A two-part mixture model for longitudinal adverse event severity data.

Area of Science:

• Biostatistics
• Pharmacovigilance
• Clinical Trial Analysis

Background:

• Mixed effects logistic regression models are commonly used for longitudinal adverse event (AE) severity data.
• Extreme bimodality in predicted random effects distributions can lead to biased parameter estimates and poor predictive performance.
• Investigational drug safety assessments require accurate modeling of dose-response relationships for adverse events.

Purpose of the Study:

• To address the issue of extreme bimodality in the distribution of predicted interindividual random effects from a mixed effects logistic regression model.
• To propose and evaluate a two-part mixture modeling approach for analyzing longitudinal AE severity data.
• To improve the statistical and predictive performance of AE severity modeling in clinical trials.

Main Methods:

• A mixed effects logistic regression model was initially fitted to four-point ordered categorical AE severity data.
• A novel two-part mixture model was developed, with the first part modeling AE incidence and the second modeling severity grade.
• Simulation studies were conducted to assess the predictive and statistical performance (bias, precision) of the proposed approach.

Main Results:

• The initial mixed effects model exhibited extreme bimodality in predicted random effects, suggesting potential estimation and prediction issues.
• The proposed two-part mixture model effectively resolved the bimodality issue.
• Simulation studies demonstrated favorable predictive and statistical performance of the two-part mixture approach.

Conclusions:

• The two-part mixture modeling approach offers a robust solution for analyzing longitudinal AE severity data with extreme bimodality.
• This method enhances the reliability of dose-AE severity response assessments for investigational drugs.
• The proposed approach improves the accuracy of safety profile evaluations in clinical research.