactive drug safety assessments.