Bayesian competing risks survival modeling for assessing the cause of death of patients with heart failure

2024-09-24

AI Summary

This study addresses the challenge of assessing competing risks of cardiovascular versus non-cardiovascular mortality in heart failure patients undergoing cardiac resynchronization therapy (CRT). Methodologically, we develop a Bayesian competing risks model incorporating baseline covariates and propose a novel generalized Bayesian model selection and diagnostic framework tailored to competing risks. For the first time, we systematically apply posterior inference of transition probabilities to clinical cause-of-death attribution. Using a Cox-type structure, Markov chain Monte Carlo (MCMC) sampling, and posterior predictive checks, we differentially model and interpret the risks associated with each cause of death. Key contributions include: (1) establishing a new Bayesian diagnostic paradigm for competing risks models, and (2) uncovering heterogeneous effects of key covariates across distinct mortality pathways. Results demonstrate significantly improved accuracy in cause-of-death attribution and provide an interpretable, individualized prognostic tool for clinical decision support.

Technology Category

• Machine Learning: Bayesian Learning
• Reasoning under Uncertainty: Relational Probabilistic Models
• Knowledge Representation and Reasoning: Diagnosis and Abductive Reasoning

Application Category

• User Modeling, Personalization and Recommendation: Attacks and countermeasures in recommendation systems
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Graph Algorithms and Modeling for the Web: Algorithms and analysis for incomplete, noisy, or partially observed Web-related graphs

Abstract

Competing risk models are survival models with several events of interest acting in competition and whose occurrence is only observed for the event that occurs first in time. This paper presents a Bayesian approach to these models in which the issue of model selection is treated in a special way by proposing generalizations of some of the Bayesian procedures used in univariate survival analysis. This research is motivated by a study on the survival of patients with hearth failure undergoing cardiac resynchronization therapy, a procedure which involves the implant of a device to stabilize the heartbeat. Two different causes of causes of death have been considered: cardiovascular and non-cardiovascular, and a set of baseline covariates are examined in order to better understand their relationship with both causes of death. Model selection procedures and model checking analyses have been implemented and assessed. The posterior distribution of some relevant outputs such as transition probabilities have been computed and discussed.

Problem

• Bayesian modeling for competing causes of death in heart failure patients
• Generalizing Bayesian procedures for survival analysis model selection
• Assessing cardiovascular vs non-cardiovascular death risk factors

Innovation

• Bayesian competing risks survival modeling
• Generalized Bayesian model selection procedures
• Posterior distribution computation for transition probabilities