Ascertainment correction in frailty models for recurrent events data

Area of Science:

• Biostatistics
• Epidemiology
• Survival Analysis

Background:

• Retrospective studies with recurrent events often select individuals based on past event history.
• This ascertainment can lead to non-representative samples, introducing selection bias.
• Ignoring this bias can distort statistical inference.

Purpose of the Study:

• To investigate the impact of unadjusted data analysis in the presence of selection bias.
• To propose a corrected statistical method for analyzing recurrent event data with ascertainment.
• To provide unbiased and consistent estimators for regression models.

Main Methods:

• Utilized Andersen-Gill and shared frailty regression models.
• Investigated the consequences of ignoring selection mechanisms in data analysis.
• Developed a simple likelihood adjustment for corrected analysis.
• Assessed the proposed method through simulation studies.

Main Results:

• Failure to adjust for ascertainment introduces bias in estimators of covariate effects, incidence, and frailty variance.
• The proposed likelihood adjustment yields unbiased and consistent estimators.
• Simulation results confirm the effectiveness of the corrected method.

Conclusions:

• Selection bias is a critical issue in retrospective recurrent event studies.
• A simple likelihood adjustment effectively corrects for ascertainment bias.
• The proposed method ensures reliable statistical inference in such studies.