Penalized Bayesian forward continuation ratio model with application to high-dimensional data with discrete survival...

Area of Science:

• Biostatistics
• Computational Biology
• Genomics

Background:

• Time-to-event data are often continuous, but discrete survival data can be more appropriate in certain scenarios.
• Existing discrete survival models include the forward continuation ratio model, which is related to the Cox proportional hazards model.
• Previous implementations in high-dimensional settings used frequentist algorithms, limiting variable selection capabilities.

Purpose of the Study:

• To propose a Bayesian penalized forward continuation ratio model for discrete survival data.
• To explore the use of different priors for variable selection and regularization in high-dimensional settings.
• To evaluate the model's performance against existing frequentist methods and apply it to a real-world dataset.

Main Methods:

• Developed a Bayesian penalized forward continuation ratio model with a complementary log-log link.
• Investigated various prior inclusion probabilities (1%, 10%, 50%) for variable selection.
• Applied the model to a publicly available acute myeloid leukemia dataset to identify genomic features.

Main Results:

• The proposed Bayesian model demonstrated superior variable selection performance compared to the frequentist approach.
• A 10% prior inclusion probability yielded better results than 1% or 50% in simulations.
• Identified nine genomic features, mapping to ten genes, with five previously linked to leukemia.

Conclusions:

• The Bayesian penalized forward continuation ratio model offers flexibility for discrete survival data analysis.
• The model enables simultaneous variable selection and uncertainty quantification.
• The approach is effective for identifying disease-associated genomic features, as shown in the acute myeloid leukemia study.