Bayesian analysis of semi-parametric Cox models with latent variables

Area of Science:

• Oncology
• Biostatistics
• Epidemiology

Background:

• Respiratory cancer is a leading cause of cancer death globally.
• Reducing respiratory cancer mortality remains a significant public health challenge.
• Existing models may not fully capture all risk factors influencing survival.

Purpose of the Study:

• To propose a novel semi-parametric Cox model incorporating latent variables.
• To assess the impact of both observed and unobserved risk factors on respiratory cancer survival.
• To enhance the accuracy of survival time predictions in respiratory cancer patients.

Main Methods:

• Developed a semi-parametric Cox model with latent variables.
• Utilized confirmatory factor analysis to characterize latent risk factors via observed indicators.
• Employed a Bayesian estimation procedure for parameter estimation.
• Validated the methodology through simulation studies.

Main Results:

• The proposed statistical model demonstrated satisfactory performance in simulations.
• The model effectively integrates observed and latent variables for survival analysis.
• Application to Surveillance, Epidemiology, and End Results (SEER) data showcased its utility.

Conclusions:

• The novel statistical approach provides a more comprehensive understanding of respiratory cancer survival.
• This methodology can improve risk factor assessment and inform targeted interventions.
• The findings contribute to advancing statistical techniques in cancer research and epidemiology.