Bayesian mixed hidden Markov models: a multi-level approach to modeling categorical outcomes with differential...

Area of Science:

• Biostatistics
• Epidemiology
• Health Outcomes Research

Background:

• Questionnaire-based health status outcomes are susceptible to misclassification, potentially biasing effect estimates of risk factors.
• Analyzing misclassified outcomes becomes more complex in multi-level settings when factors influencing both health processes and misclassification are explored simultaneously.

Purpose of the Study:

• To propose a novel statistical model for handling differential misclassification in categorical health outcomes within a multi-level framework.
• To jointly estimate health status prevalence, transition probabilities, and misclassification probabilities while accounting for cluster-level heterogeneity.

Main Methods:

• Development of a fully Bayesian mixed hidden Markov model (BMHMM).
• Incorporation of random effects into Hidden Markov Model (HMM) parameters for joint estimation.
• Modeling true health status prevalence and transitions as functions of covariates in a multi-level structure.

Main Results:

• Simulation studies validated the BMHMM's estimation procedure and computational efficiency.
• The proposed method demonstrated gains over existing approaches that ignore misclassification and heterogeneity.
• Application to the Southern California Children's Health Study identified risk factors for asthma transition and misclassification.

Conclusions:

• The BMHMM effectively addresses differential misclassification and cluster-level heterogeneity in categorical health outcomes.
• This approach provides more accurate estimates of health status prevalence and transitions compared to traditional methods.
• The model offers a robust framework for analyzing complex health data in multi-level settings.