Streamlining Considerations for Safety Measures: A Predictive Model for Addition of Clinically Significant Adverse...

Area of Science:

• Pharmacovigilance and Drug Safety
• Machine Learning in Healthcare
• Regulatory Science

Background:

• Clinically significant adverse reactions (CSARs) are crucial for patient safety, requiring timely updates to Japanese package inserts (PIs).
• Identifying the need for CSAR additions is often a complex and time-consuming process for regulatory bodies and pharmaceutical companies.

Purpose of the Study:

• To develop a machine learning model for the early prediction of CSAR additions to Japanese PIs.
• To leverage accumulated adverse drug reaction (ADR) data from both domestic (Japan) and international (US) sources for enhanced prediction accuracy.

Main Methods:

• Utilized ADR case accumulation data from the Japanese Adverse Drug Event Report and the US FDA Adverse Event Reporting System.
• Constructed a predictive model using DataRobot, employing a generalized linear model with informative features.
• Evaluated model performance using the Matthews correlation coefficient, achieving a cross-validation score of 0.8754 and a holdout score of 0.8995.

Main Results:

• The study identified 414 cases of CSAR additions, categorized by data source (domestic, international, or company core data sheet revisions).
• The best-performing generalized linear model demonstrated high predictive accuracy for CSAR additions.
• The model effectively integrated data from both Japanese and US ADR databases.

Conclusions:

• The developed machine learning model accurately predicts the addition of CSARs to PIs based on ADR case accumulation.
• This predictive capability can expedite safety updates, potentially improving patient safety by informing healthcare professionals of risks sooner.
• The model's effectiveness highlights the value of integrating international ADR data for robust pharmacovigilance.