Regression models using parametric pseudo-observations.

Area of Science:

• Biostatistics
• Survival Analysis
• Statistical Modeling

Background:

• Censored time-to-event data analysis commonly employs the Cox proportional hazards model.
• Nonparametric Kaplan-Meier pseudo-observations are an alternative but can exhibit higher variability.
• Spline-based survival estimators offer potential advantages in reducing variability.

Purpose of the Study:

• To propose and evaluate pseudo-observations derived from a flexible parametric survival estimator.
• To compare the performance of parametric pseudo-observations against nonparametric ones in regression models.
• To estimate parameters associated with cumulative risk more efficiently.

Main Methods:

• Development of pseudo-observations using a flexible parametric survival function estimator.
• Application of these parametric pseudo-observations in regression models.
• Conducting a simulation study to compare empirical standard errors with nonparametric pseudo-observations.

Main Results:

• Parametric pseudo-observations demonstrated substantial reductions in variability in various simulation settings.
• Empirical standard errors were reduced by up to approximately one-third compared to nonparametric methods.
• This reduction in variability is equivalent to needing a 125% larger sample size for nonparametric methods.

Conclusions:

• Parametric pseudo-observations provide a more statistically efficient method for analyzing censored time-to-event data.
• The proposed method offers significant gains in precision, particularly in cumulative risk estimation.
• This approach enhances the analysis of survival data, offering a valuable alternative to existing methods.