Estimating the parameters of multi-state models with time-dependent covariates through likelihood decomposition

Area of Science:

• Biostatistics
• Medical Informatics
• Health Services Research

Background:

• Multi-state models are essential for medical research but become complex with numerous states, bidirectional transitions, and time-dependent covariates.
• Modeling patient treatments, such as for end-stage renal disease (ESRD), requires handling these complexities, including age and treatment variations.

Purpose of the Study:

• To develop a specific multi-state model capable of handling complex medical scenarios with numerous states and time-dependent factors.
• To address limitations in existing modeling tools for intricate patient treatment pathways.

Main Methods:

• Designed a specialized multi-state model utilizing likelihood decomposition and separate maximizations for parameter estimation.
• Employed Poisson likelihoods based on time at risk and observed transitions within short, age-constant intervals.
• The method accommodates models with many parameters, such as 10 renal replacement therapies.

Main Results:

• Successfully estimated one hundred age-dependent transitions by assuming constant model parameters across seven time intervals.
• Demonstrated the method's scalability, unaffected by the number of parameters to estimate.

Conclusions:

• The developed multi-state modeling approach is adaptable for various diseases with multiple states or grades.
• Applicability is contingent on the absence of interactions between patient treatment courses.