A Bayesian analysis of doubly censored data using a hierarchical Cox model

Area of Science:

• Biostatistics
• Epidemiology
• Survival Analysis

Background:

• Pooling survival data from multiple studies presents statistical challenges, including doubly censored data and population heterogeneity.
• Interval-censored origins, where the exact start time is unknown but falls within a range, complicate survival analyses.
• Heterogeneity among study populations can bias results if not properly accounted for.

Purpose of the Study:

• To develop and compare methods for handling interval-censored origins in pooled survival data.
• To utilize a random-effects hierarchical Cox proportional hazards model to address study heterogeneity.
• To examine the effect of GB virus type C (GBV-C) coinfection on survival in HIV-infected individuals using pooled data.

Main Methods:

• Developed two novel approaches to incorporate uncertainty from interval-censored origins.
• Compared these new methods against traditional single-value imputation techniques.
• Applied a random-effects hierarchical Cox proportional hazards model to pooled survival data from three studies.
• Utilized an approximate Bayesian approach with partial likelihood for handling interval-censored HIV infection times.

Main Results:

• The developed methods for interval-censored origins offer improved incorporation of uncertainty compared to single-value imputation.
• The random-effects hierarchical Cox model effectively accounts for heterogeneity across study populations.
• The analysis identified the impact of GBV-C coinfection on survival outcomes in HIV-infected individuals.

Conclusions:

• An approximate Bayesian approach using partial likelihood is recommended for pooled survival analyses with interval-censored origins.
• Accounting for both interval-censored origins and study heterogeneity is crucial for accurate survival analysis.
• This methodology provides a robust framework for analyzing complex survival data in public health research.