Cumulative disease progression models for cross-sectional data: a review and comparison

Area of Science:

• Biostatistics
• Computational Biology
• Medical Informatics

Background:

• Accurate disease progression modeling aids early diagnosis and tailored therapies.
• Statistical models, from path models to Bayesian networks, are used for cumulative disease progression.
• Applications include modeling cancer (tumorigenesis) and HIV (drug resistance).

Purpose of the Study:

• To provide an overview of different statistical disease progression models.
• To analyze the advantages and disadvantages of various modeling approaches.
• To compare model performance, especially when assumptions are violated, using simulations.

Main Methods:

• Review of existing statistical models for disease progression.
• Simulation studies to evaluate model performance under violated assumptions.
• Comparison of model fitting for induced multivariate probability distributions and topological relationships.

Main Results:

• Model performance evaluation revealed that less or more complex models than the true data-generating class often performed better.
• Conjunctive Bayesian networks demonstrated flexibility in fitting oncogenetic trees.
• Mixture models with two components effectively fitted mixtures of oncogenetic trees with three components.

Conclusions:

• The choice of statistical model for disease progression is critical and may not always be the most complex or simplest.
• Flexible models like conjunctive Bayesian networks offer advantages for specific applications like oncogenetic tree fitting.
• Simulation studies are valuable for understanding model robustness and identifying optimal modeling strategies.