Flexible regression model selection for survival probabilities: with application to AIDS.

Area of Science:

• Biostatistics
• Survival Analysis
• Clinical Trials

Background:

• Clinicians require understanding covariate effects on survival probabilities at specific times.
• Short- and long-term failure risks can be influenced by different factors, necessitating flexible modeling.

Purpose of the Study:

• To propose an objective methodology for constructing consistent and asymptotically normal estimators.
• To develop estimators for regression coefficients and average prediction error, free from nuisance censoring variables.
• To implement a model selection strategy using hypothesis testing to control error rates.

Main Methods:

• A flexible modeling strategy is employed, treating the problem as a missing data scenario.
• Augmented inverse probability weighted complete case estimators are utilized.
• Stepup or stepdown multiple hypothesis testing procedures are used for model selection, controlling false positive rates or generalized familywise error rates.

Main Results:

• The proposed methodology yields consistent and asymptotically normal estimators for regression coefficients and average prediction error.
• The method effectively addresses nuisance censoring variables.
• Model selection procedures are demonstrated to control error rates.

Conclusions:

• The developed objective methodology provides a robust approach to analyzing survival data with censored observations.
• This method enhances the reliability of estimating covariate effects and prediction errors in clinical studies.
• The approach is validated through a simulation study and an AIDS trial analysis.