A monotone single index model for spatially referenced multistate current status data.

Area of Science:

• Biostatistics
• Biomedical Research
• Statistical Modeling

Background:

• Assessing multistate disease progression is crucial in biomedical research, exemplified by periodontal disease (PD).
• Current status endpoints, providing only a single data snapshot of disease progression, complicate traditional inferential frameworks.
• Disease progression data can exhibit clustering and spatial correlation within subjects, necessitating advanced analytical approaches.

Purpose of the Study:

• To develop a novel Bayesian semiparametric model for analyzing multistate disease progression with current status data.
• To incorporate spatial random effects and flexible error distributions to better model complex disease dynamics.
• To provide a computationally efficient and clinically interpretable method for estimating disease state and transition probabilities.

Main Methods:

• Proposed a Bayesian semiparametric accelerated failure time model.
• Utilized an inverse-Wishart proposal for spatial random effects and a Dirichlet process mixture of Gaussians for flexible errors.
• Employed a monotone single index model with integrated basis expansion and constrained Gaussian process priors for the link function.

Main Results:

• Established parameter identifiability for the proposed model.
• Developed scalable computational methods using elliptical slice sampling and fast circulant embedding.
• Demonstrated straightforward estimation of parameters, state occupation, and transition probabilities.

Conclusions:

• The developed Bayesian model offers a robust framework for analyzing complex multistate disease progression data, particularly current status endpoints with spatial correlations.
• The proposed computational techniques ensure efficient estimation and practical applicability.
• The method was validated using synthetic data and successfully applied to a real-world periodontal disease dataset.