Cox model with interval-censored covariate in cohort studies

Area of Science:

• Epidemiology
• Biostatistics
• Survival Analysis

Background:

• Cohort studies often track time-to-event outcomes with periodic follow-up visits.
• Secondary events, recorded at visits, act as time-varying covariates but are incompletely observed (interval-censored).
• Existing methods for missing covariates are inadequate for interval-censored data, leading to biased estimators.

Purpose of the Study:

• To develop statistical methods for handling interval-censored time-varying covariates in Cox proportional hazards models.
• To address the bias introduced by current practices of using the latest observed covariate status.
• To provide valid and practical approaches for analyzing time-to-event data with partially observed covariates.

Main Methods:

• Proposed an available-data estimator.
• Developed a doubly robust-type estimator.
• Introduced a maximum likelihood estimator via the Expectation-Maximization (EM) algorithm.
• Presented asymptotic properties and practical validation approaches.

Main Results:

• The proposed methods effectively handle interval-censored covariates in Cox models.
• Demonstrated improved estimation accuracy compared to current practices.
• Validated the methods using a real-world example from the Northern Manhattan Study.

Conclusions:

• The novel estimators provide valid and robust analysis for time-to-event data with interval-censored time-varying covariates.
• These methods offer significant improvements over existing techniques for handling partially observed covariate data in cohort studies.
• The study provides practical tools for epidemiologists and biostatisticians analyzing complex longitudinal data.