Rank-based estimating equations with general weight for accelerated failure time models: an induced smoothing...

Area of Science:

• Biostatistics
• Survival Analysis
• Statistical Modeling

Background:

• Rank-based estimating functions for accelerated failure time (AFT) models often suffer from non-smoothness, complicating analysis.
• Existing induced smoothing techniques are primarily effective for Gehan's weight, limiting their application to general weights.

Purpose of the Study:

• To develop an iterative-induced smoothing procedure applicable to general weights in AFT models.
• To improve computational efficiency and maintain accuracy compared to non-smooth methods.
• To extend the methodology for handling missing data and various sampling schemes in survival analysis.

Main Methods:

• An iterative-induced smoothing procedure is proposed, using an initial estimator from Gehan's weight.
• Asymptotic properties of the new estimators are consistent with non-smooth estimating equations.
• An efficient resampling approach is used for variance estimation, avoiding repeated solving of equations.
• The method is generalized to incorporate additional weights for missing data and complex sampling designs.

Main Results:

• The proposed estimators achieve similar asymptotic properties as non-smooth estimators but with significantly faster computation.
• Variance estimators demonstrate good approximation of estimation variability.
• The methodology effectively handles missing covariates and diverse sampling schemes in real-world datasets.

Conclusions:

• The iterative-induced smoothing procedure offers an efficient and accurate alternative for analyzing AFT models with general weights.
• The approach is robust and adaptable to complex data structures, including missing data.
• An R package (aftgee) is available for implementing this advanced statistical methodology.