Modeling unmeasured baseline information in observational time-to-event data subject to delayed study entry

Area of Science:

• Biostatistics
• Epidemiology
• Medical Data Science

Background:

• Left-truncated data is common in observational time-to-event analyses, particularly in longitudinal studies where baseline information may be unmeasured.
• Standard approaches using Cox proportional hazards models with delayed entry measurements can be biased due to potential time-dependency of covariates.
• Missing baseline data complicates accurate analysis of treatment effects and risk factors.

Purpose of the Study:

• To develop and validate a novel Bayesian joint model to handle unmeasured baseline covariates in left-truncated time-to-event data.
• To leverage longitudinal data trajectories to infer missing baseline measurements.
• To improve the accuracy of risk prediction and effect estimation in observational studies with data limitations.

Main Methods:

• A Bayesian joint model was developed, integrating a mixed-effects model for longitudinal data with a proportional hazards model for event occurrence.
• The model specifically addresses left-truncated data by inferring unmeasured baseline covariates using observed longitudinal trajectories.
• Simulations were conducted to compare the proposed method against a simpler two-stage approach.

Main Results:

• The proposed Bayesian joint model demonstrated favorable performance compared to a simpler two-stage approach in simulations.
• The method effectively utilizes longitudinal covariate trajectories to make inferences about missing baseline measurements in left-truncated data.
• Application to a German diabetes register showed the model's utility in investigating the impact of baseline blood glucose on antidiabetic treatment failure.

Conclusions:

• The Bayesian joint model offers a robust solution for analyzing left-truncated time-to-event data with unmeasured baseline covariates.
• This approach enhances the reliability of inferences in observational studies by effectively imputing missing baseline information.
• The findings have significant implications for epidemiological research and clinical practice, particularly in managing chronic diseases like diabetes.