A bayesian approach to robust modeling of skewed biomedical data

Area of Science:

• Biostatistics
• Biomedical Data Analysis
• Statistical Modeling

Background:

• Biomedical data often exhibit continuous, strictly positive, and asymmetric distributions.
• Traditional regression models may struggle with skewness, heavy tails, and outliers common in biomedical datasets.
• Accurate modeling is crucial for reliable inference in physiological and diagnostic measurements.

Purpose of the Study:

• To propose a robust Bayesian regression approach for continuous, strictly positive, and asymmetric biomedical data.
• To enhance the flexibility and reliability of statistical modeling for complex biological data.
• To provide accurate estimation and uncertainty quantification for biomedical research.

Main Methods:

• Utilizing log-symmetric distributions to model skewed and heavy-tailed data.
• Employing robust prior specifications for improved model resilience.
• Conducting full posterior inference using Markov Chain Monte Carlo (MCMC) techniques.
• Evaluating model performance with classical and Bayesian criteria (AIC, BIC, DIC, WAIC, PSIS-LOO).

Main Results:

• The proposed Bayesian approach effectively models asymmetric and positive biomedical data.
• Demonstrated robustness to outliers and improved uncertainty quantification.
• Simulation studies and real-world applications confirmed the method's effectiveness and adaptability.
• Achieved superior model fit and predictive ability compared to standard methods.

Conclusions:

• Robust Bayesian regression with log-symmetric distributions is a valuable tool for analyzing complex biomedical data.
• The methodology offers enhanced inferential reliability and accuracy for physiological and diagnostic measurements.
• This approach is particularly beneficial when dealing with distributional asymmetry and the need for outlier resilience.