Pseudo-observation regression for sequentially truncated data

Area of Science:

• Biostatistics
• Epidemiology
• Survival Analysis

Background:

• Observational studies often face truncation, where event times are only observed within a specific range.
• Sequential truncation, requiring specific event time orderings, presents a more complex challenge in data analysis.
• Existing methods for estimating event time distributions under sequential truncation have limitations for regression modeling.

Purpose of the Study:

• To develop regression modeling methods for time-to-event data with sequential truncation.
• To adapt pseudo-observation techniques for this complex data structure.
• To evaluate the performance of proposed methods in simulations and real-world data.

Main Methods:

• Development of simple and modified pseudo-observation methods.
• Application to Cox and accelerated failure time (AFT) regression models.
• Validation through simulation studies and analysis of an Alzheimer's disease cohort.

Main Results:

• The proposed pseudo-observation methods provide valid regression modeling under sequential truncation.
• Modified pseudo-observations address limitations when truncation depends on covariates.
• The methods are applicable to complex observational data, including health cohort studies.

Conclusions:

• Pseudo-observation methods, particularly the modified approach, are effective for regression analysis in the presence of sequential truncation.
• These methods enhance the ability to model time-to-event data in complex observational settings.
• The study provides valuable tools for biostatisticians and epidemiologists analyzing cohort data.