Bayesian joint and individual component regression for multigroup physiological data.

Area of Science:

• Statistics
• Biomedical Data Analysis
• Radiological Imaging

Background:

• Heterogeneous multigroup data present challenges for standard regression models.
• Medical research, especially in radiology and imaging, frequently encounters such complex data structures.
• Existing methods like Joint and Individual Component Regression (JICO) do not adequately quantify uncertainty or incorporate prior knowledge.

Purpose of the Study:

• To propose a Bayesian Joint and Individual Component Regression (Bayesian-JICO) framework.
• To extend JICO by incorporating a probabilistic formulation for uncertainty quantification and prior knowledge integration.
• To offer a more reliable inference method, particularly for datasets with limited sample sizes.

Main Methods:

• Developed a Bayesian probabilistic formulation for Joint and Individual Component Regression (JICO).
• Employed Markov Chain Monte Carlo (MCMC) for posterior estimation.
• Validated the Bayesian-JICO model using simulated data and the Australian Institute of Sport (AIS) dataset.

Main Results:

• Bayesian-JICO demonstrated superior predictive accuracy, interpretability, and robustness compared to traditional methods.
• The framework successfully quantified uncertainty through posterior distributions and credible intervals.
• Credible intervals were provided for parameter estimates, enhancing reliability.

Conclusions:

• The proposed Bayesian-JICO framework offers a comprehensive, uncertainty-aware solution for analyzing heterogeneous multigroup data.
• This approach provides more reliable statistical inference, especially beneficial in biomedical and radiological research.
• Bayesian-JICO has broad applicability across various scientific disciplines dealing with complex data structures.