Estimation in the progressive illness-death model: A nonexhaustive review

Area of Science:

• Biostatistics
• Epidemiology
• Medical Statistics

Background:

• Multistate models effectively describe complex event history data, such as disease progression.
• The illness-death model is a fundamental structure for analyzing time-to-event data with multiple endpoints in medical studies.
• Biomedical researchers require interpretable results beyond transition rates, including predictive probabilities.

Purpose of the Study:

• To review existing methods for estimating predictive probabilities in progressive illness-death models, considering covariates.
• To introduce novel, feasible estimation methods for cumulative incidence functions, especially those conditional on covariates.
• To evaluate the performance of different estimation approaches through simulation and discuss available software.

Main Methods:

• Review of estimation techniques for transition probabilities, occupation probabilities, cumulative incidence functions, and sojourn times.
• Application of subsampling (landmarking) and presmoothing techniques to address censoring and variability.
• Development and illustration of new methods for estimating cumulative incidence functions conditionally on covariates.

Main Results:

• Presmoothed estimators for cumulative incidences are presented as a novel approach.
• Feasible estimation methods for covariate-conditional cumulative incidence functions are introduced.
• A comparative simulation study assesses the performance of various estimation strategies.

Conclusions:

• The reviewed and proposed methods enhance the analysis of complex event history data in biomedical research.
• New presmoothed estimators and covariate-conditional methods offer improved insights into cumulative incidences.
• Software solutions in R packages facilitate the application of these advanced statistical models.