Survival curve estimation for informatively coarsened discrete event-time data

Area of Science:

• Biostatistics
• Survival Analysis
• Epidemiology

Background:

• Event-time data with irregular observations (interval-censored data) are common in medical research.
• Standard analysis assumes non-informative censoring, which can lead to biased survival function estimates if violated.
• Informative censoring, where observation times depend on event occurrence, complicates standard survival analysis.

Purpose of the Study:

• To extend standard survival analysis methods for interval-censored data to accommodate informative censoring mechanisms.
• To develop a Bayesian approach for estimating survival functions with informatively interval-censored data.
• To provide a robust method for analyzing event-time data where censoring may be related to the event itself.

Main Methods:

• Developed a statistical framework to incorporate various assumptions about the censoring mechanism into survival function estimation.
• Implemented a Bayesian extension that mixes estimates over a distribution of assumed censoring mechanisms.
• Applied the methods to a natural history study of HIV-infected individuals.

Main Results:

• The proposed methods allow for more accurate survival function estimation when censoring is informative.
• The Bayesian approach provides a flexible way to handle uncertainty in assumptions about the censoring process.
• Demonstrated the practical application and potential benefits of the methods in a real-world HIV cohort study.

Conclusions:

• The developed methods offer a significant improvement over standard approaches for informatively interval-censored data.
• Incorporating explicit assumptions about censoring mechanisms enhances the reliability of survival estimates.
• This work provides valuable tools for researchers dealing with complex event-time data in medical and epidemiological studies.