Semi-parametric accelerated failure time regression analysis with application to interval-censored HIV/AIDS data

Area of Science:

• Biostatistics
• Survival Analysis
• Statistical Modeling

Background:

• Investigating non-linear relationships in time-to-event data is crucial for understanding complex biological processes.
• Accelerated failure time (AFT) models are commonly used for survival data analysis.
• Existing methods may not adequately capture non-linear effects of continuous covariates.

Purpose of the Study:

• To develop a statistical framework for analyzing interval-censored data with potentially non-linear covariate effects.
• To propose a sieve maximum likelihood estimator (MLE) for estimating model parameters.
• To assess the performance and robustness of the proposed estimators.

Main Methods:

• Incorporation of a non-linear effect of a continuous explanatory variable into an AFT model, creating a partial linear model.
• Development and application of a sieve maximum likelihood estimator (MLE) for simultaneous parameter estimation.
• Validation through simulation studies and application to real-world HIV observational data.

Main Results:

• The sieve MLE is asymptotically efficient and normally distributed.
• Simulation studies demonstrate robust and efficient estimation of scale and regression parameters.
• The estimator for the non-linear function effectively captures various smooth non-linear patterns.
• The model was successfully applied to HIV data, analyzing the non-linear effect of baseline viral load on time to viral load suppression.

Conclusions:

• The proposed partial linear model and sieve MLE provide a powerful tool for analyzing interval-censored survival data with non-linear covariate effects.
• The method is statistically sound, computationally efficient, and practically applicable to biomedical research.
• This approach enhances the understanding of factors influencing treatment outcomes, such as HIV viral load suppression.