Variable selection in covariate dependent random partition models: an application to urinary tract infection.

Area of Science:

• Medical Informatics
• Statistics
• Urology

Background:

• Lower urinary tract symptoms (LUTS) can indicate urinary tract infections (UTIs), a condition impacting quality of life and healthcare costs.
• Early detection of UTI presence and severity is crucial for effective management.
• Current assessment typically relies on white blood cell (WBC) counts in urine samples.

Purpose of the Study:

• To develop a novel Bayesian nonparametric regression model for patient clustering.
• To identify patterns in symptom profiles associated with UTI.
• To provide clinicians with a tool for meaningful patient stratification based on WBC counts and symptoms.

Main Methods:

• Utilized clinical data from 1341 patients diagnosed with UTI (WBC ≥ 1).
• Recorded a clinical profile of 34 symptoms for each patient.
• Applied a Bayesian nonparametric regression model with a Dirichlet process prior and spike and slab priors for regression coefficients.
• Employed Markov Chain Monte Carlo (MCMC) methods for posterior inference.

Main Results:

• The proposed model successfully clusters patients based on WBC counts and symptom profiles.
• Identified specific symptoms associated with different patient clusters, aiding in UTI assessment.
• Demonstrated the model's capability to handle complex relationships between symptoms and infection indicators.

Conclusions:

• The Bayesian model offers a valuable tool for understanding UTI heterogeneity.
• Patient clustering based on symptoms and WBC counts can improve diagnostic insights.
• This approach supports personalized medicine strategies for managing urinary tract infections.