Probabilistic subgroup identification using Bayesian finite mixture modelling: a case study in Parkinson's disease...

Area of Science:

• Computational biology
• Biostatistics
• Medical informatics

Background:

• Complex diseases often present with heterogeneous phenotypes that are difficult to identify using traditional methods.
• Cross-sectional data is commonly available but poses challenges for capturing disease progression and subgroup dynamics.
• Finite mixture modelling provides a probabilistic framework for classifying individuals into latent subgroups based on observed data.

Purpose of the Study:

• To explore the application of finite mixture modelling for identifying complex disease phenotypes from cross-sectional data.
• To investigate the utility of quantifying uncertainty in posterior probabilities of subgroup membership for enhanced analysis.
• To demonstrate practical applications of uncertainty quantification in describing subgroup membership and characteristics.

Main Methods:

• Utilized finite mixture modelling, a probabilistic classification technique.
• Employed a Bayesian approach to quantify uncertainty in posterior probabilities of subgroup membership.
• Applied the methodology to a case study involving Parkinson's disease (PD) using data from the Unified Parkinson's Disease Rating Scale (UPDRS).

Main Results:

• Demonstrated that quantifying uncertainty in posterior probabilities allows for nuanced descriptions of individual subgroup membership.
• Showcased how uncertainty quantification can be used to characterize identified subgroups using patient-centered covariates not initially included in the model.
• Successfully identified latent subgroups within Parkinson's disease based on UPDRS symptom data, highlighting the practical utility of the approach.

Conclusions:

• Finite mixture modelling, particularly with uncertainty quantification, is a valuable tool for dissecting complex disease phenotypes from cross-sectional data.
• The proposed methods offer improved ways to describe individual and subgroup characteristics, enhancing clinical interpretability.
• This approach holds promise for advancing the understanding and classification of diseases like Parkinson's.