Risk Factor Identification in Heterogeneous Disease Progression with L1-Regularized Multi-state Models.

Area of Science:

• Biostatistics
• Longitudinal Data Analysis
• Disease Progression Modeling

Background:

• Multi-state models (MSM) are crucial for analyzing longitudinal data in disease progression.
• Extending regularization methods for variable selection to MSMs is an underexplored area.

Purpose of the Study:

• To develop and evaluate a novel L1-regularized multi-state model (L1MSTATE) framework.
• To enable simultaneous parameter estimation and variable selection in MSMs.
• To enhance the identification of risk factors in disease progression analysis.

Main Methods:

• Developed the L1MSTATE framework for integrated parameter estimation and variable selection.
• Employed a one-step coordinate descent algorithm for efficient optimization.
• Validated the approach through extensive simulation studies and analysis of the EBMT dataset.

Main Results:

• L1MSTATE demonstrated superior performance in identifying risk factors compared to existing regularized MSMs.
• L1MSTATE outperformed un-regularized multi-state models (MSTATE) in identifying key risk factors, particularly in small sample size scenarios.
• The model showed enhanced power in predicting transition probabilities compared to MSTATE using real-world data.

Conclusions:

• L1MSTATE offers an effective and computationally efficient method for variable selection and parameter estimation in multi-state modeling.
• The developed framework improves the accuracy of risk factor identification and prediction of disease progression.
• The L1MSTATE methodology is accessible via the open-source R package 'L1mstate'.