A semiparametric recurrent events model with time-varying coefficients

Area of Science:

• Biostatistics
• Clinical Epidemiology
• Survival Analysis

Background:

• Recurrent events are common in clinical studies, but traditional models often assume constant effects.
• Time-varying effects are crucial for accurately modeling disease progression and treatment impacts.
• Existing methods may struggle with complex time-dependent patterns in recurrent event data.

Purpose of the Study:

• To develop and validate a flexible statistical model for recurrent events with time-varying coefficients.
• To accurately estimate time-varying and time-constant effects in recurrent event data.
• To apply the novel methodology to real-world clinical datasets.

Main Methods:

• Utilized a random effects (Gaussian frailty) model to capture event intensity.
• Employed penalized splines for robust estimation of time-varying coefficients.
• Applied Laplace approximation for efficient penalized likelihood evaluation.
• Estimated smoothing parameters analogous to variance components estimation.

Main Results:

• Simulations demonstrated the model's effectiveness in estimating both time-varying and time-independent coefficients.
• The penalized spline approach provided accurate estimates for complex coefficient functions.
• Successful application to stroke and child wheeze clinical datasets.

Conclusions:

• The proposed recurrent events model with time-varying coefficients offers a powerful tool for clinical research.
• This method enhances the analysis of longitudinal data with recurrent events.
• The approach provides valuable insights into time-dependent risk factors in clinical populations.