Performance of three estimation methods in repeated time-to-event modeling.

Area of Science:

• Pharmacometrics
• Statistical Modeling
• Clinical Trial Analysis

Background:

• Clinical trial outcomes often involve event data, such as adverse events or symptoms.
• Mixed-effects modeling software like NONMEM can struggle with low-information, ordered categorical event data.
• Traditional methods may show poor performance with sparse event data, impacting trial analysis.

Purpose of the Study:

• To evaluate the performance of three estimation methods for repeated time-to-event data in clinical trials.
• To compare the Laplace method, Stochastic Approximation Expectation-Maximization (SAEM), and importance sampling within NONMEM 7.
• To assess method performance under varying event frequencies and interindividual variability.

Main Methods:

• A stochastic simulation and estimation study was conducted.
• A repeated time-to-event model with exponential interindividual variability was used.
• Performance was assessed based on parameter bias and precision across different simulation conditions.

Main Results:

• All methods showed bias and imprecision with very few observed events, most notably the Laplace method.
• SAEM and importance sampling generally outperformed the Laplace method when the event frequency was below 43%.
• Above 43% event frequency, all three methods demonstrated comparable performance.

Conclusions:

• SAEM and importance sampling offer improved performance over the Laplace method for modeling repeated time-to-event data, particularly in scenarios with limited events.
• The choice of estimation method is crucial for accurate analysis of event data in clinical trials.
• Further investigation into these methods is warranted for optimizing clinical trial outcome analysis.