Regression models for expected length of stay

Area of Science:

• Biostatistics
• Survival Analysis
• Longitudinal Data Analysis

Background:

• Traditional multi-state models often use transition intensities, making covariate effects on expected length of stay (ELOS) indirect.
• These methods typically rely on the Markov assumption, which may not hold in real-world scenarios.

Purpose of the Study:

• To propose a new method for modeling ELOS in multi-state models using pseudo-observations.
• To allow direct interpretation of covariate effects on ELOS without the Markov assumption.
• To extend the method for time-varying covariates and longitudinal data.

Main Methods:

• Constructing pseudo-observations for ELOS from multi-state data.
• Developing regression models for ELOS using these pseudo-observations.
• Combining pseudo-observations with landmarking for longitudinal data analysis.
• Utilizing generalized estimating equations and sandwich estimators for model fitting.

Main Results:

• The proposed method allows direct covariate effect interpretation on ELOS.
• It successfully avoids the Markov assumption.
• The approach accommodates time-varying covariates and can be extended to longitudinal data.
• The method's performance is evaluated under various conditions like censoring and non-Markovianity.

Conclusions:

• Pseudo-observations offer a flexible and interpretable approach to modeling ELOS in multi-state models.
• This method enhances the analysis of time-to-event data, particularly when dealing with complex dependencies and time-varying factors.
• The approach is validated through simulation studies and application to health economics data.