Regression analysis of interval-censored survival data with covariates using log-linear models

Area of Science:

• Biostatistics
• Survival Analysis
• Medical Statistics

Background:

• Analyzing event time data with censored observations (left-, right-, or interval-censored) is crucial in medical research.
• Existing life-table techniques require extension to effectively handle interval-censored failures.

Purpose of the Study:

• To extend life-table techniques for censored survival data using log-linear models.
• To incorporate interval-censored failures into survival analysis.
• To apply these methods to analyze the recurrence time of treated melanoma patients.

Main Methods:

• Utilized log-linear models to extend life-table techniques for censored survival data.
• Employed the Expectation-Maximization (EM) algorithm for maximum likelihood estimation.
• Assumed a stepwise hazard function over time intervals, allowing for nonparametric, parametric (exponential), and semiparametric (Cox proportional hazard) models.
• Adapted the restricted EM algorithm for hypothesis testing and confidence interval construction.

Main Results:

• Successfully incorporated interval-censored failures into survival data analysis.
• Demonstrated the flexibility of the approach by implementing various survival models (nonparametric, exponential, Cox).
• Applied the developed methods to a real-world dataset of melanoma patient recurrence times.

Conclusions:

• The extended life-table techniques provide a robust framework for analyzing event time data with various censoring types.
• The EM algorithm and restricted EM algorithm are effective tools for parameter estimation and inference in these models.
• The methodology offers valuable insights into melanoma recurrence patterns, aiding clinical understanding and treatment strategies.