Variable Selection in Multistate Models for Correlated Data With Application in a COVID-19 Vaccination Study.

Area of Science:

• Health Services Research
• Epidemiological Research
• Biostatistics

Background:

• Multistate models (MSM) are crucial for studying patient transitions across clinical states.
• MSM complexity poses challenges in parameter estimation and interpretation.
• Within-subject correlations in transition times can lead to inefficient estimation and questionable inference.

Purpose of the Study:

• To propose a novel method for variable selection in multistate models with correlated data.
• To address computational and interpretational challenges in complex MSMs.
• To enforce sparsity in multistate models for improved clarity.

Main Methods:

• Reparameterization of the likelihood function.
• Approximation of the penalty term using a smooth hyperbolic tangent function.
• Variable selection for correlated data within MSM framework.

Main Results:

• The proposed method demonstrates accuracy in variable selection and parameter estimation through extensive simulations.
• The method was successfully applied to analyze care transitions in a COVID-19 vaccine cohort.
• Identified key factors influencing transitions among healthy, infection, and hospitalization states.

Conclusions:

• The developed method offers a robust approach for variable selection in correlated multistate models.
• This facilitates more efficient and reliable inference in health services and epidemiological studies.
• Enhances understanding of patient pathways, exemplified by COVID-19 transitions.