Improving estimation efficiency for survival data analysis by integrating a coarsened time-to-event outcome from an...

Area of Science:

• Biostatistics
• Data Science
• Epidemiology

Background:

• Big data era necessitates combining diverse data sources for improved accuracy.
• Data integration is challenged by heterogeneity in data forms across studies.
• Survival analysis often faces incomplete or irregularly observed data.

Purpose of the Study:

• To develop a robust data integration method for survival analysis with heterogeneous data.
• To improve estimation efficiency by incorporating external data with different measurement structures.
• To address challenges posed by continuous vs. interval-censored outcomes and partial covariate information.

Main Methods:

• Utilizing empirical likelihood to derive informative weights for data integration.
• Constructing a weighted estimator that combines primary and external study data.
• Developing theoretical guarantees for the proposed weighted estimator's efficiency.
• Validating the method's robustness to working model misspecification through simulations.

Main Results:

• The proposed weighted estimator demonstrates higher estimation efficiency than conventional methods.
• The method's advantage in efficiency is robust even when working models are misspecified.
• Simulation studies confirm the theoretical findings and the method's practical performance.
• The approach successfully integrates external data for improved analysis in a real-world application.

Conclusions:

• The novel data integration technique effectively enhances survival analysis with heterogeneous data.
• The method offers a statistically sound and efficient way to leverage external data sources.
• This approach has significant implications for complex observational studies, such as those in neurodegenerative disease research.